v1.5.5 release

- deploy_site now defaults to new VPS (94.130.36.94, /var/www/lynglang);
  use --old to deploy to d.lynglang.com as before
- infra/setup_vps.yml: Ansible playbook installs nginx + certbot, obtains
  SSL cert for lynglang.com, fixes Debian buster EOL apt sources
- infra/inventory.ini and infra/templates/nginx_lynglang.conf.j2 included

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitignore

@@ -21,3 +21,13 @@ xcuserdata
 debug.log
 /build.log
 /test.md
+/build_output.txt
+/build_output_full.txt
+/check_output.txt
+/compile_jvm_output.txt
+/compile_metadata_output.txt
+test_output*.txt
+/site/src/version-template/lyng-version.js
+/bugcontents.db
+/bugs/
+contents.db

.junie/guidelines.md

@@ -0,0 +1,11 @@
+# Lyng Project Guidelines
+
+This project uses the Lyng scripting language for multiplatform scripting.
+
+## Coding in Lyng
+When writing, refactoring, or analyzing Lyng code:
+- **Reference**: Always use `LYNG_AI_SPEC.md` in the project root as the primary source of truth for syntax and idioms.
+- **File Extensions**: Use `.lyng` for all script files.
+- **Implicit Coroutines**: Remember that all Lyng functions are implicitly coroutines; do not look for `async/await`.
+- **Everything is an Expression**: Leverage the fact that blocks, if-statements, and loops return values.
+- **Maps vs Blocks**: Be careful: `{}` is a block/lambda, use `Map()` for an empty map.

AGENTS.md

@@ -0,0 +1,42 @@
+# AI Agent Notes
+
+## Canonical AI References
+- Use `docs/ai_language_reference.md` as the primary, compiler-verified Lyng language reference for code generation.
+- For generics-heavy code generation, follow `docs/ai_language_reference.md` section `7.1 Generics Runtime Model and Bounds` and `7.2 Differences vs Java / Kotlin / Scala`.
+- Use `docs/ai_stdlib_reference.md` for default runtime/module APIs and stdlib surface.
+- Treat `LYNG_AI_SPEC.md` and older docs as secondary if they conflict with the two files above.
+- Prefer the shortest clear loop: use `for` for straightforward iteration/ranges; use `while` only when loop state/condition is irregular or changes in ways `for` cannot express cleanly.
+- In Lyng code, slice strings with range indexing (`text[a..<b]`, `text[..<n]`, `text[n..]`) and avoid Java/Kotlin-style `substring(...)`.
+
+## Lyng-First API Declarations
+- Use `.lyng` declarations as the single source of truth for Lyng-facing API docs and types (especially module extern declarations).
+- Prefer defining Lyng entities (enums/classes/type shapes) in `.lyng` files; only define them in Kotlin when there is Kotlin/platform-specific implementation detail that cannot be expressed in Lyng.
+- Avoid hardcoding Lyng API documentation in Kotlin registrars when it can be declared in `.lyng`; Kotlin-side docs should be fallback/bridge only.
+- For mixed pluggable modules (Lyng + Kotlin), embed module `.lyng` sources as generated Kotlin string literals, evaluate them into module scope during registration, then attach Kotlin implementations/bindings.
+
+## Kotlin/Wasm generation guardrails
+- Avoid creating suspend lambdas for compiler runtime statements. Prefer explicit `object : Statement()` with `override suspend fun execute(...)`.
+- Do not use `statement { ... }` or other inline suspend lambdas in compiler hot paths (e.g., parsing/var declarations, initializer thunks).
+- If you need a wrapper for delegated properties, check for `getValue` explicitly and return a concrete `Statement` object when missing; avoid `onNotFoundResult` lambdas.
+- For any code in `commonMain`, verify it is Kotlin Multiplatform compatible before finishing. Do not use JVM-only APIs or Java-backed convenience methods such as `Map.putIfAbsent`; prefer stdlib/common equivalents and run at least the relevant compile/test task that exercises the `commonMain` source set.
+- If wasmJs browser tests hang, first run `:lynglib:wasmJsNodeTest` and look for wasm compilation errors; hangs usually mean module instantiation failed.
+- Do not increase test timeouts to mask wasm generation errors; fix the invalid IR instead.
+
+## Type inference notes (notes/new_lyng_type_system_spec.md)
+- Nullability is Kotlin-style: `T` non-null, `T?` nullable, `!!` asserts non-null.
+- `void` is a singleton of class `Void` (syntax sugar for return type).
+- Object members are always allowed even on unknown types; non-Object members require explicit casts. Remove `inspect` from Object and use `toInspectString()` instead.
+- Type expression checks: `x is T` is value instance check; `T1 is T2` is type-subset; `A in T` means `A` is subset of `T`; `==` is structural type equality.
+- Type aliases: `type Name = TypeExpr` (generic allowed) expand to their underlying type expressions; no nominal distinctness.
+- Bounds and variance: `T: A & B` / `T: A | B` for bounds; declaration-site variance with `out` / `in`.
+- Do not reintroduce bytecode fallback opcodes (e.g., `GET_NAME`, `EVAL_*`, `CALL_FALLBACK`) or runtime name-resolution fallbacks; all symbol resolution must stay compile-time only.
+
+## Bytecode frame-first migration plan
+- Treat frame slots as the only storage for locals/temps by default; avoid pre-creating scope slot mappings for compiled functions.
+- Create closure references only when a capture is detected; use a direct frame+slot reference (foreign slot ref) instead of scope slots.
+- Keep Scope as a lazy reflection facade: resolve name -> slot only on demand for Kotlin interop (no eager name mapping on every call).
+- Avoid PUSH_SCOPE/POP_SCOPE in bytecode for loops/functions unless dynamic name access or Kotlin reflection is requested.
+
+## ABI proposal notes
+- Runtime generic metadata for generic extern classes is tracked in `proposals/extern_generic_runtime_abi.md`.
+- Keep this design `Obj`-centric: do not assume extern-class values are `ObjInstance`; collection must be enabled on `ObjClass`.

CHANGELOG.md

@@ -1,119 +1,152 @@
-## Changelog
-
-### Unreleased
-
-- Language: Abstract Classes and Interfaces
-  - Support for `abstract` modifier on classes, methods, and variables.
-  - Introduced `interface` as a synonym for `abstract class`, supporting full state (constructors, fields, `init` blocks) and implementation by parts via MI.
-  - New `closed` modifier (antonym to `open`) to prevent overriding class members.
-  - Refined `override` logic: mandatory keyword when re-declaring members that exist in the ancestor chain (MRO).
-  - MI Satisfaction: Abstract requirements are automatically satisfied by matching concrete members found later in the C3 MRO chain without requiring explicit proxy methods.
-  - Integration: Updated highlighters (lynglib, lyngweb, IDEA plugin), IDEA completion, and Grazie grammar checking.
-  - Documentation: Updated `docs/OOP.md` with sections on "Abstract Classes and Members", "Interfaces", and "Overriding and Virtual Dispatch".
-
-- Language: Class properties with accessors
-  - Support for `val` (read-only) and `var` (read-write) properties in classes.
-  - Syntax: `val name [ : Type ] get() { body }` or `var name [ : Type ] get() { body } set(value) { body }`.
-  - Laconic Expression Shorthand: `val prop get() = expression` and `var prop get() = read set(v) = write`.
-  - Properties are pure accessors and do **not** have automatic backing fields.
-  - Validation: `var` properties must have both accessors; `val` must have only a getter.
-  - Integration: Updated TextMate grammar and IntelliJ plugin (highlighting + keywords).
-  - Documentation: New "Properties" section in `docs/OOP.md`.
-
-- Language: Restricted Setter Visibility
-  - Support for `private set` and `protected set` modifiers on `var` fields and properties.
-  - Allows members to be publicly readable but only writable from within the declaring class or its subclasses.
-  - Enforcement at runtime: throws `AccessException` on unauthorized writes.
-  - Supported only for declarations in class bodies (fields and properties).
-  - Documentation: New "Restricted Setter Visibility" section in `docs/OOP.md`.
-
-- Language: Late-initialized `val` fields in classes
-  - Support for declaring `val` without an immediate initializer in class bodies.
-  - Compulsory initialization: every late-init `val` must be assigned at least once within the class body or an `init` block.
-  - Write-once enforcement: assigning to a `val` is allowed only if its current value is `Unset`.
-  - Access protection: reading a late-init `val` before it is assigned returns the `Unset` singleton; using `Unset` for most operations throws an `UnsetException`.
-  - Extension properties do not support late-init.
-  - Documentation: New "Late-initialized `val` fields" and "The `Unset` singleton" sections in `docs/OOP.md`.
-
-- Docs: OOP improvements
-  - New page: `docs/scopes_and_closures.md` detailing `ClosureScope` resolution order, recursion‑safe helpers (`chainLookupIgnoreClosure`, `chainLookupWithMembers`, `baseGetIgnoreClosure`), cycle prevention, and capturing lexical environments for callbacks (`snapshotForClosure`).
-  - Updated: `docs/advanced_topics.md` (link to the new page), `docs/parallelism.md` (closures in `launch`/`flow`), `docs/OOP.md` (visibility from closures with preserved `currentClassCtx`), `docs/exceptions_handling.md` (compatibility alias `SymbolNotFound`).
-  - Tutorial: added quick link to Scopes and Closures.
-
-- IDEA plugin: Lightweight autocompletion (experimental)
-  - Global completion: local declarations, in‑scope parameters, imported modules, and stdlib symbols.
-  - Member completion: after a dot, suggests only members of the inferred receiver type (incl. chained calls like `Path(".." ).lines().` → `Iterator` methods). No global identifiers appear after a dot.
-  - Inheritance-aware: direct class members first, then inherited (e.g., `List` includes `Collection`/`Iterable` methods).
-  - Heuristics: handles literals (`"…"` → `String`, numbers → `Int/Real`, `[...]` → `List`, `{...}` → `Dict`) and static `Namespace.` members.
-  - Performance: capped results, early prefix filtering, per‑document MiniAst cache, cancellation checks.
-  - Toggle: Settings | Lyng Formatter → "Enable Lyng autocompletion (experimental)" (default ON).
-  - Stabilization: DEBUG completion/Quick Doc logs are OFF by default; behavior aligned between IDE and isolated engine tests.
-
-- Language: Named arguments and named splats
-  - New call-site syntax for named arguments using colon: `name: value`.
-    - Positional arguments must come before named; positionals after a named argument inside parentheses are rejected.
-    - Trailing-lambda interaction: if the last parameter is already assigned by name (or via a named splat), a trailing `{ ... }` block is illegal.
-  - Named splats: `...` can now expand a Map into named arguments.
-    - Only string keys are allowed; non-string keys raise a clear error.
-    - Duplicate assignment across named args and named splats is an error.
-    - Ellipsis (variadic) parameters remain positional-only and cannot be named.
-  - Rationale: `=` is assignment and an expression in Lyng; `:` at call sites avoids ambiguity. Declarations keep `name: Type`; call-site casts continue to use `as` / `as?`.
-  - Documentation updated: proposals and declaring-arguments sections now cover named args/splats and error cases.
-  - Tests added covering success cases and errors for named args/splats and trailing-lambda interactions.
-
-- Tooling: Highlighters and TextMate bundle updated for named args
-  - Website/editor highlighter (lyngweb + site) works with `name: value` and `...Map("k" => v)`; added JS tests covering punctuation/operator spans for `:` and `...`.
-  - TextMate grammar updated to recognize named call arguments: `name: value` after `(` or `,` with `name` highlighted as `variable.parameter.named.lyng` and `:` as punctuation; excludes `::`.
-  - TextMate bundle version bumped to 0.0.3; README updated with details and guidance.
-
-- Multiple Inheritance (MI) completed and enabled by default:
-  - Active C3 Method Resolution Order (MRO) for deterministic, monotonic lookup across complex hierarchies and diamonds.
-  - Qualified dispatch:
-    - `this@Type.member(...)` inside class bodies starts lookup at the specified ancestor.
-    - Cast-based disambiguation: `(expr as Type).member(...)`, `(expr as? Type)?.member(...)` (works with existing safe-call `?.`).
-  - Field inheritance (`val`/`var`) under MI:
-    - Instance storage is disambiguated per declaring class; unqualified read/write resolves to the first match in MRO.
-    - Qualified read/write targets the chosen ancestor’s storage.
-  - Constructors and initialization:
-    - Direct bases are initialized left-to-right; each ancestor is initialized at most once (diamond-safe de-duplication).
-    - Header-specified constructor arguments are passed to direct bases.
-  - Visibility enforcement under MI:
-    - `private` visible only inside the declaring class body.
-    - `protected` visible inside the declaring class and any of its transitive subclasses; unrelated contexts cannot access it (qualification/casts do not bypass).
-  - Diagnostics improvements:
-    - Missing member/field messages include receiver class and linearization order; hints for `this@Type` or casts when helpful.
-    - Invalid `this@Type` reports that the qualifier is not an ancestor and shows the receiver lineage.
-    - `as`/`as?` cast errors include actual and target type names.
-
-- Documentation updated (docs/OOP.md and tutorial quick-start) to reflect MI with active C3 MRO.
-
-Notes:
-- Existing single-inheritance code continues to work; resolution reduces to the single base.
-- If code previously relied on non-deterministic parent set iteration, C3 MRO provides a predictable order; disambiguate explicitly if needed using `this@Type`/casts.
-
 # Changelog
 
-All notable changes to this project will be documented in this file.
+This file tracks user-visible Lyng language/runtime/tooling changes.
+
+History note:
+- The project had periods where changelog maintenance lagged behind commits.
+- Entries below are synchronized and curated for `1.5.x`.
+- Earlier history may be incomplete and should be cross-checked with git tags/commits when needed.
 
 ## Unreleased
 
-- CLI: Added `fmt` as a first-class Clikt subcommand.
-  - Default behavior: formats files to stdout (no in-place edits by default).
-  - Options:
-    - `--check`: check only; print files that would change; exit with code 2 if any changes are needed.
-    - `-i, --in-place`: write formatted result back to files.
-    - `--spacing`: apply spacing normalization.
-    - `--wrap`, `--wrapping`: enable line wrapping.
-  - Mutually exclusive: `--check` and `--in-place` together now produce an error and exit with code 1.
-  - Multi-file stdout prints headers `--- <path> ---` per file.
-  - `lyng --help` shows `fmt`; `lyng fmt --help` displays dedicated help.
-  - Fix: Property accessors (`get`, `set`, `private set`, `protected set`) are now correctly indented relative to the property declaration.
-  - Fix: Indentation now correctly carries over into blocks that start on extra‑indented lines (e.g., nested `if` statements or property accessor bodies).
-  - Fix: Formatting Markdown files no longer deletes content in `.lyng` code fences and works correctly with injected files (resolves clobbering, `StringIndexOutOfBoundsException`, and `nonempty text is not covered by block` errors).
+- No unreleased entries yet.
 
-- CLI: Preserved legacy script invocation fast-paths:
-  - `lyng script.lyng [args...]` executes the script directly.
-  - `lyng -- -file.lyng [args...]` executes a script whose name begins with `-`.
+## 1.5.5 (2026-04-23)
 
-- CLI: Fixed a regression where the root help banner could print before subcommands.
-  - Root command no longer prints help when a subcommand (e.g., `fmt`) is invoked.
+### Concurrency and collections
+- Added coroutine coordination primitives and helpers for everyday parallel code:
+  - `Channel` for coroutine-to-coroutine communication
+  - `LaunchPool` for bounded-concurrency task execution
+  - `Iterable<Deferred>.joinAll()` to await a whole collection of deferreds in input order
+  - `CompletableDeferred.completeExceptionally(...)` and `Deferred.cancelAndJoin()`
+- Added docs and examples for the new concurrency APIs, including `joinAll()` coverage in iterable and parallelism references.
+
+### Database and time APIs
+- Added the portable `lyng.io.db` SQL contract and the first concrete providers:
+  - `lyng.io.db.sqlite` on JVM and Linux Native
+  - `lyng.io.db.jdbc` on JVM
+- Added SQLite/JDBC release hardening:
+  - nested transactions via savepoints
+  - detached materialized rows
+  - generated-key support through `ExecutionResult.getGeneratedKeys()`
+  - schema-driven value conversion for `Bool`, `Decimal`, `Date`, `DateTime`, and `Instant`
+  - portable SQLite linker/deployment fixes and documented runtime options
+- Added `Date` to `lyng.time` and the core runtime as a first-class calendar-date type, plus conversions and arithmetic across `Instant`, `DateTime`, and `Date`.
+
+### Language, stdlib, and tooling
+- Added extensions on singleton `object` declarations, including object-scoped indexer overrides for bracket syntax.
+- Added backtick string literals and formatter support.
+- Added `lyng.legacy_digest` for SHA-1 compatibility work, `String.replace`, and `buffer.base64std`.
+- Improved CLI/runtime behavior with `atExit` shutdown handlers, native release-binary work, and follow-up CLI packaging/import fixes.
+- Expanded docs across the tutorial, stdlib references, database docs, networking docs, and release notes.
+
+### Runtime/compiler stability and performance
+- Extended exact-call and higher-order lambda inlining through the bytecode compiler, including compiled fast paths for simple lambdas, wrappers, captures, and common higher-order helpers.
+- Fixed import caching and class/object bytecode dispatch on JVM.
+- Fixed immutable `val` compound assignments so true mutating `*Assign` operations continue to work while fallback reassignments report the correct read-only error.
+- Fixed closure/capture and import regressions across launched loops, singleton/object extensions, aliasing, transitive re-exports, and immutable capture escaping.
+- Improved list-fill/list-append fast paths, nullable-let inference, Decimal/Complex interop, and related regression coverage.
+
+### Release notes
+- Release metadata, homepage samples, docs, and README now point to `1.5.5`.
+
+## 1.5.4 (2026-04-03)
+
+### Runtime and compiler stability
+- Stabilized the recent `piSpigot` benchmark/compiler work for release.
+- Fixed numeric-mix regressions introduced by overly broad int-coercion in bytecode compilation.
+- Restored correct behavior for decimal arithmetic, mixed real/int flows, list literals, list size checks, and national-character script cases.
+- Fixed plain-list index fast paths so they no longer bypass subclass behavior such as `ObservableList` hooks and flow notifications.
+- Hardened local numeric compare fast paths to correctly handle primitive-coded frame slots.
+
+### Performance and examples
+- Added `piSpigot` benchmark/example coverage:
+  - `examples/pi-test.lyng`
+  - `examples/pi-bench.lyng`
+  - JVM benchmark test for release-baseline verification
+- Kept the safe list/index/runtime wins that improve the optimized `piSpigot` path without reintroducing type-unsound coercions.
+- Changed the default `RVAL_FASTPATH` setting off on JVM/Android and in the benchmark preset after verification that it no longer helps the stabilized `piSpigot` workload.
+
+### Release notes
+- Full JVM and wasm test gates pass on the release tree.
+- Benchmark findings and remaining post-release optimization targets are documented in `notes/pi_spigot_benchmark_baseline_2026-04-03.md`.
+
+## 1.5.1 (2026-03-25)
+
+### Language
+- Added string interpolation:
+  - `"$name"` identifier interpolation.
+  - `"${expr}"` expression interpolation.
+- Added literal-dollar forms in strings:
+  - `"\$"` -> `$`
+  - `"$$"` -> `$`
+  - `\\$x` is parsed as backslash + interpolation of `x`.
+- Added per-file interpolation opt-out via leading directive comment:
+  - `// feature: interpolation: off`
+
+### Docs and AI references
+- Updated compiler-accurate AI language docs:
+  - interpolation syntax and escaping
+  - per-file feature switch behavior
+- Refreshed tutorial examples and doctests to reflect new interpolation semantics.
+- Added/reworked current proposal/reference materials for Lyng common-platform guidance.
+
+### Compatibility notes
+- Interpolation is enabled by default for normal string literals.
+- Existing code that intentionally used `$name` as literal text should use `\$name`, `$$name`, or the file directive `// feature: interpolation: off`.
+
+## 1.5.0 (2026-03-22)
+
+### Major runtime/compiler direction
+- Completed migration to bytecode-first/bytecode-only execution paths.
+- Removed interpreter fallback behavior in core execution hot paths.
+- Continued frame-slot-first local/capture model improvements and related diagnostics.
+
+### Language features and semantics
+- Added/finished `return` semantics including labeled non-local forms (`return@label`).
+- Added abstract classes/members and `interface` support (as abstract-class-style construct).
+- Completed and enabled multiple inheritance with C3 MRO by default.
+- Added class properties with accessors (`get`/`set`) and restricted setter visibility (`private set`, `protected set`).
+- Added late-initialized class `val` support with `Unset` protection rules.
+- Added named arguments (`name: value`) and named splats (`...map`) with stricter validation.
+- Added assign-if-null operator `?=`.
+- Improved nullable/type-checking behavior (including `T is nullable` and related type checks).
+- Added variadic function types (`...` in function type declarations) and tighter lambda type checks.
+
+### Type system and collections
+- Added immutable collections hierarchy (`ImmutableList`, `ImmutableSet`, `ImmutableMap`).
+- Improved generic runtime binding/checking for explicit type arguments and bounds.
+- Added smarter type-aware collection ops (`+=`, `-=`) and stronger declared-member type checks.
+
+### Extern/Kotlin bridge
+- Tightened extern declaration rules:
+  - explicit extern members are required for extern class/object declarations.
+- Improved extern generic class behavior and diagnostics.
+- Extended bridge APIs for binding global functions/variables and object/member interop scenarios.
+
+### Standard library and modules
+- Added `lyng.observable` improvements (`ObservableList` hooks/events).
+- Added `Random` stdlib API used by updated samples.
+- Added/extended `lyngio.console` support and CLI integration for console interaction.
+- Migrated time APIs to `kotlin.time` (`Instant` migration and related docs/tests).
+
+### CLI, IDE, and docs/tooling
+- CLI:
+  - added first-class `fmt` command
+  - preserved direct script fast-path invocation
+  - improved command help/dispatch behavior
+- IntelliJ plugin:
+  - improved lightweight completion and documentation/inspection behavior
+  - continued highlighter and Grazie/spellchecking integration work
+- Docs:
+  - substantial updates across tutorial/OOP/type/runtime references
+  - expanded bytecode and advanced topics coverage
+
+## Migration checklist for 1.5.x
+
+- If you rely on literal `$...` strings:
+  - replace with `\$...` or `$$...`, or
+  - add `// feature: interpolation: off` at file top.
+- Review any code relying on interpreter-era fallback behavior; 1.5.x assumes bytecode-first execution.
+- For extern declarations, ensure members are explicitly declared where required.
+- For named arguments/splats, verify call sites follow stricter ordering/duplication rules.

LYNG_AI_SPEC.md

@@ -0,0 +1,130 @@
+# Lyng Language AI Specification (V1.5.0-SNAPSHOT)
+
+High-density specification for LLMs. Reference this for all Lyng code generation.
+
+## 1. Core Philosophy & Syntax
+- **Everything is an Expression**: Blocks, `if`, `when`, `for`, `while`, `do-while` return their last expression (or `void`).
+- **Static Types + Inference**: Every declaration has a compile-time type (explicit or inferred). Types are Kotlin‑style: non‑null by default, nullable with `?`.
+- **Loops with `else`**: `for`, `while`, and `do-while` support an optional `else` block.
+    - `else` executes **only if** the loop finishes normally (without a `break`).
+    - `break <value>` exits the loop and sets its return value.
+    - Loop Return Value:
+        1. Value from `break <value>`.
+        2. Result of `else` block (if loop finished normally and `else` exists).
+        3. Result of the last iteration (if loop finished normally and no `else`).
+        4. `void` (if loop body never executed and no `else`).
+- **Implicit Coroutines**: All functions are coroutines. No `async/await`. Use `launch { ... }` (returns `Deferred`) or `flow { ... }`.
+- **Functions**: Use `fun` or the short form `fn`. Function declarations are expressions returning a callable.
+- **Variables**: `val` (read-only), `var` (mutable). Supports late-init `val` in classes (must be assigned in `init` or body).
+- **Serialization**: Use `@Transient` attribute before `val`/`var` or constructor parameters to exclude them from Lynon/JSON serialization. Transient fields are also ignored during `==` structural equality checks.
+- **Null Safety**: `?` (nullable type), `?.` (safe access), `?( )` (safe invoke), `?{ }` (safe block invoke), `?[ ]` (safe index), `?:` or `??` (elvis), `?=` (assign-if-null).
+- **Equality**: `==` (equals), `!=` (not equals), `===` (ref identity), `!==` (ref not identity).
+- **Comparison**: `<`, `>`, `<=`, `>=`, `<=>` (shuttle/spaceship, returns -1, 0, 1).
+- **Destructuring**: `val [a, b, rest...] = list`. Supports nested `[a, [b, c]]` and splats.
+- **Compile-Time Resolution Only**: All names/members must resolve at compile time. No runtime name lookup or fallback opcodes.
+
+## 2. Object-Oriented Programming (OOP)
+- **Multiple Inheritance**: Supported with **C3 MRO** (Python-style). Diamond-safe.
+- **Header Arguments**: `class Foo(a, b) : Base(a)` defines fields `a`, `b` and passes `a` to `Base`.
+- **Members**: `fun name(args) { ... }`, `val`, `var`, `static val`, `static fun`.
+- **Properties (Get/Set)**: Pure accessors, no auto-backing fields.
+    ```lyng
+    var age
+        get() = _age
+        private set(v) { if(v >= 0) _age = v }
+    // Laconic syntax:
+    val area get = π * r * r
+    ```
+- **Mandatory `override`**: Required for all members existing in the ancestor chain.
+- **Visibility**: `public` (default), `protected` (subclasses and ancestors for overrides), `private` (this class instance only). `private set` / `protected set` allowed on properties.
+- **Disambiguation**: `this@Base.member()` or `(obj as Base).member()`. `as` returns a qualified view.
+- **Abstract/Interface**: `interface` is a synonym for `abstract class`. Both support state and constructors.
+- **Extensions**: `fun Class.ext()` or `val Class.ext get = ...`. Scope-isolated.
+- **Member Access**: Object members (`toString`, `toInspectString`, `let`, `also`, `apply`, `run`) are allowed on unknown types; all other members require a statically known receiver type or explicit cast.
+
+## 2.1 Type System (2026)
+- **Root Type**: Everything is an `Object` (root of the hierarchy).
+- **Nullability**: Non-null by default (`T`), nullable with `T?`, `!!` asserts non-null.
+- **Untyped params**: `fun foo(x)` -> `x: Object`, `fun foo(x?)` -> `x: Object?`.
+- **Untyped vars**: `var x` is `Unset` until first assignment locks the type (including nullability).
+  - `val x = null` -> type `Null`; `var x = null` -> type `Object?`.
+- **Inference**:
+  - List literals infer union element types; empty list defaults to `List<Object>` unless constrained.
+  - Map literals infer key/value types; empty map defaults to `Map<Object, Object>` unless constrained.
+  - Mixed numeric ops promote `Int` + `Real` to `Real`.
+- **Type aliases**: `type Name = TypeExpr` (generic allowed). Aliases expand to their underlying type expressions (no nominal distinctness).
+- **Generics**: Bounds with `T: A & B` or `T: A | B`; variance uses `out`/`in` (declaration‑site only).
+- **Casts**: `as` is a runtime-checked cast; `as?` is safe-cast returning `null`. If the value is nullable, `as T` implies `!!`.
+
+## 2.2 Type Expressions and Checks
+- **Value checks**: `x is T` (runtime instance check).
+- **Type checks**: `T1 is T2` and `A in T` are subset checks between type expressions (compile-time where possible).
+- **Type equality**: `T1 == T2` is structural (unions/intersections are order‑insensitive).
+- **Compile-time enforcement**: Bounds are checked at call sites; runtime checks only appear when the compile‑time type is too general.
+
+## 3. Delegation (`by`)
+Unified model for `val`, `var`, and `fun`.
+```lyng
+val x by MyDelegate()
+var y by Map() // Uses "y" as key in map
+fn f(a, b) by RemoteProxy() // Calls Proxy.invoke(thisRef, "f", a, b)
+```
+Delegate Methods:
+- `getValue(thisRef, name)`: for `val`/`var`.
+- `setValue(thisRef, name, val)`: for `var`.
+- `invoke(thisRef, name, args...)`: for `fn` (called if `getValue` is absent).
+- `bind(name, access, thisRef)`: optional hook called at declaration/binding time. `access` is `DelegateAccess.Val`, `Var`, or `Callable`.
+
+## 4. Standard Library & Functional Built-ins
+- **Scope Functions**:
+    - `obj.let { it... }`: result of block. `it` is `obj`.
+    - `obj.apply { this... }`: returns `obj`. `this` is `obj`.
+    - `obj.also { it... }`: returns `obj`. `it` is `obj`.
+    - `obj.run { this... }`: result of block. `this` is `obj`.
+    - `with(obj, { ... })`: result of block. `this` is `obj`.
+- **Functional**: `forEach`, `map`, `filter`, `any`, `all`, `sum`, `count`, `sortedBy`, `flatten`, `flatMap`, `associateBy`.
+- **Lazy**: `val x = cached { expensive() }` (call as `x()`) or `val x by lazy { ... }`.
+- **Collections**: `List` ( `[a, b]` ), `Map` ( `Map(k => v)` ), `Set` ( `Set(a, b)` ). `MapEntry` ( `k => v` ).
+
+## 5. Patterns & Shorthands
+- **Map Literals**: `{ key: value, identifier: }` (identifier shorthand `x:` is `x: x`). Empty map is `{:}`.
+- **Named Arguments**: `fun(y: 10, x: 5)`. Shorthand: `Point(x:, y:)`.
+- **Varargs & Splats**: `fun f(args...)`, `f(...otherList)`.
+- **Labels**: `loop@ for(x in list) { if(x == 0) break@loop }`.
+- **Dynamic**: `val d = dynamic { get { name -> ... } }` allows `d.anyName` via explicit dynamic handler (not implicit fallback).
+
+## 6. Operators & Methods to Overload
+| Op | Method | Op | Method |
+| :--- | :--- | :--- | :--- |
+| `+` | `plus` | `==` | `equals` |
+| `-` | `minus` | `<=>` | `compareTo` |
+| `*` | `mul` | `[]` | `getAt` / `putAt` |
+| `/` | `div` | `!` | `logicalNot` |
+| `%` | `mod` | `-` | `negate` (unary) |
+| `=~` | `operatorMatch` | `+=` | `plusAssign` |
+
+## 7. Common Snippets
+```lyng
+// Multiple Inheritance and Properties
+class Warrior(id, hp) : Character(id), HealthPool(hp) {
+    override fun toString() = "Warrior #%s (%s HP)"(id, hp)
+}
+
+// Map entry and merging
+val m = Map("a" => 1) + ("b" => 2)
+m += "c" => 3
+
+// Destructuring with splat
+val [first, middle..., last] = [1, 2, 3, 4, 5]
+
+// Safe Navigation and Elvis
+val companyName = person?.job?.company?.name ?: "Freelancer"
+```
+
+## 8. Standard Library Discovery
+To collect data on the standard library and available APIs, AI should inspect:
+- **Global Symbols**: `lynglib/src/commonMain/kotlin/net/sergeych/lyng/Script.kt` (root functions like `println`, `sqrt`, `assert`).
+- **Core Type Members**: `lynglib/src/commonMain/kotlin/net/sergeych/lyng/obj/*.kt` (e.g., `ObjList.kt`, `ObjString.kt`, `ObjMap.kt`) for methods on built-in types.
+- **Lyng-side Extensions**: `lynglib/stdlib/lyng/root.lyng` for high-level functional APIs (e.g., `map`, `filter`, `any`, `lazy`).
+- **I/O & Processes**: `lyngio/src/commonMain/kotlin/net/sergeych/lyng/io/` for `fs` and `process` modules.
+- **Documentation**: `docs/*.md` (e.g., `tutorial.md`, `lyngio.md`) for high-level usage and module overviews.

README.md

@@ -1,6 +1,14 @@
-# Lyng: modern scripting for kotlin multiplatform
+# Lyng: ideal scripting for kotlin multiplatform
+
+__Please visit the project homepage: [https://lynglang.com](https://lynglang.com) and a [telegram channel](https://t.me/lynglang).__
+
+__Main development site:__ [https://gitea.sergeych.net/SergeychWorks/lyng](https://gitea.sergeych.net/SergeychWorks/lyng)
+__github mirror__: [https://github.com/sergeych/lyng](https://github.com/sergeych/lyng)
+
+We keep github as a mirror and backup for the project, while the main development site is hosted on gitea.sergeych.net. We use gitea for issues and pull requests, and as a main point of trust, as github access now is a thing that can momentarily be revoked for no apparent reason.
+
+We encourage using the github if the main site is not accessible from your country and vice versa. We recommend to `publishToMavenLocal` and not depend on politics.
 
-Please visit the project homepage: [https://lynglang.com](https://lynglang.com) and a [telegram channel](https://t.me/lynglang) for updates.
 
 - simple, compact, intuitive and elegant modern code:
 
@@ -17,45 +25,48 @@ Point(x:, y:).dist() //< 5
 fun swapEnds(first, args..., last, f) {
     f( last, ...args, first)
 } 
+
+class A {
+    class B(x?)
+    object Inner { val foo = "bar" }
+    enum E* { One, Two }
+}
+val ab = A.B()
+assertEquals(null, ab.x)
+assertEquals("bar", A.Inner.foo)
+assertEquals(A.E.One, A.One)
 ```
 
 - extremely simple Kotlin integration on any platform (JVM, JS, WasmJS, Lunux, MacOS, iOS, Windows)
 - 100% secure: no access to any API you didn't explicitly provide
-- 100% coroutines! Every function/script is a coroutine, it does not block the thread, no async/await/suspend keyword garbage, see [parallelism]
-
-```
-  val deferred = launch {
-    delay(1.5) // coroutine is delayed for 1.5s, thread is not blocked!
-    "done"
-  }
-  // ...
-  // suspend current coroutine, no thread is blocked again,
-  // and wait for deferred to return something:
-  assertEquals("donw", deferred.await())
-```
-and it is multithreaded on platforms supporting it (automatically, no code changes required, just
-`launch` more coroutines and they will be executed concurrently if possible). See [parallelism]
-
-- functional style and OOP together, multiple inheritance, implementing interfaces for existing classes, writing extensions.
+- 100% coroutines! Every function/script is a coroutine, it does not block the thread, no async/await/suspend keyword garbage, see [parallelism]. it is multithreaded on platforms supporting it (automatically, no code changes required, just `launch` more coroutines and they will be executed concurrently if possible). See [parallelism]
+- functional style and OOP together: multiple inheritance (so you got it all - mixins, interfaces, etc.), delegation, sigletons, anonymous classes,extensions.
+- nice literals for maps and arrays, destructuring assignment, ranges.
 - Any Unicode letters can be used as identifiers: `assert( sin(π/2) == 1 )`.
 
  ## Resources: 
 
 - [Language home](https://lynglang.com)
 - [introduction and tutorial](docs/tutorial.md) - start here please
+- [Latest release notes (1.5.5)](docs/whats_new.md)
+- [What's New in 1.5](docs/whats_new_1_5.md)
 - [Testing and Assertions](docs/Testing.md)
+- [Filesystem and Processes (lyngio)](docs/lyngio.md)
+- [SQL Databases (lyng.io.db)](docs/lyng.io.db.md)
+- [Time and Calendar Types](docs/time.md)
+- [Return Statement](docs/return_statement.md)
 - [Efficient Iterables in Kotlin Interop](docs/EfficientIterables.md)
 - [Samples directory](docs/samples)
 - [Formatter (core + CLI + IDE)](docs/formatter.md)
 - [Books directory](docs)
+- [AI agent guidance](AGENTS.md)
 
 ## Integration in Kotlin multiplatform
 
 ### Add dependency to your project
 
 ```kotlin
-// update to current please:
-val lyngVersion = "0.6.1-SNAPSHOT"
+val lyngVersion = "1.5.5"
 
 repositories {
     // ...
@@ -79,28 +90,30 @@ Now you can import lyng and use it:
 ### Execute script:
 
 ```kotlin
-import net.sergeyh.lyng.*
+import net.sergeych.lyng.*
 
 // we need a coroutine to start, as Lyng
 // is a coroutine based language, async topdown
 runBlocking {
-    assert(5 == eval(""" 3*3 - 4 """).toInt())
-    eval(""" println("Hello, Lyng!") """)
+    val session = EvalSession()
+    assert(5 == session.eval(""" 3*3 - 4 """).toInt())
+    session.eval(""" println("Hello, Lyng!") """)
 }
 ```
 
 ### Exchanging information
 
-Script is executed over some `Scope`. Create instance,
-add your specific vars and functions to it, and call:
+The preferred host runtime is `EvalSession`. It owns the script scope and any coroutines
+started with `launch { ... }`. Create a session, grab its scope when you need low-level
+binding APIs, then execute scripts through the session:
 
 ```kotlin
+import net.sergeych.lyng.*
 
-import com.sun.source.tree.Scope
-import new.sergeych.lyng.*
-
+runBlocking {
+    val session = EvalSession()
+    val scope = session.getScope().apply {
         // simple function
-val scope = Script.newScope().apply {
         addFn("sumOf") {
             var sum = 0.0
             for (a in args) sum += a.toDouble()
@@ -118,10 +131,13 @@ val scope = Script.newScope().apply {
             doSomeWork(args[0]).toObj()
         }
     }
-// adding constant:
-scope.eval("sumOf(1,2,3)") // <- 6
+
+    // execute through the session:
+    session.eval("sumOf(1,2,3)") // <- 6
+}
 ```
-Note that the scope stores all changes in it so you can make calls on a single scope to preserve state between calls.
+Note that the session reuses one scope, so state persists across `session.eval(...)` calls.
+Use raw `Scope.eval(...)` only when you intentionally want low-level control without session-owned coroutine lifecycle.
 
 ## IntelliJ IDEA plugin: Lightweight autocompletion (experimental)
 
@@ -145,6 +161,12 @@ Tips:
 - After a dot, globals are intentionally suppressed (e.g., `lines().Path` is not valid), only the receiver’s members are suggested.
 - If completion seems sparse, make sure related modules are imported (e.g., `import lyng.io.fs` so that `Path` and its methods are known).
 
+## AI Assistant Support
+
+To help AI assistants (like Cursor, Windsurf, or GitHub Copilot) understand Lyng with minimal effort, we provide a high-density language specification:
+
+- **[LYNG_AI_SPEC.md](LYNG_AI_SPEC.md)**: A concise guide for AI models to learn Lyng syntax, idioms, and core philosophy. We recommend pointing your AI tool to this file or including it in your project's custom instructions.
+
 ## Why? 
 
 Designed to add scripting to kotlin multiplatform application in easy and efficient way. This is attempt to achieve what Lua is for C/++.
@@ -164,8 +186,7 @@ Designed to add scripting to kotlin multiplatform application in easy and effici
 
 # Language Roadmap
 
-We are now at **v1.0**: basic optimization performed, battery included: standard library is 90% here, initial
-support in HTML, popular editors, and IDEA; tools to syntax highlight and format code are ready. It was released closed to schedule.
+The current stable release is **v1.5.5**: the 1.5 cycle now includes the database/date/concurrency additions as well as the latest compiler/runtime stabilization work, and the language, tooling, and site are aligned around this release.
 
 Ready features:
 
@@ -176,6 +197,7 @@ Ready features:
 - [x] ranges, lists, strings, interfaces: Iterable, Iterator, Collection, Array
 - [x] when(value), if-then-else
 - [x] exception handling: throw, try-catch-finally, exception classes.
+- [x] user-defined exception classes
 - [x] multiplatform maven publication
 - [x] documentation for the current state
 - [x] maps, sets and sequences (flows?)
@@ -191,22 +213,31 @@ Ready features:
 - [x] regular exceptions + extended `when`
 - [x] multiple inheritance for user classes
 - [x] class properties (accessors)
+- [x] `return` statement for local and non-local exit
+- [x] Unified Delegation model: val, var and fun
+- [x] `lazy val` using delegation
+- [x] singletons `object TheOnly { ... }`
+- [x] object expressions `object: List { ... }`
+- [x] late-init vals in classes
+- [x] properties with getters and setters
+- [x] assign-if-null operator `?=`
+- [x] user-defined exception classes
 
-## plan: towards v1.5 Enhancing
+All of this is documented on the [language site](https://lynglang.com) and locally in [docs/tutorial.md](docs/tutorial.md). The site reflects the current release, while development snapshots continue in the private Maven repository.  
+
+## plan: towards v2.0 Next Generation
 
 - [x] site with integrated interpreter to give a try
 - [x] kotlin part public API good docs, integration focused
-- [ ] type specifications
+- [x] type specifications
 - [x] Textmate Bundle
 - [x] IDEA plugin
-- [ ] source docs and maybe lyng.md to a standard
-- [ ] metadata first class access from lyng
+- [x] source docs and maybe lyng.md to a standard
 - [x] aggressive optimizations
-- [ ] compile to JVM bytecode optimization
 
 ## After 1.5 "Ideal scripting"
 
-Estimated summer 2026
+* __we are here now ;)__
 
 - propose your feature! 
 
@@ -214,8 +245,12 @@ Estimated summer 2026
 
 @-links are for contacting authors on [project home](https://gitea.sergeych.net/SergeychWorks/lyng): this simplest s to open issue for the person you need to convey any information about this project.
 
-__Sergey Chernov__ @sergeych: Initial idea and architecture, language concept, design, implementation.
+<img src="https://www.gravatar.com/avatar/7e3a56ff8a090fc9ffbd1909dea94904?s=32&d=identicon" alt="Sergey Chernov" width="32" height="32" style="vertical-align: middle; margin-right: 0.5em;" /> <b>Sergey Chernov</b> @sergeych, real.sergeych@gmail.com: Initial idea and architecture, language concept, design, implementation.
+
+<br/>
+
+<img src="https://www.gravatar.com/avatar/53a90bca30c85a81db8f0c0d8dea43a1?s=32&d=identicon" alt="Yulia Nezhinskaya" width="32" height="32" style="vertical-align: middle; margin-right: 0.5em;" /> <b>Yulia Nezhinskaya</b> @AlterEgoJuliaN, neleka88@gmail.com: System analysis, math and feature design.
+
 
-__Yulia Nezhinskaya__ @AlterEgoJuliaN: System analysis, math and features design.
 
 [parallelism]: docs/parallelism.md

archived/README.md

@@ -0,0 +1,4 @@
+# Obsolete files
+
+
+__Do not rely on contents of the files in this directory. They are kept for historical reference only and may not be up-to-date or relevant.__

archived/migrate_time_to_2_2.md

@@ -4,7 +4,7 @@
 
 Before kotlin 2.0, there was an excellent library, kotlinx.datetime, which was widely used everywhere, also in Lyng and its dependencies.
 
-When kotlin 2.0 was released, or soon after, JetBrains made an exptic decision to remove `Instant` and `Clock` from kotlinx.datetime and replace it with _yet experimental_ analogs in `kotlin.time`.
+When Kotlin 2.0 was released, or soon after, JetBrains made a perplexing decision to remove `Instant` and `Clock` from kotlinx.datetime and replace it with _yet experimental_ analogs in `kotlin.time`.
 
 The problem is, these were not quite the same (these weren't `@Serializable`!), so people didn't migrate with ease. Okay, then JetBrains decided to not only deprecate it but also make them unusable on Apple targets. It sort of split auditories of many published libraries to those who hate JetBrains and Apple and continue to use 1.9-2.0 compatible versions that no longer work with Kotlin 2.2 on Apple targets (but work pretty well with earlier Kotlin or on other platforms). 
 
@@ -12,14 +12,14 @@ Later JetBrains added serializers for their new `Instant` and `Clock` types, but
 
 ## Solution
 
-We hereby publish a new version of Lyng, 1.0.8-SNAPSHOT, which uses `ktlin.time.Instant` and `kotlin.time.Clock` instead of `kotlinx.datetime.Instant` and `kotlinx.datetime.Clock; it is in other aspects compatible also with Lynon encoded binaries. Still you might need to migrate your code to use `kotlinx.datetime` types.
+We hereby publish a new version of Lyng, 1.0.8-SNAPSHOT, which uses `kotlin.time.Instant` and `kotlin.time.Clock` instead of `kotlinx.datetime.Instant` and `kotlinx.datetime.Clock`. It is in other aspects compatible also with Lynon encoded binaries. You might need to migrate your code to use `kotlin.time` types. (LocalDateTime/TimeZone still come from `kotlinx.datetime`.)
 
-So, if you are getting errors with new version, plase do:
+So, if you are getting errors with new version, please do:
 
 - upgrade to Kotlin 2.2
 - upgrade to Lyng 1.0.8-SNAPSHOT
 - replace in your code imports (or other uses) of `kotlinx.datetime.Clock` to `kotlin.time.Clock` and `kotlinx.datetime.Instant` to `kotlin.time.Instant`.
 
-This should solve the problem and hopefully we'll see no more suh a brillant ideas from IDEA ideologspersons.
+This should solve the problem and hopefully we'll see no more such "brilliant" ideas from IDEA ideologspersons.
 
-Sorry for inconvenicence and send a ray of hate to JetBrains ;)
+Sorry for inconvenience and send a ray of hate to JetBrains ;)

archived/proposals/delegates.lyng

@@ -0,0 +1,117 @@
+/*
+    This is a tech proposal under construction, please do not use it yet
+    for any purpose
+*/
+
+/*
+    Abstract delegate can be used to proxy read/wrtie field access
+    or method call. Default implementation reports error.
+*/
+interface Delegate {
+    fun getValue() = Unset
+    fun setValue(newValue) { throw NotImplementedException("delegate setter is not implemented") }
+    fun invoke(args...) { throw NotImplementedException("delegate setter is not implemented") }
+}
+
+/*
+    Delegate cam be used to implement a val, var or fun, so there are
+    access type enum to distinguish:
+*/
+enum DelegateAccess {
+    Val,
+    Var,
+    Callable
+}
+
+// Delegate can be associated by a val/var/fun in a declaraion site using `by` keyword
+
+val proxiedVal by proxy(1)
+var proxiedVar by proxy(2, 3)
+fun proxiedFun by proxy()
+
+// each proxy is a Lyng expression returning instance of the Proxy interface:
+
+/*
+    Proxy interface is connecting some named property of a given kind with the `Delegate`.
+    It removes the burden of dealing with property name and this ref on each get/set value
+    or invoke allowing having one delegate per instance, execution buff.
+*/
+interface Proxy {
+    fun getDelegate(propertyName: String,access: DelegateAccess,thisRef: Obj?): Delegate
+}
+
+// val, var and fun can be delegated, local or class instance:
+class TestProxy: Proxy {
+    override getDelegate(name,access,thisRef) {
+        Delegate()
+    }
+}
+
+val proxy = TestProxy()
+
+class Allowed {
+    val v1 by proxy
+    var v2 by proxy
+    fun f1 by proxy
+}
+val v3 by proxy
+var v4 by proxy
+fun f2 by proxy
+
+/*
+    It means that for example
+    Allowed().f1("foo")
+    would call a delegate.invoke("foo") on the `Delegate` instance supplied by `proxy`, etc.
+*/
+
+// The practic sample: lazy value
+
+/*
+    The delegate that caches single time evaluated value
+*/
+class LazyDelegate(creator): Delegate {
+    private var currentValue=Unset
+
+    override fun getValue() {
+        if( currentValue == Unset )
+            currentValue = creator()
+        currentValue
+    }
+}
+
+/*
+    The proxy to assign it
+*/
+class LazyProxy(creator) {
+    fun getDelegate(name,access,thisRef) {
+        if( access != DelegateAccess.Val )
+            throw IllegalArgumentException("only lazy val are allowed")
+        LazyDelegate(creator)
+    }
+}
+
+
+/*
+    A helper function to simplify creation:
+*/
+fun lazy(creator) {
+    LazyProxy(creator)
+}
+
+// Usage sample and the test:
+var callCounter = 0
+assertEquals(0, clallCounter)
+
+val lazyText by lazy { "evaluated text" }
+
+// the lazy property is not yet evaluated:
+assertEquals(0, clallCounter)
+// now evaluate it by using it:
+assertEquals("evaluated text", lazyText)
+assertEquals(1, callCounter)
+
+// lazy delegate should fail on vars or funs:
+assertThrows { var bad by lazy { "should not happen" } }
+assertThrows { fun bad by lazy { 42 } }
+
+

archived/wasm_generation_bug.md

@@ -0,0 +1,27 @@
+# Wasm generation hang in wasmJs browser tests
+
+## Summary
+The wasmJs browser test runner hung after commit 5f819dc. The root cause was invalid WebAssembly generated by the Kotlin/Wasm backend when certain compiler paths emitted suspend lambdas for `Statement` execution. The invalid module failed to instantiate in the browser, and Karma kept the browser connected but never ran tests.
+
+## Symptoms
+- `:lynglib:wasmJsBrowserTest` hangs indefinitely in ChromeHeadless.
+- `:lynglib:wasmJsNodeTest` fails with a WebAssembly compile error similar to:
+  - `struct.set expected type (ref null XXXX), found global.get of type (ref null YYYY)`
+- The failing function name in the wasm name section looks like:
+  - `net.sergeych.lyng.$invokeCOROUTINE$.doResume`
+
+## Root cause
+The delegation/var-declaration changes introduced compiler-generated suspend lambdas inside `Statement` construction (e.g., `statement { ... }` wrappers). Kotlin/Wasm generates extra coroutine state for those suspend lambdas, which in this case produced invalid wasm IR (mismatched GC reference types). The browser loader then waits forever because the module fails to instantiate.
+
+## Fix
+Avoid suspend-lambda `Statement` construction in compiler code paths. Replace `statement { ... }` and other anonymous suspend lambdas with explicit `object : Statement()` implementations and move logic into `override suspend fun execute(...)`. This keeps the resulting wasm IR valid while preserving behavior.
+
+## Where it was fixed
+- `lynglib/src/commonMain/kotlin/net/sergeych/lyng/Compiler.kt`
+- `lynglib/src/commonMain/kotlin/net/sergeych/lyng/Scope.kt`
+
+## Verification
+- `./gradlew :lynglib:wasmJsNodeTest --info`
+- `./gradlew :lynglib:wasmJsBrowserTest --info`
+
+Both tests finish quickly after the change.

bin/deploy_all

@@ -0,0 +1,39 @@
+#!/bin/bash
+
+#
+# Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+#
+
+set -e
+echo "publishing all artifacts"
+echo
+./gradlew publishToMavenLocal site:jsBrowserDistribution publish buildInstallablePlugin :lyng:linkReleaseExecutableLinuxX64 :lyng:installJvmDist --parallel --no-configuration-cache
+
+#echo
+#echo "Creating plugin"
+#echo
+#./gradlew buildInstallablePlugin
+
+echo
+echo "building CLI tools"
+echo
+bin/local_jrelease
+bin/local_release
+
+echo
+echo "Deploying site"
+echo
+./bin/deploy_site

bin/deploy_plugin

@@ -0,0 +1,26 @@
+#!/bin/sh
+
+#
+# Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+#
+
+set -e
+
+echo
+echo "Creating plugin"
+echo
+./gradlew buildInstallablePlugin
+deploy_site -u

bin/deploy_site

@@ -1,7 +1,7 @@
 #!/bin/bash
 
 #
-# Copyright 2025 Sergey S. Chernov real.sergeych@gmail.com
+# Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,6 +17,16 @@
 #
 #
 
+upload_only=false
+target=vps   # default: new server; use --old for d.lynglang.com
+for arg in "$@"; do
+  if [[ "$arg" == "-u" || "$arg" == "--upload-only" ]]; then
+    upload_only=true
+  elif [[ "$arg" == "--old" ]]; then
+    target=com
+  fi
+done
+
 function checkState() {
   if [[ $? != 0 ]]; then
     echo
@@ -24,9 +34,10 @@ function checkState() {
     echo
     exit 100
   fi
-
 }
 
+
+
 # Update docs/idea_plugin.md to point to the latest built IDEA plugin zip
 # from ./distributables before building the site. The change is temporary and
 # the original file is restored right after the build.
@@ -79,19 +90,20 @@ function updateIdeaPluginDownloadLink() {
   fi
 }
 
-# default target settings
-case "com" in
+# target settings  (-t com | -t vps)
+case "$target" in
   com)
-    SSH_HOST=sergeych@d.lynglang.com # host to deploy to
-    SSH_PORT=22                             # ssh port on it
-    ROOT=/bigstore/sergeych_pub/lyng                 # directory to rsync to
+    SSH_HOST=sergeych@d.lynglang.com
+    SSH_PORT=22
+    ROOT=/bigstore/sergeych_pub/lyng
+    ;;
+  vps)
+    SSH_HOST=sergeych@94.130.36.94
+    SSH_PORT=22
+    ROOT=/var/www/lynglang
     ;;
-#  com)
-#    SSH_HOST=vvk@front-01.neurodatalab.com
-#    ROOT=/home/vvk
-#    ;;
   *)
-    echo "*** ERROR: target not specified (use deploy com | dev)"
+    echo "*** ERROR: unknown target '$target' (use -t com | -t vps)"
     echo "*** stop"
     exit 101
 esac
@@ -107,6 +119,8 @@ function refreshTextmateZip() {
   (cd editors && zip -rq ../distributables/lyng-textmate.zip .)
 }
 
+
+if [[ "$upload_only" == false ]]; then
   # Update the IDEA plugin download link in docs (temporarily), then build, then restore the doc
   refreshTextmateZip
   updateIdeaPluginDownloadLink || echo "WARN: proceeding without updating IDEA plugin download link"
@@ -125,9 +139,7 @@ if [[ $BUILD_RC -ne 0 ]]; then
     echo
     exit 100
   fi
-
-
-#exit 0
+fi
 
 # Prepare working dir
 ssh -p ${SSH_PORT} ${SSH_HOST} "
@@ -143,12 +155,15 @@ ssh -p ${SSH_PORT} ${SSH_HOST} "
   fi
 ";
 
+if [[ "$upload_only" == false ]]; then
   # sync files
   SRC=./site/build/dist/js/productionExecutable
   rsync -e "ssh -p ${SSH_PORT}" -avz -r -d --delete ${SRC}/* ${SSH_HOST}:${ROOT}/build/dist
   checkState
   #rsync -e "ssh -p ${SSH_PORT}" -avz  ./static/* ${SSH_HOST}:${ROOT}/build/dist
   #checkState
+fi
+
 rsync -e "ssh -p ${SSH_PORT}" -avz -r -d --delete distributables/* ${SSH_HOST}:${ROOT}/build/dist/distributables
 checkState
 

bin/local_jrelease

@@ -1,7 +1,7 @@
 #!/bin/bash
 
 #
-# Copyright 2025 Sergey S. Chernov real.sergeych@gmail.com
+# Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -19,13 +19,15 @@
 
 set -e
 
-root=./lyng/build/install/lyng-jvm/
+archive=./lyng/build/distributions/lyng-jvm.zip
+install_root="$HOME/bin/jlyng-jvm"
+launcher="$install_root/lyng-jvm/bin/lyng"
 
-./gradlew :lyng:installJvmDist
-#strip $file
-#upx   $file
-rm -rf ~/bin/jlyng-jvm || true
-rm ~/bin/jlyng 2>/dev/null || true
-mkdir -p ~/bin/jlyng-jvm
-cp -R $root ~/bin/jlyng-jvm
-ln -s ~/bin/jlyng-jvm/lyng-jvm/bin/lyng ~/bin/jlyng
+./gradlew :lyng:jvmDistZip
+mkdir -p ./distributables
+cp "$archive" ./distributables/lyng-jvm.zip
+rm -rf "$install_root" || true
+rm "$HOME/bin/jlyng" 2>/dev/null || true
+mkdir -p "$install_root"
+unzip -q ./distributables/lyng-jvm.zip -d "$install_root"
+ln -s "$launcher" "$HOME/bin/jlyng"

build.gradle.kts

@@ -35,3 +35,27 @@ tasks.register<Exec>("generateDocs") {
     description = "Generates a single-file documentation HTML using bin/generate_docs.sh"
     commandLine("./bin/generate_docs.sh")
 }
+
+// Sample generator task for .lyng.d definition files (not wired into build).
+// Usage: ./gradlew generateLyngDefsSample
+tasks.register("generateLyngDefsSample") {
+    group = "lyng"
+    description = "Generate a sample .lyng.d file under build/generated/lyng/defs"
+    outputs.dir(layout.buildDirectory.dir("generated/lyng/defs"))
+    doLast {
+        val outDir = layout.buildDirectory.dir("generated/lyng/defs").get().asFile
+        outDir.mkdirs()
+        val outFile = outDir.resolve("sample.lyng.d")
+        outFile.writeText(
+            """
+            /** Generated API */
+            extern fun ping(): Int
+
+            /** Generated class */
+            class Generated(val name: String) {
+                fun greet(): String = "hi " + name
+            }
+            """.trimIndent()
+        )
+    }
+}

docs/Array.md

@@ -1,8 +1,15 @@
 # Array
 
 It's an interface if the [Collection] that provides indexing access, like `array[3] = 0`.
-Array therefore implements [Iterable] too. The well known implementatino of the `Array` is
-[List].
+Array therefore implements [Iterable] too. Well known implementations of `Array` are
+[List] and [ImmutableList].
+
+The language-level bracket syntax supports one or more selectors:
+
+- `value[i]`
+- `value[i, j]`
+
+Concrete array-like types decide what selectors they accept. Built-in list-like arrays use one selector at a time; custom types such as matrices may interpret multiple selectors.
 
 Array adds the following methods:
 
@@ -35,3 +42,4 @@ To pre-sort and array use `Iterable.sorted*` or in-place `List.sort*` families,
 [Collection]: Collection.md
 [Iterable]: Iterable.md
 [List]: List.md
+[ImmutableList]: ImmutableList.md

docs/Buffer.md

@@ -23,11 +23,11 @@ There are a lo of ways to construct a buffer:
     assertEquals( 5, Buffer("hello").size )
 
     // from bytes, e.g. integers in range 0..255
-    assertEquals( 255, Buffer(1,2,3,255).last() )
+    assertEquals( 255, Buffer(1,2,3,255).last )
 
     // from whatever iterable that produces bytes, e.g.
     // integers in 0..255 range:
-    assertEquals( 129, Buffer([1,2,129]).last() )
+    assertEquals( 129, Buffer([1,2,129]).last )
 
     // Empty buffer of fixed size:
     assertEquals(100, Buffer(100).size)
@@ -121,7 +121,7 @@ which is used in `toString`) and hex encoding:
 ## Members
 
 | name                       | meaning                                        | type          |
-|----------------------------|-----------------------------------------|---------------|
+|----------------------------|------------------------------------------------|---------------|
 | `size`                     | size                                           | Int           |
 | `decodeUtf8`               | decode to String using UTF8 rules              | Any           |
 | `+`                        | buffer concatenation                           | Any           |
@@ -129,6 +129,7 @@ which is used in `toString`) and hex encoding:
 | `hex`                      | encode to hex strign                           | String        |
 | `Buffer.decodeHex(hexStr)  | decode hex string                              | Buffer        |
 | `base64`                   | encode to base64 (url flavor) (2)              | String        |
+| `base64std`                | encode to base64 (default vocabulary, filling) | String        |
 | `Buffer.decodeBase64(str)` | decode base64 to new Buffer (2)                | Buffer        |
 | `toBitInput()`             | create bit input from a byte buffer (3)        |               |
 

docs/BytecodeSpec.md

@@ -0,0 +1,280 @@
+# Lyng Bytecode VM Spec v0 (Draft)
+
+This document describes a register-like (3-address) bytecode for Lyng with
+dynamic slot width (8/16/32-bit slot IDs), a slot-tail argument model, and
+typed lanes for Obj/Int/Real/Bool. The VM is intended to run as a suspendable
+interpreter and fall back to the existing AST execution when needed.
+
+## 1) Frame & Slot Model
+
+### Frame metadata
+- localCount: number of local slots for this function (fixed at compile time).
+- argCount: number of arguments passed at call time.
+- scopeSlotNames: optional debug names for scope slots (locals/params), aligned to slot mapping.
+- argBase = localCount.
+
+### Slot layout
+slots[0 .. localCount-1]              locals
+slots[localCount .. localCount+argCount-1]  arguments
+
+### Typed lanes
+- slotType[]: UNKNOWN/OBJ/INT/REAL/BOOL
+- objSlots[], intSlots[], realSlots[], boolSlots[]
+- A slot is a logical index; active lane is selected by slotType.
+
+### Parameter access
+- param i => slot localCount + i
+- variadic extra => slot localCount + declaredParamCount + k
+
+### Debug metadata (optional)
+- scopeSlotNames: array sized scopeSlotCount, each entry nullable.
+- Intended for disassembly/debug tooling; VM semantics do not depend on it.
+
+### Constant pool extras
+- SlotPlan: map of name -> slot index, used by PUSH_SCOPE to pre-allocate and map loop locals.
+- CallArgsPlan: ordered argument specs (name/splat) + tailBlock flag, used when argCount has the plan flag set.
+
+## 2) Slot ID Width
+
+Per frame, select:
+- 8-bit if localCount + argCount < 256
+- 16-bit if < 65536
+- 32-bit otherwise
+
+The decoder uses a dedicated loop per width. All slot operands are expanded to
+Int internally.
+
+## 3) CALL Semantics (Model A)
+
+Instruction:
+CALL_DIRECT fnId, argBase, argCount, dst
+
+Behavior:
+- Allocate a callee frame sized localCount + argCount.
+- Copy caller slots [argBase .. argBase+argCount-1] into callee slots
+  [localCount .. localCount+argCount-1].
+- Callee returns via RET slot or RET_VOID.
+- Caller stores return value to dst.
+
+Other calls:
+- CALL_VIRTUAL recvSlot, methodId, argBase, argCount, dst
+- CALL_FALLBACK stmtId, argBase, argCount, dst
+- CALL_SLOT calleeSlot, argBase, argCount, dst
+
+## 4) Binary Encoding Layout
+
+All instructions are:
+  [opcode:U8] [operands...]
+
+Operand widths:
+- slotId: S = 1/2/4 bytes (per frame slot width)
+- constId: K = 2 bytes (U16), extend to 4 if needed
+- ip: I = 2 bytes (U16) or 4 bytes (U32) per function size
+- fnId/methodId/stmtId: F/M/T = 2 bytes (U16) unless extended
+- argCount: C = 2 bytes (U16), extend to 4 if needed
+
+Endianness: little-endian for multi-byte operands.
+
+Common operand patterns:
+- S: one slot
+- SS: two slots
+- SSS: three slots
+- K S: constId + dst slot
+- S I: slot + jump target
+- I: jump target
+- F S C S: fnId, argBase slot, argCount, dst slot
+
+Arg count flag:
+- If high bit of C is set (0x8000), the low 15 bits encode a CallArgsPlan constId.
+- When not set, C is the raw positional count and tailBlockMode=false.
+
+## 5) Opcode Table
+
+Note: Any opcode can be compiled to FALLBACK if not implemented in a VM pass.
+
+### Data movement
+- NOP
+- MOVE_OBJ S -> S
+- MOVE_INT S -> S
+- MOVE_REAL S -> S
+- MOVE_BOOL S -> S
+- BOX_OBJ S -> S
+- CONST_OBJ K -> S
+- CONST_INT K -> S
+- CONST_REAL K -> S
+- CONST_BOOL K -> S
+- CONST_NULL -> S
+
+### Numeric conversions
+- INT_TO_REAL S -> S
+- REAL_TO_INT S -> S
+- BOOL_TO_INT S -> S
+- INT_TO_BOOL S -> S
+
+### Arithmetic: INT
+- ADD_INT S, S -> S
+- SUB_INT S, S -> S
+- MUL_INT S, S -> S
+- DIV_INT S, S -> S
+- MOD_INT S, S -> S
+- NEG_INT S -> S
+- INC_INT S
+- DEC_INT S
+
+### Arithmetic: REAL
+- ADD_REAL S, S -> S
+- SUB_REAL S, S -> S
+- MUL_REAL S, S -> S
+- DIV_REAL S, S -> S
+- NEG_REAL S -> S
+
+### Arithmetic: OBJ
+- ADD_OBJ S, S -> S
+- SUB_OBJ S, S -> S
+- MUL_OBJ S, S -> S
+- DIV_OBJ S, S -> S
+- MOD_OBJ S, S -> S
+
+### Bitwise: INT
+- AND_INT S, S -> S
+- OR_INT S, S -> S
+- XOR_INT S, S -> S
+- SHL_INT S, S -> S
+- SHR_INT S, S -> S
+- USHR_INT S, S -> S
+- INV_INT S -> S
+
+### Comparisons (typed)
+- CMP_EQ_INT S, S -> S
+- CMP_NEQ_INT S, S -> S
+- CMP_LT_INT S, S -> S
+- CMP_LTE_INT S, S -> S
+- CMP_GT_INT S, S -> S
+- CMP_GTE_INT S, S -> S
+- CMP_EQ_REAL S, S -> S
+- CMP_NEQ_REAL S, S -> S
+- CMP_LT_REAL S, S -> S
+- CMP_LTE_REAL S, S -> S
+- CMP_GT_REAL S, S -> S
+- CMP_GTE_REAL S, S -> S
+- CMP_EQ_BOOL S, S -> S
+- CMP_NEQ_BOOL S, S -> S
+
+### Mixed numeric comparisons
+- CMP_EQ_INT_REAL S, S -> S
+- CMP_EQ_REAL_INT S, S -> S
+- CMP_LT_INT_REAL S, S -> S
+- CMP_LT_REAL_INT S, S -> S
+- CMP_LTE_INT_REAL S, S -> S
+- CMP_LTE_REAL_INT S, S -> S
+- CMP_GT_INT_REAL S, S -> S
+- CMP_GT_REAL_INT S, S -> S
+- CMP_GTE_INT_REAL S, S -> S
+- CMP_GTE_REAL_INT S, S -> S
+- CMP_NEQ_INT_REAL S, S -> S
+- CMP_NEQ_REAL_INT S, S -> S
+- CMP_EQ_OBJ S, S -> S
+- CMP_NEQ_OBJ S, S -> S
+- CMP_REF_EQ_OBJ S, S -> S
+- CMP_REF_NEQ_OBJ S, S -> S
+- CMP_LT_OBJ S, S -> S
+- CMP_LTE_OBJ S, S -> S
+- CMP_GT_OBJ S, S -> S
+- CMP_GTE_OBJ S, S -> S
+
+### Boolean ops
+- NOT_BOOL S -> S
+- AND_BOOL S, S -> S
+- OR_BOOL S, S -> S
+
+### Control flow
+- JMP I
+- JMP_IF_TRUE S, I
+- JMP_IF_FALSE S, I
+- RET S
+- RET_VOID
+- PUSH_SCOPE K
+- POP_SCOPE
+
+### Scope setup
+- PUSH_SCOPE uses const `SlotPlan` (name -> slot index) to create a child scope and apply slot mapping.
+- POP_SCOPE restores the parent scope.
+
+### Calls
+- CALL_DIRECT F, S, C, S
+- CALL_VIRTUAL S, M, S, C, S
+- CALL_FALLBACK T, S, C, S
+- CALL_SLOT S, S, C, S
+
+### Object access (optional, later)
+- GET_FIELD S, M -> S
+- SET_FIELD S, M, S
+- GET_INDEX S, S -> S
+- SET_INDEX S, S, S
+
+### Fallback
+- EVAL_FALLBACK T -> S
+
+## 6) Const Pool Encoding (v0)
+
+Each const entry is encoded as:
+  [tag:U8] [payload...]
+
+Tags:
+- 0x00: NULL
+- 0x01: BOOL (payload: U8 0/1)
+- 0x02: INT (payload: S64, little-endian)
+- 0x03: REAL (payload: F64, IEEE-754, little-endian)
+- 0x04: STRING (payload: U32 length + UTF-8 bytes)
+- 0x05: OBJ_REF (payload: U32 index into external Obj table)
+
+Notes:
+- OBJ_REF is reserved for embedding prebuilt Obj handles if needed.
+- Strings use UTF-8; length is bytes, not chars.
+
+## 7) Function Header (binary container)
+
+Suggested layout for a bytecode function blob:
+- magic: U32 ("LYBC")
+- version: U16 (0x0001)
+- slotWidth: U8 (1,2,4)
+- ipWidth: U8 (2,4)
+- constIdWidth: U8 (2,4)
+- localCount: U32
+- codeSize: U32 (bytes)
+- constCount: U32
+- constPool: [const entries...]
+- code: [bytecode...]
+
+Const pool entries use the encoding described in section 6.
+
+## 8) Sample Bytecode (illustrative)
+
+Example Lyng:
+  val x = 2
+  val y = 3
+  val z = x + y
+
+Assume:
+- localCount = 3 (x,y,z)
+- argCount = 0
+- slot width = 1 byte
+- const pool: [INT 2, INT 3]
+
+Bytecode:
+  CONST_INT k0 -> s0
+  CONST_INT k1 -> s1
+  ADD_INT s0, s1 -> s2
+  RET_VOID
+
+Encoded (opcode values symbolic):
+  [OP_CONST_INT][k0][s0]
+  [OP_CONST_INT][k1][s1]
+  [OP_ADD_INT][s0][s1][s2]
+  [OP_RET_VOID]
+
+## 9) Notes
+
+- Mixed-mode is allowed: compiler can emit FALLBACK ops for unsupported nodes.
+- The VM must be suspendable; on suspension, store ip + minimal operand state.
+- Source mapping uses a separate ip->Pos table, not part of core bytecode.

docs/Channel.md

@@ -0,0 +1,211 @@
+# Channel
+
+A `Channel` is a **hot, bidirectional pipe** for passing values between concurrently running coroutines.
+Unlike a [Flow], which is cold and replayed on every collection, a `Channel` is stateful: each value
+sent is consumed by exactly one receiver.
+
+Channels model the classic _producer / consumer_ pattern and are the right tool when:
+
+- two or more coroutines need to exchange individual values at their own pace;
+- you want back-pressure (rendezvous) or explicit buffering control;
+- you need a push-based, hot data source (opposite of the pull-based, cold [Flow]).
+
+## Constructors
+
+```
+Channel()            // rendezvous — sender and receiver must meet
+Channel(n: Int)      // buffered — sender may run n items ahead of the receiver
+Channel(Channel.UNLIMITED)  // no limit on buffered items
+```
+
+**Rendezvous** (`Channel()`, capacity 0): `send` suspends until a matching `receive` is ready,
+and vice-versa. This gives the tightest synchronisation and the smallest memory footprint.
+
+**Buffered** (`Channel(n)`): `send` only suspends when the internal buffer is full. Allows the
+producer to get up to _n_ items ahead of the consumer.
+
+**Unlimited** (`Channel(Channel.UNLIMITED)`): `send` never suspends. Useful when the producer is
+bursty and you do not want it blocked, but be careful not to grow the buffer without bound.
+
+## Sending and receiving
+
+```lyng
+val ch = Channel()         // rendezvous channel
+
+val producer = launch {
+    ch.send("hello")       // suspends until the receiver is ready
+    ch.send("world")
+    ch.close()             // signal: no more values
+}
+
+val a = ch.receive()       // suspends until "hello" arrives
+val b = ch.receive()       // suspends until "world" arrives
+val c = ch.receive()       // channel is closed and drained → null
+assertEquals("hello", a)
+assertEquals("world", b)
+assertEquals(null,    c)
+```
+
+`receive()` returns `null` when the channel is both **closed** _and_ **fully drained** — that is
+the idiomatic loop termination condition:
+
+```lyng
+val ch = Channel(4)
+
+launch {
+    for (i in 1..5) ch.send(i)
+    ch.close()
+}
+
+var item = ch.receive()
+while (item != null) {
+    println(item)
+    item = ch.receive()
+}
+```
+
+## Non-suspending poll
+
+`tryReceive()` never suspends. It returns the next buffered value, or `null` if the buffer is
+empty or the channel is closed.
+
+```lyng
+val ch = Channel(8)
+ch.send(42)
+println(ch.tryReceive())   // 42
+println(ch.tryReceive())   // null — nothing buffered right now
+```
+
+Use `tryReceive` for _polling_ patterns where blocking would be unacceptable, for example when
+combining channel checks with other work inside a coroutine loop.
+
+## Closing a channel
+
+`close()` marks the channel so that no further `send` calls are accepted. Any items already in the
+buffer can still be received. Once the buffer is drained, `receive()` returns `null` and
+`isClosedForReceive` becomes `true`.
+
+```lyng
+val ch = Channel(2)
+ch.send(1)
+ch.send(2)
+ch.close()
+
+assert(ch.isClosedForSend)
+assert(!ch.isClosedForReceive)   // still has 2 buffered items
+
+ch.receive()  // 1
+ch.receive()  // 2
+assert(ch.isClosedForReceive)    // drained
+```
+
+Calling `send` after `close()` throws `IllegalStateException`.
+
+## Properties
+
+| property            | type   | description                                              |
+|---------------------|--------|----------------------------------------------------------|
+| `isClosedForSend`   | `Bool` | `true` after `close()` is called                         |
+| `isClosedForReceive`| `Bool` | `true` when closed _and_ every buffered item is consumed |
+
+## Methods
+
+| method          | suspends | description                                                                      |
+|-----------------|----------|----------------------------------------------------------------------------------|
+| `send(value)`   | yes      | send a value; suspends when buffer full (rendezvous: always until partner ready) |
+| `receive()`     | yes      | receive next value; suspends when empty; returns `null` when closed + drained    |
+| `tryReceive()`  | no       | return next buffered value or `null`; never suspends                             |
+| `close()`       | no       | signal end of production; existing buffer items are still receivable             |
+
+## Static constants
+
+| constant            | value            | description                         |
+|---------------------|------------------|-------------------------------------|
+| `Channel.UNLIMITED` | `Int.MAX_VALUE`  | capacity for an unlimited-buffer channel |
+
+## Common patterns
+
+### Producer / consumer
+
+```lyng
+val ch = Channel()
+val results = []
+val mu = Mutex()
+
+val consumer = launch {
+    var item = ch.receive()
+    while (item != null) {
+        mu.withLock { results += item }
+        item = ch.receive()
+    }
+}
+
+launch {
+    for (i in 1..5) ch.send("msg:$i")
+    ch.close()
+}.await()
+
+consumer.await()
+println(results)
+```
+
+### Fan-out: one channel, many consumers
+
+```lyng
+val ch = Channel(16)
+
+// multiple consumers
+val workers = (1..4).map { id ->
+    launch {
+        var task = ch.receive()
+        while (task != null) {
+            println("worker $id handles $task")
+            task = ch.receive()
+        }
+    }
+}
+
+// single producer
+for (i in 1..20) ch.send(i)
+ch.close()
+
+workers.forEach { it.await() }
+```
+
+### Ping-pong between two coroutines
+
+```lyng
+val ping = Channel()
+val pong = Channel()
+
+launch {
+    repeat(3) {
+        val msg = ping.receive()
+        println("got: $msg → sending pong")
+        pong.send("pong")
+    }
+}
+
+repeat(3) {
+    ping.send("ping")
+    println(pong.receive())
+}
+```
+
+## Channel vs Flow
+
+| | [Flow] | Channel |
+|---|---|---|
+| **temperature** | cold (lazy) | hot (eager) |
+| **replay** | every collector gets a fresh run | each item is consumed once |
+| **consumers** | any number; each gets all items | one receiver per item |
+| **back-pressure** | built-in via rendezvous | configurable (rendezvous / buffered / unlimited) |
+| **typical use** | transform pipelines, sequences | producer–consumer, fan-out |
+
+## See also
+
+- [parallelism] — `launch`, `Deferred`, `Mutex`, `Flow`, and the full concurrency picture
+- [Flow] — cold async sequences
+
+[Flow]: parallelism.md#flow
+[parallelism]: parallelism.md

docs/Collection.md

@@ -6,6 +6,13 @@ Is a [Iterable] with known `size`, a finite [Iterable]:
         val size
     }
 
+`Collection` is a read/traversal contract shared by mutable and immutable collections.
+Concrete collection classes:
+
+- Mutable: [List], [Set], [Map]
+- Immutable: [ImmutableList], [ImmutableSet], [ImmutableMap]
+- Observable mutable lists (opt-in module): [ObservableList]
+
 | name                   | description                                          |
 |------------------------|------------------------------------------------------|
 
@@ -17,3 +24,8 @@ See [List], [Set], [Iterable] and [Efficient Iterables in Kotlin Interop](Effici
 [Iterable]: Iterable.md
 [List]: List.md
 [Set]: Set.md
+[Map]: Map.md
+[ImmutableList]: ImmutableList.md
+[ImmutableSet]: ImmutableSet.md
+[ImmutableMap]: ImmutableMap.md
+[ObservableList]: ObservableList.md

docs/Complex.md

@@ -0,0 +1,82 @@
+# Complex Numbers (`lyng.complex`)
+
+`lyng.complex` adds a pure-Lyng `Complex` type backed by `Real` components.
+
+Import it when you want ordinary complex arithmetic:
+
+```lyng
+import lyng.complex
+```
+
+## Construction
+
+Use any of these:
+
+```lyng
+import lyng.complex
+
+val a = Complex(1.0, 2.0)
+val b = complex(1.0, 2.0)
+val c = 2.i
+val d = 3.re
+
+assertEquals(Complex(1.0, 2.0), 1 + 2.i)
+```
+
+Convenience extensions:
+
+- `Int.re`, `Real.re`: embed a real value into the complex plane
+- `Int.i`, `Real.i`: create a pure imaginary value
+- `cis(angle)`: shorthand for `cos(angle) + i sin(angle)`
+
+## Core Operations
+
+`Complex` supports:
+
+- `+`
+- `-`
+- `*`
+- `/`
+- unary `-`
+- `conjugate`
+- `magnitude`
+- `phase`
+
+Mixed arithmetic with `Int` and `Real` is enabled through `lyng.operators`, so both sides work naturally:
+
+```lyng
+import lyng.complex
+
+assertEquals(Complex(1.0, 2.0), 1 + 2.i)
+assertEquals(Complex(1.5, 2.0), 1.5 + 2.i)
+assertEquals(Complex(2.0, 2.0), 2.i + 2)
+```
+
+Mixed equality with built-in numeric types is intentionally not promised yet. Keep equality checks in the `Complex` domain for now.
+
+## Transcendental Functions
+
+For now, use member-style calls:
+
+```lyng
+import lyng.complex
+
+val z = 1 + π.i
+val w = z.exp()
+val s = z.sin()
+val r = z.sqrt()
+```
+
+This is deliberate. Lyng already has built-in top-level real-valued functions such as `exp(x)` and `sin(x)`, and imported modules do not currently replace those root bindings. So plain `exp(z)` is not yet the right extension mechanism for complex math.
+
+## Design Scope
+
+This module intentionally uses `Complex` with `Real` parts, not `Complex<T>`.
+
+Reasons:
+
+- the existing math runtime is `Real`-centric
+- the operator interop registry works with concrete runtime classes
+- transcendental functions (`exp`, `sin`, `ln`, `sqrt`) are defined over the `Real` math backend here
+
+If Lyng later gets a more general numeric-trait or callable-overload registry, a generic algebraic `Complex<T>` can be revisited on firmer ground.

docs/Decimal.md

@@ -0,0 +1,325 @@
+# Decimal (`lyng.decimal`)
+
+`lyng.decimal` adds an arbitrary-precision decimal type to Lyng as a normal library module.
+
+Import it when you need decimal arithmetic that should not inherit `Real`'s binary floating-point behavior:
+
+```lyng
+import lyng.decimal
+```
+
+## What `Decimal` Is For
+
+Use `Decimal` when values are fundamentally decimal:
+
+- money
+- human-entered quantities
+- exact decimal text
+- predictable decimal rounding
+- user-facing formatting and tests
+
+Do not use it just because a number has a fractional part. `Real` is still the right type for ordinary double-precision numeric work.
+
+## Creating Decimal Values
+
+There are three supported conversions:
+
+```lyng
+import lyng.decimal
+
+val a = 1.d
+val b = 2.2.d
+val c = "2.2".d
+
+assertEquals("1", a.toStringExpanded())
+assertEquals("2.2", b.toStringExpanded())
+assertEquals("2.2", c.toStringExpanded())
+```
+
+The three forms mean different things:
+
+- `1.d`: convert `Int -> Decimal`
+- `2.2.d`: convert `Real -> Decimal`
+- `"2.2".d`: parse exact decimal text
+
+That distinction is intentional.
+
+### `Real.d` vs `"..." .d`
+
+`Real.d` preserves the current `Real` value. It does not pretend the source code was exact decimal text.
+
+```lyng
+import lyng.decimal
+
+assertEquals("0.30000000000000004", (0.1 + 0.2).d.toStringExpanded())
+assertEquals("0.3", "0.3".d.toStringExpanded())
+```
+
+This follows the "minimal confusion" rule:
+
+- if you start from a `Real`, you get a decimal representation of that `Real`
+- if you want exact decimal source text, use a `String`
+
+## Factory Functions
+
+The explicit factory methods are:
+
+```lyng
+import lyng.decimal
+
+Decimal.fromInt(10)
+Decimal.fromReal(2.5)
+Decimal.fromString("12.34")
+```
+
+These are equivalent to the conversion-property forms, but sometimes clearer in APIs or generated code.
+
+## From Kotlin
+
+If you already have an ionspin `BigDecimal` on the host side, the simplest supported way to create a Lyng `Decimal` is:
+
+```kotlin
+import com.ionspin.kotlin.bignum.decimal.BigDecimal
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.asFacade
+import net.sergeych.lyng.newDecimal
+
+val scope = EvalSession().getScope()
+val decimal = scope.asFacade().newDecimal(BigDecimal.parseStringWithMode("12.34"))
+```
+
+Notes:
+
+- `newDecimal(...)` loads `lyng.decimal` if needed
+- it returns a real Lyng `Decimal` object instance
+- this is the preferred Kotlin-side construction path when you already hold a host `BigDecimal`
+
+## Core Operations
+
+`Decimal` supports:
+
+- `+`
+- `-`
+- `*`
+- `/`
+- `%`
+- unary `-`
+- comparison operators
+- equality operators
+
+Examples:
+
+```lyng
+import lyng.decimal
+
+assertEquals("3.75", ("1.5".d + "2.25".d).toStringExpanded())
+assertEquals("1.25", ("2.5".d - "1.25".d).toStringExpanded())
+assertEquals("3.0", ("1.5".d * 2.d).toStringExpanded())
+assertEquals("0.5", (1.d / 2.d).toStringExpanded())
+assert("2.0".d > "1.5".d)
+assert("2.0".d == 2.d)
+```
+
+## Interoperability With `Int` and `Real`
+
+The decimal module registers mixed-operand operator bridges so both sides read naturally:
+
+```lyng
+import lyng.decimal
+
+assertEquals(3.d, 1 + 2.d)
+assertEquals(3.d, 2.d + 1)
+assertEquals(1.5.d, 1.d + 0.5)
+assertEquals(1.5.d, 0.5 + 1.d)
+assert(2 == 2.d)
+assert(3 > 2.d)
+```
+
+Without this registration mechanism, only the cases directly implemented on the left-hand class would work. The bridge fills the gap for expressions such as `Int + Decimal` and `Real + Decimal`.
+
+See [OperatorInterop.md](OperatorInterop.md) for the generic mechanism behind that.
+
+## String Representation
+
+Use `toStringExpanded()` when you want plain decimal output without scientific notation:
+
+```lyng
+import lyng.decimal
+
+assertEquals("12.34", "12.34".d.toStringExpanded())
+```
+
+This is the recommended representation for:
+
+- tests
+- user-visible diagnostics
+- decimal formatting checks
+
+## Conversions Back To Built-ins
+
+```lyng
+import lyng.decimal
+
+assertEquals(2, "2.9".d.toInt())
+assertEquals(2.9, "2.9".d.toReal())
+```
+
+Use `toReal()` only when you are willing to return to binary floating-point semantics.
+
+## Non-Finite Checks
+
+`Decimal` values are always finite, so these helpers exist for API symmetry with `Real` and always return `false`:
+
+```lyng
+import lyng.decimal
+
+assertEquals(false, "2.9".d.isInfinite())
+assertEquals(false, "2.9".d.isNaN())
+```
+
+## Division Context
+
+Division is the operation where precision and rounding matter most.
+
+By default, decimal division uses:
+
+- precision: `34` significant digits
+- rounding: `HalfEven`
+
+Example:
+
+```lyng
+import lyng.decimal
+
+assertEquals("0.3333333333333333333333333333333333", (1.d / 3.d).toStringExpanded())
+assertEquals("0.6666666666666666666666666666666667", ("2".d / 3.d).toStringExpanded())
+```
+
+## `withDecimalContext(...)`
+
+Use `withDecimalContext(...)` to override decimal division rules inside a block:
+
+```lyng
+import lyng.decimal
+
+assertEquals(
+    "0.3333333333",
+    withDecimalContext(10) { (1.d / 3.d).toStringExpanded() }
+)
+```
+
+You can also pass an explicit context object:
+
+```lyng
+import lyng.decimal
+
+val ctx = DecimalContext(6, DecimalRounding.HalfAwayFromZero)
+assertEquals("0.666667", withDecimalContext(ctx) { ("2".d / 3.d).toStringExpanded() })
+```
+
+The context is dynamic and local to the block. After the block exits, the previous context is restored.
+
+## Rounding Modes
+
+Available rounding modes:
+
+- `HalfEven`
+- `HalfAwayFromZero`
+- `HalfTowardsZero`
+- `Ceiling`
+- `Floor`
+- `AwayFromZero`
+- `TowardsZero`
+
+Tie example at precision `2`:
+
+```lyng
+import lyng.decimal
+
+assertEquals("0.12", withDecimalContext(2, DecimalRounding.HalfEven) { (1.d / 8.d).toStringExpanded() })
+assertEquals("0.13", withDecimalContext(2, DecimalRounding.HalfAwayFromZero) { (1.d / 8.d).toStringExpanded() })
+assertEquals("0.12", withDecimalContext(2, DecimalRounding.HalfTowardsZero) { (1.d / 8.d).toStringExpanded() })
+assertEquals("0.13", withDecimalContext(2, DecimalRounding.Ceiling) { (1.d / 8.d).toStringExpanded() })
+assertEquals("0.12", withDecimalContext(2, DecimalRounding.Floor) { (1.d / 8.d).toStringExpanded() })
+```
+
+Negative values follow the same named policy in the obvious direction:
+
+```lyng
+import lyng.decimal
+
+assertEquals("-0.13", withDecimalContext(2, DecimalRounding.HalfAwayFromZero) { (-1.d / 8.d).toStringExpanded() })
+assertEquals("-0.12", withDecimalContext(2, DecimalRounding.HalfTowardsZero) { (-1.d / 8.d).toStringExpanded() })
+```
+
+## Recommended Usage Rules
+
+## Decimal With Stdlib Math Functions
+
+Core math helpers such as `abs`, `floor`, `ceil`, `round`, `sin`, `exp`, `ln`, `sqrt`, `log10`, `log2`, and `pow`
+now also accept `Decimal`.
+
+Current behavior is intentionally split:
+
+- exact decimal implementation:
+  - `abs(x)`
+  - `floor(x)`
+  - `ceil(x)`
+  - `round(x)`
+  - `pow(x, y)` when `x` is `Decimal` and `y` is an integral exponent
+- temporary bridge through `Real`:
+  - `sin`, `cos`, `tan`
+  - `asin`, `acos`, `atan`
+  - `sinh`, `cosh`, `tanh`
+  - `asinh`, `acosh`, `atanh`
+  - `exp`, `ln`, `log10`, `log2`
+  - `sqrt`
+  - `pow` for the remaining non-integral decimal exponent cases
+
+The temporary bridge is:
+
+```lyng
+Decimal -> Real -> host math -> Decimal
+```
+
+This is a compatibility step, not the long-term design. Native decimal implementations will replace these bridge-based
+paths over time.
+
+Examples:
+
+```lyng
+import lyng.decimal
+
+assertEquals("2.5", (abs("-2.5".d) as Decimal).toStringExpanded())
+assertEquals("2", (floor("2.9".d) as Decimal).toStringExpanded())
+
+// Temporary Real bridge:
+assertEquals((exp(1.25) as Real).d.toStringExpanded(), (exp("1.25".d) as Decimal).toStringExpanded())
+assertEquals((sqrt(2.0) as Real).d.toStringExpanded(), (sqrt("2".d) as Decimal).toStringExpanded())
+```
+
+If you care about exact decimal source text:
+
+```lyng
+"12.34".d
+```
+
+If you intentionally convert an existing binary floating-point value:
+
+```lyng
+someReal.d
+```
+
+If you want local control over division:
+
+```lyng
+withDecimalContext(precision, rounding) { ... }
+```
+
+If you want custom mixed operators for your own type, follow the same pattern as Decimal and use the operator registry:
+
+- define the operators on your own class
+- choose a common class
+- register mixed operand bridges
+
+See [OperatorInterop.md](OperatorInterop.md).

docs/ImmutableList.md

@@ -0,0 +1,37 @@
+# ImmutableList built-in class
+
+`ImmutableList` is an immutable, indexable list value.
+It implements [Array], therefore [Collection] and [Iterable].
+
+Use it when API contracts require a list that cannot be mutated through aliases.
+
+## Creating
+
+    val a = ImmutableList(1,2,3)
+    val b = [1,2,3].toImmutable()
+    val c = (1..3).toImmutableList()
+    >>> void
+
+## Converting
+
+    val i = ImmutableList(1,2,3)
+    val m = i.toMutable()
+    m += 4
+    assertEquals( ImmutableList(1,2,3), i )
+    assertEquals( [1,2,3,4], m )
+    >>> void
+
+## Members
+
+| name          | meaning                                 |
+|---------------|-----------------------------------------|
+| `size`        | number of elements                      |
+| `[index]`     | element access by index                 |
+| `[Range]`     | immutable slice                         |
+| `+`           | append element(s), returns new immutable list |
+| `-`           | remove element(s), returns new immutable list |
+| `toMutable()` | create mutable copy                     |
+
+[Array]: Array.md
+[Collection]: Collection.md
+[Iterable]: Iterable.md

docs/ImmutableMap.md

@@ -0,0 +1,36 @@
+# ImmutableMap built-in class
+
+`ImmutableMap` is an immutable map of key-value pairs.
+It implements [Collection] and [Iterable] of [MapEntry].
+
+## Creating
+
+    val a = ImmutableMap("a" => 1, "b" => 2)
+    val b = Map("a" => 1, "b" => 2).toImmutable()
+    val c = ["a" => 1, "b" => 2].toImmutableMap
+    >>> void
+
+## Converting
+
+    val i = ImmutableMap("a" => 1)
+    val m = i.toMutable()
+    m["a"] = 2
+    assertEquals( 1, i["a"] )
+    assertEquals( 2, m["a"] )
+    >>> void
+
+## Members
+
+| name            | meaning                                  |
+|-----------------|------------------------------------------|
+| `size`          | number of entries                        |
+| `[key]`         | get value by key, or `null` if absent    |
+| `getOrNull(key)`| same as `[key]`                          |
+| `keys`          | list of keys                             |
+| `values`        | list of values                           |
+| `+`             | merge (rightmost wins), returns new immutable map |
+| `toMutable()`   | create mutable copy                      |
+
+[Collection]: Collection.md
+[Iterable]: Iterable.md
+[MapEntry]: Map.md

docs/ImmutableSet.md

@@ -0,0 +1,34 @@
+# ImmutableSet built-in class
+
+`ImmutableSet` is an immutable set of unique elements.
+It implements [Collection] and [Iterable].
+
+## Creating
+
+    val a = ImmutableSet(1,2,3)
+    val b = Set(1,2,3).toImmutable()
+    val c = [1,2,3].toImmutableSet
+    >>> void
+
+## Converting
+
+    val i = ImmutableSet(1,2,3)
+    val m = i.toMutable()
+    m += 4
+    assertEquals( ImmutableSet(1,2,3), i )
+    assertEquals( Set(1,2,3,4), m )
+    >>> void
+
+## Members
+
+| name          | meaning                                             |
+|---------------|-----------------------------------------------------|
+| `size`        | number of elements                                  |
+| `contains(x)` | membership test                                     |
+| `+`, `union`  | union, returns new immutable set                    |
+| `-`, `subtract` | subtraction, returns new immutable set            |
+| `*`, `intersect` | intersection, returns new immutable set         |
+| `toMutable()` | create mutable copy                                 |
+
+[Collection]: Collection.md
+[Iterable]: Iterable.md

docs/Iterable.md

@@ -55,6 +55,23 @@ Here is the sample:
     assertEquals( (1..3).joinToString { it * 10 }, "10 20 30")
     >>> void
 
+## joinAll
+
+`joinAll()` is an `Iterable<Deferred>` helper that awaits every deferred in iteration order and returns a `List`
+with the collected results.
+
+    val jobs = (1..4).map { n ->
+        launch { n * n }
+    }
+    assertEquals([1, 4, 9, 16], jobs.joinAll())
+    >>> void
+
+Notes:
+
+- it does not start any task by itself; it only awaits the deferreds already present in the iterable.
+- awaiting happens in iteration order, so the result list keeps the same order as the input iterable.
+- if any deferred fails or was cancelled, that `await()` error is propagated from `joinAll()`.
+
 ## `sum` and `sumOf`
 
 These, again, does the thing:
@@ -108,8 +125,8 @@ You can also use flow variations that return a cold `Flow` instead of a `List`,
 Find the minimum or maximum value of a function applied to each element:
 
     val source = ["abc", "de", "fghi"]
-    assertEquals(2, source.minOf { it.length })
-    assertEquals(4, source.maxOf { it.length })
+    assertEquals(2, source.minOf { (it as String).length })
+    assertEquals(4, source.maxOf { (it as String).length })
     >>> void
 
 ## flatten and flatMap
@@ -147,12 +164,15 @@ Search for the first element that satisfies the given predicate:
 | fun/method             | description                                                                     |
 |------------------------|---------------------------------------------------------------------------------|
 | toList()               | create a list from iterable                                                     |
+| toImmutableList()      | create an immutable list from iterable                                          |
 | toSet()                | create a set from iterable                                                      |
+| toImmutableSet         | create an immutable set from iterable                                           |
 | contains(i)            | check that iterable contains `i`                                                |
 | `i in iterable`        | same as `contains(i)`                                                           |
 | isEmpty()              | check iterable is empty                                                         |
 | forEach(f)             | call f for each element                                                         |
 | toMap()                | create a map from list of key-value pairs (arrays of 2 items or like)           |
+| toImmutableMap         | create an immutable map from list of key-value pairs                             |
 | any(p)                 | true if any element matches predicate `p`                        |
 | all(p)                 | true if all elements match predicate `p`                         |
 | map(f)                 | create a list of values returned by `f` called for each element of the iterable |
@@ -181,6 +201,7 @@ Search for the first element that satisfies the given predicate:
 | sortedWith(comparator) | sort using a comparator that compares elements (1)                              |
 | sortedBy(predicate)    | sort by comparing results of the predicate function                             |
 | joinToString(s,t)      | convert iterable to string, see (2)                                             |
+| joinAll()              | for `Iterable<Deferred>`, await all items in order and collect results to [List] |
 | reversed()             | create a list containing items from this in reverse order                       |
 | shuffled()             | create a list of shuffled elements                                              |
 
@@ -206,16 +227,20 @@ For high-performance Kotlin-side interop and custom iterable implementation deta
 
 ## Implemented in classes:
 
-- [List], [Range], [Buffer](Buffer.md), [BitBuffer], [Buffer], [Set], [RingBuffer]
+- [List], [ImmutableList], [Range], [Buffer](Buffer.md), [BitBuffer], [Buffer], [Set], [ImmutableSet], [Map], [ImmutableMap], [RingBuffer]
 
 [Collection]: Collection.md
 
 [List]: List.md
+[ImmutableList]: ImmutableList.md
 
 [Flow]: parallelism.md#flow
 
 [Range]: Range.md
 
 [Set]: Set.md
+[ImmutableSet]: ImmutableSet.md
+[Map]: Map.md
+[ImmutableMap]: ImmutableMap.md
 
 [RingBuffer]: RingBuffer.md

docs/LaunchPool.md

@@ -0,0 +1,110 @@
+# LaunchPool
+
+`LaunchPool` is a bounded-concurrency task pool: you submit lambdas with `launch`, and the pool runs them using a fixed number of worker coroutines.
+
+## Constructor
+
+```
+LaunchPool(maxWorkers, maxQueueSize = Channel.UNLIMITED)
+```
+
+| Parameter | Description |
+|-----------|-------------|
+| `maxWorkers` | Maximum number of tasks that run in parallel. |
+| `maxQueueSize` | Maximum number of tasks that may wait in the queue. When the queue is full, `launch` suspends the caller until space becomes available. Defaults to `Channel.UNLIMITED` (no bound). |
+
+## Methods
+
+### `launch(lambda): Deferred`
+
+Schedules `lambda` for execution and returns a `Deferred` for its result.
+
+- Suspends if the queue is full (`maxQueueSize` reached).
+- Throws `IllegalStateException` if the pool is already closed or cancelled.
+- Any exception thrown by `lambda` is captured in the returned `Deferred` and **does not escape the pool**.
+
+```lyng
+val pool = LaunchPool(4)
+val d1 = pool.launch { computeSomething() }
+val d2 = pool.launch { computeOther() }
+pool.closeAndJoin()
+println(d1.await())
+println(d2.await())
+```
+
+### `closeAndJoin()`
+
+Stops accepting new tasks and suspends until all queued and running tasks complete normally. After this call, any further `launch` throws `IllegalStateException`. Idempotent — safe to call multiple times.
+
+### `cancel()`
+
+Immediately closes the queue and cancels all worker coroutines. Queued but unstarted tasks are discarded. After this call, `launch` throws `IllegalStateException`. Idempotent.
+
+### `cancelAndJoin()`
+
+Like `cancel()`, but also suspends until all worker coroutines have stopped. Useful when you need to be sure no worker code is still running before proceeding. Idempotent.
+
+## Exception handling
+
+Exceptions from submitted lambdas are captured per-task in the returned `Deferred`. The pool itself continues running after a task failure:
+
+```lyng
+val pool = LaunchPool(2)
+val good = pool.launch { 42 }
+val bad  = pool.launch { throw IllegalArgumentException("boom") }
+pool.closeAndJoin()
+
+assertEquals(42, good.await())
+assertThrows(IllegalArgumentException) { bad.await() }
+```
+
+## Bounded queue / back-pressure
+
+When `maxQueueSize` is set, the producer suspends if the queue fills up, providing automatic back-pressure:
+
+```lyng
+// 1 worker, queue of 2 — producer can be at most 2 tasks ahead of what's running
+val pool = LaunchPool(1, 2)
+val d1 = pool.launch { delay(10); "a" }
+val d2 = pool.launch { delay(10); "b" }
+val d3 = pool.launch { delay(10); "c" }   // suspends until d1 is picked up by the worker
+pool.closeAndJoin()
+```
+
+## Collecting all results
+
+`launch` returns a `Deferred`, so you can collect results with `joinAll()`:
+
+```lyng
+val pool = LaunchPool(4)
+val jobs = (1..10).map { n -> pool.launch { n * n } }
+pool.closeAndJoin()
+val results = jobs.joinAll()
+// results == [1, 4, 9, 16, 25, 36, 49, 64, 81, 100]
+```
+
+## Concurrency limit in practice
+
+With `maxWorkers = 2`, at most 2 tasks run simultaneously regardless of how many are queued:
+
+```lyng
+val mu = Mutex()
+var active = 0
+var maxSeen = 0
+
+val pool = LaunchPool(2)
+(1..8).map {
+    pool.launch {
+        mu.withLock { active++; if (active > maxSeen) maxSeen = active }
+        delay(5)
+        mu.withLock { active-- }
+    }
+}
+pool.closeAndJoin()
+assert(maxSeen <= 2)
+```
+
+## See also
+
+- [parallelism.md](parallelism.md) — `launch`, `Deferred`, `Mutex`, `Channel`, and coroutine basics
+- [Channel.md](Channel.md) — the underlying channel primitive used by `LaunchPool`

docs/LegacyDigest.md

@@ -0,0 +1,66 @@
+# Legacy Digest Functions (`lyng.legacy_digest`)
+
+> ⚠️ **Security warning:** The functions in this module use cryptographically broken
+> algorithms.  Do **not** use them for passwords, digital signatures, integrity
+> verification against adversarial tampering, or any other security-sensitive
+> purpose.  They exist solely for compatibility with legacy protocols and file
+> formats that require specific hash values.
+
+Import when you need to produce a SHA-1 digest for an existing protocol or format:
+
+```lyng
+import lyng.legacy_digest
+```
+
+## `LegacyDigest` Object
+
+### `sha1(data): String`
+
+Computes the SHA-1 digest of `data` and returns it as a 40-character lowercase
+hex string.
+
+`data` can be:
+
+| Type     | Behaviour                              |
+|----------|----------------------------------------|
+| `String` | Encoded as UTF-8, then hashed          |
+| `Buffer` | Raw bytes hashed directly              |
+| anything | Falls back to `toString()` then UTF-8  |
+
+```lyng
+import lyng.legacy_digest
+
+// String input
+val h = LegacyDigest.sha1("abc")
+assertEquals("a9993e364706816aba3e25717850c26c9cd0d89d", h)
+
+// Empty string
+assertEquals("da39a3ee5e6b4b0d3255bfef95601890afd80709", LegacyDigest.sha1(""))
+```
+
+```lyng
+import lyng.legacy_digest
+import lyng.buffer
+
+// Buffer input (raw bytes)
+val buf = Buffer.decodeHex("616263")   // 0x61 0x62 0x63 = "abc"
+assertEquals("a9993e364706816aba3e25717850c26c9cd0d89d", LegacyDigest.sha1(buf))
+```
+
+## Implementation Notes
+
+- Pure Kotlin/KMP — no native libraries or extra dependencies.
+- Follows FIPS 180-4.
+- The output is always lowercase hex, never uppercase or binary.
+
+## When to Use
+
+Use `lyng.legacy_digest` only when an external system you cannot change requires
+a SHA-1 value, for example:
+
+- old git-style content addresses
+- some OAuth 1.0 / HMAC-SHA1 signature schemes
+- legacy file checksums defined in published specs
+
+For any new design choose a current hash function (SHA-256 or better) once
+Lyng adds a `lyng.digest` module.

docs/List.md

@@ -1,6 +1,8 @@
 # List built-in class
 
 Mutable list of any objects.
+For immutable list values, see [ImmutableList].
+For observable mutable lists and change hooks, see [ObservableList].
 
 It's class in Lyng is `List`:
 
@@ -28,6 +30,13 @@ There is a shortcut for the last:
 
 __Important__ negative indexes works wherever indexes are used, e.g. in insertion and removal methods too.
 
+The language also allows multi-selector indexing syntax such as `value[i, j]`, but `List` itself uses a single selector only:
+
+- `list[index]` for one element
+- `list[range]` for a slice copy
+
+Multi-selector indexing is intended for custom indexers such as `Matrix`.
+
 ## Concatenation
 
 You can concatenate lists or iterable objects:
@@ -36,6 +45,16 @@ You can concatenate lists or iterable objects:
     assert( [4,5] + (1..3) == [4, 5, 1, 2, 3])
     >>> void
 
+## Constructing lists
+
+Besides literals, you can build a list by size using `List.fill`:
+
+    val squares = List.fill(5) { i -> i * i }
+    assertEquals([0, 1, 4, 9, 16], squares)
+    >>> void
+
+`List.fill(size) { ... }` calls the block once for each index from `0` to `size - 1` and returns a new mutable list.
+
 ## Appending
 
 To append to lists, use `+=` with elements, lists and any [Iterable] instances, but beware it will
@@ -155,6 +174,9 @@ List could be sorted in place, just like [Collection] provide sorted copies, in
 | `[index]`                     | get or set element at index                  | Int         |
 | `[Range]`                     | get slice of the array (copy)                | Range       |
 | `+=`                          | append element(s) (2)                        | List or Obj |
+| `List.fill(size, block)`      | build a new list from indices `0..<size`     | Int, Callable |
+| `List.fill(size,capacity,block)` | same, pre-allocating capacity slots       | Int, Int, Callable |
+| `ensureCapacity(count)`       | pre-allocate storage for at least `count` elements without reallocation (5) | Int |
 | `sort()`                      | in-place sort, natural order                 | void        |
 | `sortBy(predicate)`           | in-place sort bu `predicate` call result (3) | void        |
 | `sortWith(comparator)`        | in-place sort using `comarator` function (4) | void        |
@@ -177,8 +199,57 @@ order, e.g. is same as `list.sortWith { a,b -> predicate(a) <=> predicate(b) }`
 positive if first is greater, and zero if they are equal. For example, the equvalent comparator
 for `sort()` will be `sort { a, b -> a <=> b }
 
+(5)
+: if the current capacity is already ≥ `count`, this is a no-op. Otherwise the internal storage
+is reallocated to hold at least `count` elements. Use this before a bulk `+=` loop to avoid
+repeated reallocations. `List.fill(size, capacity, block)` calls this automatically.
+
 It inherits from [Iterable] too and thus all iterable methods are applicable to any list.
 
+## Observable list hooks
+
+Observable hooks are provided by module `lyng.observable` and are opt-in:
+
+    import lyng.observable
+
+    val src = [1,2,3]
+    val xs = src.observable()
+    assert(xs is ObservableList<Int>)
+
+    var before = 0
+    var after = 0
+    xs.beforeChange { before++ }
+    xs.onChange { after++ }
+
+    xs += 4
+    xs[0] = 100
+    assertEquals([100,2,3,4], xs)
+    assertEquals(2, before)
+    assertEquals(2, after)
+    >>> void
+
+`beforeChange` runs before mutation commit and may reject it by throwing exception (typically `ChangeRejectionException` from the same module):
+
+    import lyng.observable
+    val xs = [1,2].observable()
+    xs.beforeChange { throw ChangeRejectionException("read only") }
+    assertThrows(ChangeRejectionException) { xs += 3 }
+    assertEquals([1,2], xs)
+    >>> void
+
+`changes()` returns `Flow<ListChange<T>>` of committed events:
+
+    import lyng.observable
+    val xs = [10,20].observable()
+    val it = xs.changes().iterator()
+    xs += 30
+    assert(it.hasNext())
+    val e = it.next()
+    assert(e is ListInsert<Int>)
+    assertEquals([30], (e as ListInsert<Int>).values)
+    it.cancelIteration()
+    >>> void
+
 ## Member inherited from Array
 
 | name             | meaning                        | type  |
@@ -197,3 +268,5 @@ It inherits from [Iterable] too and thus all iterable methods are applicable to
 [Range]: Range.md
 
 [Iterable]: Iterable.md
+[ImmutableList]: ImmutableList.md
+[ObservableList]: ObservableList.md

docs/Map.md

@@ -3,6 +3,7 @@
 Map is a mutable collection of key-value pairs, where keys are unique. You can create maps in two ways:
 - with the constructor `Map(...)` or `.toMap()` helpers; and
 - with map literals using braces: `{ "key": value, id: expr, id: }`.
+For immutable map values, see [ImmutableMap].
 
 When constructing from a list, each list item must be a [Collection] with exactly 2 elements, for example, a [List].
 
@@ -94,7 +95,8 @@ Or iterate its key-value pairs that are instances of [MapEntry] class:
 
     val map = Map( ["foo", 1], ["bar", "buzz"], [42, "answer"] )
     for( entry in map ) {
-        println("map[%s] = %s"(entry.key, entry.value))
+        val e: MapEntry = entry as MapEntry
+        println("map[%s] = %s"(e.key, e.value))
     }
     void
     >>> map[foo] = 1
@@ -176,3 +178,4 @@ Notes:
 - When you need computed or non-string keys, use the constructor form `Map(...)`, map literals with computed keys (if supported), or build entries with `=>` and then merge.
 
 [Collection](Collection.md)
+[ImmutableMap]: ImmutableMap.md

docs/Matrix.md

@@ -0,0 +1,192 @@
+# Matrix (`lyng.matrix`)
+
+`lyng.matrix` adds dense immutable `Matrix` and `Vector` types for linear algebra.
+
+Import it when you need matrix or vector arithmetic:
+
+```lyng
+import lyng.matrix
+```
+
+## Construction
+
+Create vectors from a flat list and matrices from nested row lists:
+
+```lyng
+import lyng.matrix
+
+val v: Vector = vector([1, 2, 3])
+val m: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+
+assertEquals([1.0, 2.0, 3.0], v.toList())
+assertEquals([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]], m.toList())
+```
+
+Factory methods are also available:
+
+```lyng
+import lyng.matrix
+
+val z: Vector = Vector.zeros(3)
+val i: Matrix = Matrix.identity(3)
+val m: Matrix = Matrix.zeros(2, 4)
+```
+
+All elements are standard double-precision numeric values internally.
+
+## Shapes
+
+Matrices may have any rectangular geometry:
+
+```lyng
+import lyng.matrix
+
+val a: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+
+assertEquals(2, a.rows)
+assertEquals(3, a.cols)
+assertEquals([2, 3], a.shape)
+assertEquals(false, a.isSquare)
+```
+
+Vectors expose:
+
+- `size`
+- `length` as an alias of `size`
+
+## Matrix Operations
+
+Supported matrix operations:
+
+- `+` and `-` for element-wise matrix arithmetic
+- `*` for matrix-matrix product
+- `*` and `/` by a scalar
+- `transpose()`
+- `trace()`
+- `rank()`
+- `determinant()`
+- `inverse()`
+- `solve(rhs)` for `Vector` or `Matrix` right-hand sides
+
+Example:
+
+```lyng
+import lyng.matrix
+
+val a: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+val b: Matrix = matrix([[7, 8], [9, 10], [11, 12]])
+val product: Matrix = a * b
+assertEquals([[58.0, 64.0], [139.0, 154.0]], product.toList())
+assertEquals([[1.0, 4.0], [2.0, 5.0], [3.0, 6.0]], a.transpose().toList())
+```
+
+Inverse and solve:
+
+```lyng
+import lyng.matrix
+
+val a: Matrix = matrix([[4, 7], [2, 6]])
+val rhs: Vector = vector([1, 0])
+
+val inv: Matrix = a.inverse()
+val x: Vector = a.solve(rhs)
+
+assert(abs(a.determinant() - 10.0) < 1e-9)
+assert(abs(inv.get(0, 0) - 0.6) < 1e-9)
+assert(abs(x.get(0) - 0.6) < 1e-9)
+```
+
+## Vector Operations
+
+Supported vector operations:
+
+- `+` and `-`
+- scalar `*` and `/`
+- `dot(other)`
+- `norm()`
+- `normalize()`
+- `cross(other)` for 3D vectors
+- `outer(other)` producing a matrix
+
+```lyng
+import lyng.matrix
+
+val a: Vector = vector([1, 2, 3])
+val b: Vector = vector([2, 0, 0])
+
+assertEquals(2.0, a.dot(b))
+assertEquals([0.2672612419124244, 0.5345224838248488, 0.8017837257372732], a.normalize().toList())
+```
+
+## Indexing and Slicing
+
+`Matrix` supports both method-style indexing and bracket syntax.
+
+Scalar access:
+
+```lyng
+import lyng.matrix
+
+val m: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+
+assertEquals(6.0, m.get(1, 2))
+assertEquals(6.0, m[1, 2])
+```
+
+Bracket indexing accepts two selectors: `[row, col]`.
+Each selector may be either:
+
+- an `Int`
+- a `Range`
+
+Examples:
+
+```lyng
+import lyng.matrix
+
+val m: Matrix = matrix([[1, 2, 3, 4], [5, 6, 7, 8], [9, 10, 11, 12]])
+
+assertEquals(7.0, m[1, 2])
+val columnSlice: Matrix = m[0..2, 2]
+val topLeft: Matrix = m[0..1, 0..1]
+val tail: Matrix = m[1.., 1..]
+assertEquals([[3.0], [7.0], [11.0]], columnSlice.toList())
+assertEquals([[1.0, 2.0], [5.0, 6.0]], topLeft.toList())
+assertEquals([[6.0, 7.0, 8.0], [10.0, 11.0, 12.0]], tail.toList())
+```
+
+Shape rules:
+
+- `m[Int, Int]` returns a `Real`
+- `m[Range, Int]` returns an `Nx1` `Matrix`
+- `m[Int, Range]` returns a `1xM` `Matrix`
+- `m[Range, Range]` returns a submatrix
+
+Open-ended ranges are supported:
+
+- `m[..1, ..1]`
+- `m[1.., 1..]`
+- `m[.., 2]`
+
+Stepped ranges are not supported in matrix slicing.
+
+Slices currently return new matrices, not views.
+
+## Rows and Columns
+
+If you want plain lists instead of a sliced matrix:
+
+```lyng
+import lyng.matrix
+
+val a: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+
+assertEquals([4.0, 5.0, 6.0], a.row(1))
+assertEquals([2.0, 5.0], a.column(1))
+```
+
+## Backend Notes
+
+The matrix module uses a platform-specific backend where available and falls back to pure Kotlin where needed.
+
+The public Lyng API stays the same across platforms.

docs/OOP.md

@@ -9,7 +9,7 @@ Lyng supports first class OOP constructs, based on classes with multiple inherit
 The class clause looks like
 
     class Point(x,y)
-    assert( Point is Class ) 
+    assertEquals("Point", Point.className)
     >>> void
 
 It creates new `Class` with two fields. Here is the more practical sample:
@@ -71,14 +71,99 @@ object DefaultLogger : Logger("Default") {
 }
 ```
 
+## Object Expressions
+
+Object expressions allow you to create an instance of an anonymous class. This is useful when you need to provide a one-off implementation of an interface or inherit from a class without declaring a new named subclass.
+
+```lyng
+val worker = object : Runnable {
+    override fun run() {
+        println("Working...")
+    }
+}
+```
+
+Object expressions can implement multiple interfaces and inherit from one class:
+
+```lyng
+val x = object : Base(arg1), Interface1, Interface2 {
+    val property = 42
+    override fun method() = property * 2
+}
+```
+
+### Scoping and `this@object`
+
+Object expressions capture their lexical scope, meaning they can access local variables and members of the outer class. When `this` is rebound (for example, inside an `apply` block), you can use the reserved alias `this@object` to refer to the innermost anonymous object instance.
+
+```lyng
+val handler = object {
+    fun process() {
+        this@object.apply {
+            // here 'this' is rebound to the map/context
+            // but we can still access the anonymous object via this@object
+            println("Processing in " + this@object)
+        }
+    }
+}
+```
+
+### Serialization and Identity
+
+- **Serialization**: Anonymous objects are **not serializable**. Attempting to encode an anonymous object via `Lynon` will throw a `SerializationException`. This is because their class definition is transient and cannot be safely restored in a different session or process.
+- **Type Identity**: Every object expression creates a unique anonymous class. Two identical object expressions will result in two different classes with distinct type identities.
+
+## Nested Declarations
+
+Lyng allows classes, objects, enums, and type aliases to be declared inside another class. These declarations live in the **class namespace** (not the instance), so they do not capture an outer instance and are accessed with a qualifier.
+
+```lyng
+class A {
+    class B(x?)
+    object Inner { val foo = "bar" }
+    type Alias = B
+    enum E { One, Two }
+}
+
+val ab = A.B()
+assertEquals(ab.x, null)
+assertEquals(A.Inner.foo, "bar")
+```
+
+Rules:
+- **Qualified access**: use `Outer.Inner` for nested classes/objects/enums/aliases. Inside `Outer` you can refer to them by unqualified name unless shadowed.
+- **No inner semantics**: nested declarations do not capture an instance of the outer class. They are resolved at compile time.
+- **Visibility**: `private` restricts a nested declaration to the declaring class body (not visible from outside or subclasses).
+- **Reflection name**: a nested class reports `Outer.Inner` (e.g., `A.B::class.name` is `"A.B"`).
+- **Type aliases**: behave as aliases of the qualified nested type and are expanded by the type system.
+
+### Lifted Enum Entries
+
+Enums can optionally lift their entries into the surrounding class namespace using `*`:
+
+```lyng
+class A {
+    enum E* { One, Two }
+}
+
+assertEquals(A.One, A.E.One)
+assertEquals(A.Two, A.E.Two)
+```
+
+Notes:
+- `E*` exposes entries in `A` as if they were direct members (`A.One`).
+- If a name would conflict with an existing class member, compilation fails (no implicit fallback).
+- Without `*`, use the normal `A.E.One` form.
+
 ## Properties
 
 Properties allow you to define member accessors that look like fields but execute code when read or written. Unlike regular fields, properties in Lyng do **not** have automatic backing fields; they are pure accessors.
 
 ### Basic Syntax
 
-Properties are declared using `val` (read-only) or `var` (read-write) followed by a name and `get()`/`set()` blocks:
+Properties are declared using `val` (read-only) or `var` (read-write) followed by a name and a `get` (and optionally `set`) accessor. Unlike fields, properties do not have automatic storage and must compute their values or delegate to other members.
 
+The standard syntax uses parentheses:
 ```lyng
 class Person(private var _age: Int) {
     // Read-only property
@@ -94,21 +179,17 @@ class Person(private var _age: Int) {
             if (value >= 0) _age = value
         }
 }
-
-val p = Person(15)
-assertEquals("Minor", p.ageCategory)
-p.age = 20
-assertEquals("Adult", p.ageCategory)
 ```
 
-### Laconic Expression Shorthand
+### Laconic Syntax (Optional Parentheses)
 
-For simple accessors and methods, you can use the `=` shorthand for a more elegant and laconic form:
+For even cleaner code, you can omit the parentheses for `get` and `set`. This is especially useful for simple expression shorthand:
 
 ```lyng
 class Circle(val radius: Real) {
-    val area get() = π * radius * radius
-    val circumference get() = 2 * π * radius
+    // Laconic expression shorthand
+    val area get = π * radius * radius
+    val circumference get = 2 * π * radius
     
     fun diameter() = radius * 2
 }
@@ -117,15 +198,16 @@ fun median(a, b) = (a + b) / 2
 
 class Counter {
     private var _count = 0
-    var count get() = _count set(v) = _count = v
+    var count get = _count set(v) = _count = v
 }
 ```
 
 ### Key Rules
 
-- **`val` properties** must have a `get()` accessor and cannot have a `set()`.
-- **`var` properties** must have both `get()` and `set()` accessors.
+- **`val` properties** must have a `get` accessor (with or without parentheses) and cannot have a `set`.
+- **`var` properties** must have both `get` and `set` accessors.
 - **Functions and methods** can use the `=` shorthand to return the result of a single expression.
+- **`override` is mandatory**: If you are overriding a member from a base class, you MUST use the `override` keyword.
 - **No Backing Fields**: There is no magic `field` identifier. If you need to store state, you must declare a separate (usually `private`) field.
 - **Type Inference**: You can omit the type declaration if it can be inferred or if you don't need strict typing.
 
@@ -182,6 +264,19 @@ A delegate is any object that provides the following methods (all optional depen
 - `invoke(thisRef, name, args...)`: Called when a delegated `fun` is invoked.
 - `bind(name, access, thisRef)`: Called once during initialization to configure or validate the delegate.
 
+### Map as a Delegate
+
+Maps can also be used as delegates. When delegated to a property, the map uses the property name as the key:
+
+```lyng
+val settings = { "theme": "dark", "fontSize": 14 }
+val theme by settings
+var fontSize by settings
+
+println(theme)    // "dark"
+fontSize = 16     // Updates settings["fontSize"]
+```
+
 For more details and advanced patterns (like `lazy`, `observable`, and shared stateless delegates), see the [Delegation Guide](delegation.md).
 
 ## Instance initialization: init block
@@ -281,11 +376,10 @@ Functions defined inside a class body are methods, and unless declared
 `private` are available to be called from outside the class:
 
     class Point(x,y) {
-        // public method declaration:
-        fun length() { sqrt(d2()) }
-
         // private method:
         private fun d2() { x*x + y*y }
+        // public method declaration:
+        fun length() { sqrt(d2()) }
     }
     val p = Point(3,4)
     // private called from inside public: OK
@@ -372,7 +466,7 @@ Key rules and features:
 
 - Visibility
   - `private`: accessible only inside the declaring class body; not visible in subclasses and cannot be accessed via `this@Type` or casts.
-  - `protected`: accessible in the declaring class and in any of its transitive subclasses (including MI), but not from unrelated contexts; qualification/casts do not bypass it.
+  - `protected`: accessible in the declaring class and in any of its transitive subclasses (including MI). Additionally, ancestor classes can access protected members of their descendants if it's an override of a member known to the ancestor. Protected members are not visible from unrelated contexts; qualification/casts do not bypass it.
 
 ## Abstract Classes and Members
 
@@ -489,6 +583,99 @@ class Critical {
 
 Attempting to override a `closed` member results in a compile-time error.
 
+## Operator Overloading
+
+Lyng allows you to overload standard operators by defining specific named methods in your classes. When an operator expression is evaluated, Lyng delegates the operation to these methods if they are available.
+
+### Binary Operators
+
+To overload a binary operator, define the corresponding method that takes one argument:
+
+| Operator | Method Name |
+| :--- | :--- |
+| `a + b` | `plus(other)` |
+| `a - b` | `minus(other)` |
+| `a * b` | `mul(other)` |
+| `a / b` | `div(other)` |
+| `a % b` | `mod(other)` |
+| `a && b` | `logicalAnd(other)` |
+| `a \|\| b` | `logicalOr(other)` |
+| `a =~ b` | `operatorMatch(other)` |
+| `a & b` | `bitAnd(other)` |
+| `a \| b` | `bitOr(other)` |
+| `a ^ b` | `bitXor(other)` |
+| `a << b` | `shl(other)` |
+| `a >> b` | `shr(other)` |
+
+Example:
+```lyng
+class Vector(val x, val y) {
+    fun plus(other) = Vector(x + other.x, y + other.y)
+    override fun toString() = "Vector(${x}, ${y})"
+}
+
+val v1 = Vector(1, 2)
+val v2 = Vector(3, 4)
+assertEquals(Vector(4, 6), v1 + v2)
+```
+
+### Unary Operators
+
+Unary operators are overloaded by defining methods with no arguments:
+
+| Operator | Method Name |
+| :--- | :--- |
+| `-a` | `negate()` |
+| `!a` | `logicalNot()` |
+| `~a` | `bitNot()` |
+
+### Assignment Operators
+
+Assignment operators like `+=` first attempt to call a specific assignment method. If that method is not defined, they fall back to a combination of the binary operator and a regular assignment (e.g., `a = a + b`).
+
+| Operator | Method Name | Fallback |
+| :--- | :--- | :--- |
+| `a += b` | `plusAssign(other)` | `a = a + b` |
+| `a -= b` | `minusAssign(other)` | `a = a - b` |
+| `a *= b` | `mulAssign(other)` | `a = a * b` |
+| `a /= b` | `divAssign(other)` | `a = a / b` |
+| `a %= b` | `modAssign(other)` | `a = a % b` |
+
+Example of in-place mutation:
+```lyng
+class Counter(var value) {
+    fun plusAssign(n) {
+        value = value + n
+    }
+}
+
+val c = Counter(10)
+c += 5
+assertEquals(15, c.value)
+```
+
+### Comparison Operators
+
+Comparison operators use `compareTo` and `equals`.
+
+| Operator | Method Name |
+| :--- | :--- |
+| `a == b`, `a != b` | `equals(other)` |
+| `<`, `>`, `<=`, `>=`, `<=>` | `compareTo(other)` |
+
+- `compareTo` should return:
+  - `0` if `a == b`
+  - A negative integer if `a < b`
+  - A positive integer if `a > b`
+- The `<=>` (shuttle) operator returns the result of `compareTo` directly.
+- `equals` returns a `Bool`. If `equals` is not explicitly defined, Lyng falls back to `compareTo(other) == 0`.
+
+> **Note**: Methods that are already defined in the base `Obj` class (like `equals`, `toString`, or `contains`) require the `override` keyword when redefined in your class or as an extension. Other operator methods (like `plus` or `negate`) do not require `override` unless they are already present in your class's hierarchy.
+
+### Increment and Decrement
+
+`++` and `--` operators are implemented using `plus(1)` or `minus(1)` combined with an assignment back to the variable. If the variable is a field or local variable, it will be updated with the result of the operation.
+
 Compatibility notes:
 
 - Existing single‑inheritance code continues to work unchanged; its resolution order reduces to the single base.
@@ -572,6 +759,23 @@ Notes and limitations (current version):
 - `name` and `ordinal` are read‑only properties of an entry.
 - `entries` is a read‑only list owned by the enum type.
 
+## Exception Classes
+
+You can define your own exception classes by inheriting from the built-in `Exception` class. User-defined exceptions are regular classes and can have their own properties and methods.
+
+```lyng
+class MyError(val code, m) : Exception(m)
+
+try {
+    throw MyError(500, "Internal Server Error")
+}
+catch(e: MyError) {
+    println("Error " + e.code + ": " + e.message)
+}
+```
+
+For more details on error handling, see the [Exceptions Handling Guide](exceptions_handling.md).
+
 ## fields and visibility
 
 It is possible to add non-constructor fields:
@@ -694,23 +898,44 @@ Private fields are visible only _inside the class instance_:
     void
     >>> void
 
+### Transient fields
+
+You can mark a field or a constructor parameter as transient using the `@Transient` attribute. Transient members are ignored during serialization (Lynon and JSON) and are also excluded from structural equality (`==`) checks.
+
+```lyng
+class Session(@Transient val token, val userId) {
+    @Transient var lastAccess = time.now()
+    var data = Map()
+}
+```
+
+For more details on how transient fields behave during restoration, see the [Serialization Guide](serialization.md).
+
 ### Protected members
 
-Protected members are available to the declaring class and all of its transitive subclasses (including via MI), but not from unrelated contexts:
+Protected members are available to the declaring class and all of its transitive subclasses (including via MI). Additionally, an ancestor class can access a `protected` member of its descendant if the ancestor also defines or inherits a member with the same name (i.e., it is an override of something the ancestor knows about). 
 
-```
-class A() {
-    protected fun ping() { "pong" }
+Protected members are not available from unrelated contexts:
+
+```lyng
+class Base {
+    abstract protected fun foo()
+    
+    fun bar() {
+        // Ancestor can see foo() because it's an override 
+        // of a member it defines (even as abstract):
+        foo()
     }
-class B() : A() {
-    fun call() { this@A.ping() }
 }
 
-val b = B()
-assertEquals("pong", b.call())
+class Derived : Base {
+    override protected fun foo() { "ok" }
+}
+
+assertEquals("ok", Derived().bar())
 
 // Unrelated access is forbidden, even via cast
-assertThrows { (b as A).ping() }
+assertThrows { (Derived() as Base).foo() }
 ```
 
 It is possible to provide private constructor parameters so they can be
@@ -753,7 +978,7 @@ You can mark a field or a method as static. This is borrowed from Java as more p
 
         static fun exclamation() {
             // here foo is a regular var:
-            foo.x + "!"
+        Value.foo.x + "!"
         }
     }
     assertEquals( Value.foo.x, "foo" )
@@ -764,49 +989,29 @@ You can mark a field or a method as static. This is borrowed from Java as more p
     assertEquals( "bar!", Value.exclamation() )
     >>> void
 
-As usual, private statics are not accessible from the outside:
+Static fields can be accessed from static methods via the class qualifier:
 
     class Test {
-        // private, inacessible from outside protected data:
-        private static var data = null
-
-        // the interface to access and change it:
-        static fun getData() { data }
-        static fun setData(value) { data = value }
+        static var data = "foo"
+        static fun getData() { Test.data }
     }
 
-    // no direct access:
-    assertThrows { Test.data }
-
-    // accessible with the interface:
-    assertEquals( null, Test.getData() )
-    Test.setData("fubar")
-    assertEquals("fubar", Test.getData() )
+    assertEquals( "foo", Test.getData() )
+    Test.data = "bar"
+    assertEquals("bar", Test.getData() )
     >>> void
 
-# Extending classes
+# Extension members
 
-It sometimes happen that the class is missing some particular functionality that can be _added to it_ without rewriting its inner logic and using its private state. In this case _extension members_ could be used.
+Sometimes an existing type or named singleton object is missing some particular functionality that can be _added to it_ without rewriting its inner logic and without using its private state. In this case, _extension members_ can be used.
 
 ## Extension methods
 
-For example, we want to create an extension method that would test if some object of unknown type contains something that can be interpreted as an integer. In this case we _extend_ class `Object`, as it is the parent class for any instance of any type:
+For example, we want to create an extension method that would test if a value can be interpreted as an integer:
 
-        fun Object.isInteger() {
-            when(this) {
-                // already Int?
-                is Int -> true
-
-                // real, but with no declimal part?
-                is Real -> toInt() == this
-
-                // string with int or real reuusig code above
-                is String -> toReal().isInteger()
-                
-                // otherwise, no:
-                else -> false
-            }
-        }
+        fun Int.isInteger() { true }
+        fun Real.isInteger() { this.toInt() == this }
+        fun String.isInteger() { (this.toReal() as Real).isInteger() }
 
         // Let's test:
         assert( 12.isInteger() == true )
@@ -815,6 +1020,133 @@ For example, we want to create an extension method that would test if some objec
         assert( ! "5.2".isInteger() )
         >>> void
 
+Extension methods normally act like instance members. If declared as `static`, they are called on the type object itself:
+
+```lyng
+static fun List<T>.fill(size: Int, block: (Int)->T): List<T> { ... }
+
+val tens = List.fill(5) { it * 10 }
+```
+
+## Extending singleton `object` declarations
+
+Named singleton objects can also be extension receivers. Use the object name as the receiver type:
+
+```lyng
+object Config {
+    fun base() = "cfg"
+}
+
+fun Config.describe(value) {
+    this.base() + ":" + value.toString()
+}
+
+val Config.tag get() = this.base() + ":tag"
+
+assertEquals("cfg:42", Config.describe(42))
+assertEquals("cfg:tag", Config.tag)
+```
+
+This differs from extending a class in one important way:
+
+- `fun Point.foo()` adds a member-like extension for all `Point` instances.
+- `fun Config.foo()` adds a member-like extension only for the single named object `Config`.
+
+The same rules still apply:
+
+- Extensions on singleton objects are scope-isolated just like class extensions.
+- They cannot access the object's `private` members.
+- Inside the extension body, `this` is the singleton object itself.
+
+## Extension indexers
+
+Bracket syntax is just another form of member dispatch. When you write `value[i]` or `value[i] = x`, Lyng lowers it to `getAt(...)` and `putAt(...)`.
+
+That means indexers can be extended in the same way as methods and properties.
+
+### Extending indexers on classes
+
+Use `override fun Type.getAt(...)` and `override fun Type.putAt(...)`:
+
+```lyng
+class BoxStore {
+    val data = {"name": "alice"}
+}
+
+override fun BoxStore.getAt(key: String): Object? {
+    data[key]
+}
+
+override fun BoxStore.putAt(key: String, value: Object) {
+    data[key] = value
+}
+
+val store = BoxStore()
+assertEquals("alice", store["name"])
+store["name"] = "bob"
+assertEquals("bob", store["name"])
+```
+
+As with other extension members, this does not modify the original class declaration. It adds indexer behavior only in the scope where the extension is visible.
+
+### Extending indexers on singleton `object` declarations
+
+Named singleton objects work the same way:
+
+```lyng
+object Storage
+
+var storageData = {}
+
+override fun Storage.getAt(key: String): Object? {
+    storageData[key]
+}
+
+override fun Storage.putAt(key: String, value: Object) {
+    storageData[key] = value
+}
+
+Storage["name"] = "alice"
+val name: String? = Storage["name"]
+assertEquals("alice", name)
+```
+
+This is the indexer equivalent of `fun Config.foo()`: the extension applies to that single named object, not to all instances of some class.
+
+### Selector packing
+
+Index syntax can contain more than one selector:
+
+```lyng
+value[i]
+value[i, j]
+```
+
+The same packing rules still apply for extension indexers:
+
+- `value[i]` calls `getAt(i)` or `putAt(i, value)`
+- `value[i, j]` passes `[i, j]` as one list-like index object
+- `value[i, j, k]` passes `[i, j, k]`
+
+So if you want multi-selector indexing, define the receiver to accept that packed index object.
+
+### About types and generics
+
+In practice, extension indexers are usually best declared with `Object?` for reads and `Object` for writes:
+
+```lyng
+override fun Storage.getAt(key: String): Object? { ... }
+override fun Storage.putAt(key: String, value: Object) { ... }
+```
+
+Then use the expected type at the call site:
+
+```lyng
+val name: String? = Storage["name"]
+```
+
+Explicit generic arguments do not fit naturally onto `[]` syntax, so typed assignment on read is usually the clearest approach.
+
 ## Extension properties
 
 Just like methods, you can extend existing classes with properties. These can be defined using simple initialization (for `val` only) or with custom accessors.
@@ -910,7 +1242,7 @@ The same we can provide writable dynamic fields (var-type), adding set method:
                 // mutable field
                 "bar" -> storedValueForBar 
 
-                else -> throw SymbolNotFoundException()
+                else -> throw SymbolNotFound()
             }
         }
         set { name, value ->
@@ -978,8 +1310,24 @@ collection's sugar won't work with it:
     assertEquals("buzz", x[0])
     >>> void
 
-If you want dynamic to function like an array, create a [feature
-request](https://gitea.sergeych.net/SergeychWorks/lyng/issues).
+Multiple selectors are packed into one list index object:
+
+    val x = dynamic {
+        get {
+            if( it == [1, 2] ) "hit"
+            else null
+        }
+    }
+    assertEquals("hit", x[1, 2])
+    >>> void
+
+So:
+
+- `x[i]` passes `i`
+- `x[i, j]` passes `[i, j]`
+- `x[i, j, k]` passes `[i, j, k]`
+
+This is the same rule used by Kotlin-backed `getAt` / `putAt` indexers in embedding.
 
 # Theory
 

docs/ObservableList.md

@@ -0,0 +1,70 @@
+# ObservableList module
+
+`ObservableList` lives in explicit module `lyng.observable`.
+
+Import it first:
+
+    import lyng.observable
+    >>> void
+
+Create from a regular mutable list:
+
+    import lyng.observable
+    val xs = [1,2,3].observable()
+    assert(xs is ObservableList<Int>)
+    assertEquals([1,2,3], xs)
+    >>> void
+
+## Hook flow
+
+Event order is:
+1. `beforeChange(change)` listeners
+2. mutation commit
+3. `onChange(change)` listeners
+4. `changes()` flow emission
+
+Rejection is done by throwing in `beforeChange`.
+
+    import lyng.observable
+    val xs = [1,2].observable()
+    xs.beforeChange {
+        throw ChangeRejectionException("no mutation")
+    }
+    assertThrows(ChangeRejectionException) { xs += 3 }
+    assertEquals([1,2], xs)
+    >>> void
+
+## Subscriptions
+
+`beforeChange` and `onChange` return `Subscription`.
+Call `cancel()` to unsubscribe.
+
+    import lyng.observable
+    val xs = [1].observable()
+    var hits = 0
+    val sub = xs.onChange { hits++ }
+    xs += 2
+    sub.cancel()
+    xs += 3
+    assertEquals(1, hits)
+    >>> void
+
+## Change events
+
+`changes()` returns `Flow<ListChange<T>>` with concrete event classes:
+- `ListInsert`
+- `ListSet`
+- `ListRemove`
+- `ListClear`
+- `ListReorder`
+
+    import lyng.observable
+    val xs = [10,20].observable()
+    val it = xs.changes().iterator()
+    xs[1] = 200
+    val ev = it.next()
+    assert(ev is ListSet<Int>)
+    assertEquals(20, (ev as ListSet<Int>).oldValue)
+    assertEquals(200, ev.newValue)
+    it.cancelIteration()
+    >>> void

docs/OperatorInterop.md

@@ -0,0 +1,309 @@
+# Operator Interop Registry
+
+`lyng.operators` provides a runtime registry for mixed-class binary operators.
+
+Import it when you want expressions such as:
+
+- `1 + MyType(...)`
+- `2 < MyType(...)`
+- `3 == MyType(...)`
+
+to work without modifying the built-in `Int` or `Real` classes.
+
+```lyng
+import lyng.operators
+```
+
+## Why This Exists
+
+If your class defines:
+
+```lyng
+class Amount(val value: Int) {
+    fun plus(other: Amount) = Amount(value + other.value)
+}
+```
+
+then:
+
+```lyng
+Amount(1) + Amount(2)
+```
+
+works, because the left operand already knows how to add another `Amount`.
+
+But:
+
+```lyng
+1 + Amount(2)
+```
+
+does not naturally work, because `Int` has not been rewritten to know about `Amount`.
+
+The operator interop registry solves exactly that problem.
+
+## Mental Model
+
+Registration describes a mixed pair:
+
+- left class `L`
+- right class `R`
+- common class `C`
+
+When Lyng sees `L op R`, it:
+
+1. converts `L -> C`
+2. converts `R -> C`
+3. evaluates the operator as `C op C`
+
+So the registry is a bridge, not a separate arithmetic engine.
+
+## API
+
+```lyng
+OperatorInterop.register(
+    leftClass,
+    rightClass,
+    commonClass,
+    operators,
+    leftToCommon,
+    rightToCommon
+)
+```
+
+Parameters:
+
+- `leftClass`: original left operand class
+- `rightClass`: original right operand class
+- `commonClass`: class that will actually execute the operator methods
+- `operators`: list of operators enabled for this pair
+- `leftToCommon`: conversion from left operand to common class
+- `rightToCommon`: conversion from right operand to common class
+
+## Supported Operators
+
+`BinaryOperator` values:
+
+- `Plus`
+- `Minus`
+- `Mul`
+- `Div`
+- `Mod`
+- `Compare`
+- `Equals`
+
+Meaning:
+
+- `Compare` enables `<`, `<=`, `>`, `>=`, and `<=>`
+- `Equals` enables `==` and `!=`
+
+## Minimal Working Example
+
+```lyng
+package test.decimalbox
+import lyng.operators
+
+class DecimalBox(val value: Int) {
+    fun plus(other: DecimalBox) = DecimalBox(value + other.value)
+    fun minus(other: DecimalBox) = DecimalBox(value - other.value)
+    fun mul(other: DecimalBox) = DecimalBox(value * other.value)
+    fun div(other: DecimalBox) = DecimalBox(value / other.value)
+    fun mod(other: DecimalBox) = DecimalBox(value % other.value)
+    fun compareTo(other: DecimalBox) = value <=> other.value
+}
+
+OperatorInterop.register(
+    Int,
+    DecimalBox,
+    DecimalBox,
+    [
+        BinaryOperator.Plus,
+        BinaryOperator.Minus,
+        BinaryOperator.Mul,
+        BinaryOperator.Div,
+        BinaryOperator.Mod,
+        BinaryOperator.Compare,
+        BinaryOperator.Equals
+    ],
+    { x: Int -> DecimalBox(x) },
+    { x: DecimalBox -> x }
+)
+```
+
+Then:
+
+```lyng
+import test.decimalbox
+
+assertEquals(DecimalBox(3), 1 + DecimalBox(2))
+assertEquals(DecimalBox(1), 3 - DecimalBox(2))
+assertEquals(DecimalBox(8), 4 * DecimalBox(2))
+assertEquals(DecimalBox(4), 8 / DecimalBox(2))
+assertEquals(DecimalBox(1), 7 % DecimalBox(2))
+assert(1 < DecimalBox(2))
+assert(2 <= DecimalBox(2))
+assert(3 > DecimalBox(2))
+assert(2 == DecimalBox(2))
+assert(2 != DecimalBox(3))
+```
+
+## How Decimal Uses It
+
+`lyng.decimal` uses this same mechanism so that:
+
+```lyng
+import lyng.decimal
+
+1 + 2.d
+0.5 + 1.d
+2 == 2.d
+3 > 2.d
+```
+
+work naturally even though `Int` and `Real` themselves were not edited to know `Decimal`.
+
+The shape is:
+
+- `leftClass = Int` or `Real`
+- `rightClass = Decimal`
+- `commonClass = Decimal`
+- convert built-ins into `Decimal`
+- leave `Decimal` values unchanged
+
+## Step-By-Step Pattern For Your Own Type
+
+### 1. Pick the common class
+
+Choose one class that will be the actual arithmetic domain.
+
+For numeric-like types, that is usually your own class:
+
+```lyng
+class Rational(...)
+```
+
+### 2. Implement operators on that class
+
+The common class should define the operations you plan to register.
+
+Example:
+
+```lyng
+class Rational(val num: Int, val den: Int) {
+    fun plus(other: Rational) = Rational(num * other.den + other.num * den, den * other.den)
+    fun minus(other: Rational) = Rational(num * other.den - other.num * den, den * other.den)
+    fun mul(other: Rational) = Rational(num * other.num, den * other.den)
+    fun div(other: Rational) = Rational(num * other.den, den * other.num)
+    fun compareTo(other: Rational) = (num * other.den) <=> (other.num * den)
+
+    static fun fromInt(value: Int) = Rational(value, 1)
+}
+```
+
+### 3. Register the mixed pair
+
+```lyng
+import lyng.operators
+
+OperatorInterop.register(
+    Int,
+    Rational,
+    Rational,
+    [
+        BinaryOperator.Plus,
+        BinaryOperator.Minus,
+        BinaryOperator.Mul,
+        BinaryOperator.Div,
+        BinaryOperator.Compare,
+        BinaryOperator.Equals
+    ],
+    { x: Int -> Rational.fromInt(x) },
+    { x: Rational -> x }
+)
+```
+
+### 4. Use it
+
+```lyng
+assertEquals(Rational(3, 2), 1 + Rational(1, 2))
+assert(Rational(3, 2) == Rational(3, 2))
+assert(2 > Rational(3, 2))
+```
+
+## Registering More Than One Built-in Type
+
+If you want both `Int + MyType` and `Real + MyType`, register both pairs explicitly:
+
+```lyng
+OperatorInterop.register(
+    Int,
+    MyType,
+    MyType,
+    [BinaryOperator.Plus, BinaryOperator.Compare, BinaryOperator.Equals],
+    { x: Int -> MyType.fromInt(x) },
+    { x: MyType -> x }
+)
+
+OperatorInterop.register(
+    Real,
+    MyType,
+    MyType,
+    [BinaryOperator.Plus, BinaryOperator.Compare, BinaryOperator.Equals],
+    { x: Real -> MyType.fromReal(x) },
+    { x: MyType -> x }
+)
+```
+
+Each mixed pair is independent.
+
+## Pure Lyng Registration
+
+This mechanism is intentionally useful from pure Lyng code, not only from Kotlin-backed modules.
+
+That means you can:
+
+- declare a class in Lyng
+- define its operators in Lyng
+- register mixed operand bridges in Lyng
+
+without touching compiler internals.
+
+## Where To Register
+
+Register once during module initialization.
+
+Top-level module code is a good place:
+
+```lyng
+package my.rational
+import lyng.operators
+
+class Rational(...)
+
+OperatorInterop.register(...)
+```
+
+That keeps registration close to the type declaration and makes importing the module enough to activate the interop.
+
+## What Registration Does Not Do
+
+The registry does not:
+
+- invent operators your common class does not implement
+- change the original `Int`, `Real`, or other built-ins
+- automatically cover every class pair
+- replace normal method overload resolution when the left-hand class already knows what to do
+
+It only teaches Lyng how to bridge a specific mixed pair into a common class for the listed operators.
+
+## Recommended Design Rules
+
+If you want interop to feel natural:
+
+- choose one obvious common class
+- make conversions explicit and unsurprising
+- implement `compareTo` if you want ordering operators
+- register `Equals` whenever mixed equality should work
+- keep the registered operator list minimal and accurate
+
+For decimal-like semantics, also read [Decimal.md](Decimal.md).

docs/Range.md

@@ -25,6 +25,23 @@ Exclusive end ranges are adopted from kotlin either:
     assert(4 in r)
     >>> void
 
+Descending finite ranges are explicit too:
+
+    val r = 5 downTo 1
+    assert(r.isDescending)
+    assert(r.toList() == [5,4,3,2,1])
+    >>> void
+
+Use `downUntil` when the lower bound should be excluded:
+
+    val r = 5 downUntil 1
+    assert(r.toList() == [5,4,3,2])
+    assert(1 !in r)
+    >>> void
+
+This is explicit by design: `5..1` is not treated as a reverse range. It is an
+ordinary ascending range with no values in it when iterated.
+
 In any case, we can test an object to belong to using `in` and `!in` and
 access limits:
     
@@ -45,10 +62,11 @@ are equal or within another, taking into account the end-inclusiveness:
     assert( (1..<3) in (1..3) )
     >>> void
 
-## Finite Ranges are iterable
+## Ranges are iterable
 
-So given a range with both ends, you can assume it is [Iterable]. This automatically let
-use finite ranges in loops and convert it to lists:
+Finite ranges are [Iterable] and can be used in loops and list conversions.
+Open-ended ranges are iterable only with an explicit `step`, and open-start
+ranges are never iterable.
 
     assert( [-2, -1, 0, 1] == (-2..1).toList() )
     >>> void
@@ -62,6 +80,8 @@ In spite of this you can use ranges in for loops:
     >>> 3
     >>> void
 
+The loop variable is read-only inside the loop body (behaves like a `val`).
+
 but
 
     for( i in 1..<3 ) 
@@ -70,6 +90,52 @@ but
     >>> 2
     >>> void
 
+Descending ranges work in `for` loops exactly the same way:
+
+    for( i in 3 downTo 1 )
+        println(i)
+    >>> 3
+    >>> 2
+    >>> 1
+    >>> void
+
+And with an exclusive lower bound:
+
+    for( i in 3 downUntil 1 )
+        println(i)
+    >>> 3
+    >>> 2
+    >>> void
+
+### Stepped ranges
+
+Use `step` to change the iteration increment. The range bounds still define membership,
+so iteration ends when the next value is no longer in the range.
+
+    assert( [1,3,5] == (1..5 step 2).toList() )
+    assert( [1,3] == (1..<5 step 2).toList() )
+    assert( [5,3,1] == (5 downTo 1 step 2).toList() )
+    assert( ['a','c','e'] == ('a'..'e' step 2).toList() )
+    >>> void
+
+Descending ranges still use a positive `step`; the direction comes from
+`downTo` / `downUntil`:
+
+    assert( ['e','c','a'] == ('e' downTo 'a' step 2).toList() )
+    >>> void
+
+A negative step with `downTo` / `downUntil` is invalid.
+
+Real ranges require an explicit step:
+
+    assert( [0,0.25,0.5,0.75,1.0] == (0.0..1.0 step 0.25).toList() )
+    >>> void
+
+Open-ended ranges require an explicit step to iterate:
+
+    (0.. step 1).take(3).toList()
+    >>> [0,1,2]
+
 ## Character ranges
 
 You can use Char as both ends of the closed range:
@@ -96,12 +162,15 @@ Exclusive end char ranges are supported too:
 |-----------------|------------------------------|---------------|
 | contains(other) | used in `in`                 | Range, or Any |
 | isEndInclusive  | true for '..'                | Bool          |
+| isDescending    | true for `downTo`/`downUntil`| Bool          |
 | isOpen          | at any end                   | Bool          |
 | isIntRange      | both start and end are Int   | Bool          |
-| start           |                              | Bool          |
-| end             |                              | Bool          |
+| step            | explicit iteration step      | Any?          |
+| start           |                              | Any?          |
+| end             |                              | Any?          |
 | size            | for finite ranges, see above | Long          |
 | []              | see above                    |               |
-|                 |                              |               |
+
+Ranges are also used with the `clamp(value, range)` function and the `value.clamp(range)` extension method to limit values within boundaries.
 
 [Iterable]: Iterable.md

docs/Real.md

@@ -19,6 +19,9 @@ you can use it's class to ensure type:
 |-----------------|-------------------------------------------------------------|------|
 | `.roundToInt()` | round to nearest int like round(x)                          | Int  |
 | `.toInt()`      | convert integer part of real to `Int` dropping decimal part | Int  |
+| `.isInfinite()` | true when the value is `Infinity` or `-Infinity`            | Bool |
+| `.isNaN()`      | true when the value is `NaN`                                | Bool |
+| `.clamp(range)` | clamp value within range boundaries                         | Real |
 |                 |                                                             |      |
 |                 |                                                             |      |
 |                 |                                                             |      |

docs/Regex.md

@@ -24,13 +24,14 @@ counterpart, _not match_ operator `!~`:
 
 When you need to find groups, and more detailed match information, use `Regex.find`:
 
-    val result = Regex("abc(\d)(\d)(\d)").find( "bad456 good abc123")
+    val result: RegexMatch? = Regex("abc(\d)(\d)(\d)").find( "bad456 good abc123")
     assert( result  != null )
-    assertEquals( 12 .. 17, result.range )
-    assertEquals( "abc123", result[0] )
-    assertEquals( "1", result[1] )
-    assertEquals( "2", result[2] )
-    assertEquals( "3", result[3] )
+    val match: RegexMatch = result as RegexMatch
+    assertEquals( 12 ..< 17, match.range )
+    assertEquals( "abc123", match[0] )
+    assertEquals( "1", match[1] )
+    assertEquals( "2", match[2] )
+    assertEquals( "3", match[3] )
     >>> void
 
 Note that the object `RegexMatch`, returned by [Regex.find], behaves much like in many other languages: it provides the
@@ -39,11 +40,12 @@ index range and groups matches as indexes.
 Match operator actually also provides `RegexMatch` in `$~` reserved variable (borrowed from Ruby too):
 
     assert( "bad456 good abc123" =~ "abc(\d)(\d)(\d)".re )
-    assertEquals( 12 .. 17, $~.range )
-    assertEquals( "abc123", $~[0] )
-    assertEquals( "1", $~[1] )
-    assertEquals( "2", $~[2] )
-    assertEquals( "3", $~[3] )
+    val match2: RegexMatch = $~ as RegexMatch
+    assertEquals( 12 ..< 17, match2.range )
+    assertEquals( "abc123", match2[0] )
+    assertEquals( "1", match2[1] )
+    assertEquals( "2", match2[2] )
+    assertEquals( "3", match2[3] )
     >>> void
 
 This is often more readable than calling `find`.
@@ -59,7 +61,19 @@ string can be either left or right operator, but not both:
 
 Also, string indexing is Regex-aware, and works like `Regex.find` (_not findall!_):
 
-    assert( "cd" == "abcdef"[ "c.".re ].value )
+    assert( "cd" == ("abcdef"[ "c.".re ] as RegexMatch).value )
+    >>> void
+
+Regex replacement is exposed on `String.replace` and `String.replaceFirst`:
+
+    assertEquals( "v#.#.#", "v1.2.3".replace( "\d+".re, "#" ) )
+    assertEquals( "v[1].[2].[3]", "v1.2.3".replace( "(\d+)".re ) { m -> "[" + m[1] + "]" } )
+    assertEquals( "year-04-08", "2026-04-08".replaceFirst( "\d+".re, "year" ) )
+    >>> void
+
+When `replace` takes a plain `String`, it is treated literally, not as a regex pattern:
+
+    assertEquals( "a-b-c", "a.b.c".replace( ".", "-" ) )
     >>> void
 
 
@@ -88,4 +102,3 @@ Also, string indexing is Regex-aware, and works like `Regex.find` (_not findall!
 [List]: List.md
 
 [Range]: Range.md
-

docs/Set.md

@@ -1,7 +1,8 @@
-# List built-in class
+# Set built-in class
 
 Mutable set of any objects: a group of different objects, no repetitions.
 Sets are not ordered, order of appearance does not matter.
+For immutable set values, see [ImmutableSet].
 
     val set = Set(1,2,3, "foo")
     assert( 1 in set )
@@ -26,8 +27,8 @@ no indexing. Use [set.toList] as needed.
 
     // intersection
     assertEquals( Set(1,4), Set(3, 1, 4).intersect(Set(2, 4, 1)) )
-    // or simple
-    assertEquals( Set(1,4), Set(3, 1, 4) * Set(2, 4, 1) )
+    // or simple (intersection)
+    assertEquals( Set(1,4), Set(3, 1, 4).intersect(Set(2, 4, 1)) )
 
     // To find collection elements not present in another collection, use the 
     // subtract() or `-`:
@@ -92,3 +93,4 @@ Also, it inherits methods from [Iterable].
 
 
 [Range]: Range.md
+[ImmutableSet]: ImmutableSet.md

docs/advanced_topics.md

@@ -105,6 +105,7 @@ arguments list in almost arbitrary ways. For example:
         var result = ""
         for( a in args ) result += a
     }
+    // loop variables are read-only inside the loop body
 
     assertEquals( 
         "4231", 
@@ -153,9 +154,10 @@ Function annotation can have more args specified at call time. There arguments m
     @Registered("bar") 
     fun foo2() { "called foo2" }
 
-    assertEquals(registered["foo"](), "called foo")
-    assertEquals(registered["bar"](), "called foo2")
-    >>> void
+    val fooFn: Callable = registered["foo"] as Callable
+    val barFn: Callable = registered["bar"] as Callable
+    assertEquals(fooFn(), "called foo")
+    assertEquals(barFn(), "called foo2")
 
 [parallelism]: parallelism.md
 

docs/ai_language_reference.md

@@ -0,0 +1,244 @@
+# Lyng Language Reference for AI Agents (Current Compiler State)
+
+[//]: # (excludeFromIndex)
+
+Purpose: dense, implementation-first reference for generating valid Lyng code.
+
+Primary sources used: `lynglib/src/commonMain/kotlin/net/sergeych/lyng/{Parser,Token,Compiler,Script,TypeDecl}.kt`, `lynglib/stdlib/lyng/root.lyng`, tests in `lynglib/src/commonTest` and `lynglib/src/jvmTest`.
+
+## 1. Ground Rules
+- Resolution is compile-time-first. Avoid runtime name/member lookup assumptions.
+- `lyng.stdlib` is auto-seeded for normal scripts (default import manager).
+- Use explicit casts when receiver type is unknown (`Object`/`Obj`).
+- Prefer modern null-safe operators (`?.`, `?:`/`??`, `?=`, `as?`, `!!`).
+- Do not rely on fallback opcodes or dynamic member fallback semantics.
+
+## 2. Lexical Syntax
+- Comments: `// line`, `/* block */`.
+- Strings: `"..."` or `` `...` `` (supports escapes). Multiline string content is normalized by indentation logic.
+  - AI generation preference: use `"..."` by default, including multiline strings; `"` strings are also multiline-capable and should be preferred for ordinary code/doc/SQL text. Use backtick strings mainly when the content contains many double quotes and backticks would make the source clearer.
+  - Shared escapes: `\n`, `\r`, `\t`, `\\`, `\uXXXX` (4 hex digits).
+  - Delimiter escapes: `\"` inside `"..."`, ``\` `` inside `` `...` ``.
+  - Unicode escapes use exactly 4 hex digits (for example: `"\u0416"` -> `Ж`).
+  - Unknown `\x` escapes in strings are preserved literally as two characters (`\` and `x`).
+  - String interpolation is supported:
+    - identifier form: `"$name"` or `` `$name` ``
+    - expression form: `"${expr}"` or `` `${expr}` ``
+    - escaped dollar: `"\$"`, `"$$"`, `` `\$` ``, and `` `$$` `` all produce literal `$`.
+    - `\\$x` means backslash + interpolated `x` in either delimiter form.
+  - Per-file opt-out is supported via leading comment directive:
+    - `// feature: interpolation: off`
+    - with this directive, `$...` stays literal text.
+- Numbers: `Int` (`123`, `1_000`), `Real` (`1.2`, `1e3`), hex (`0xFF`).
+- Char: `'a'`, escaped chars supported.
+  - Supported escapes: `\n`, `\r`, `\t`, `\'`, `\\`, `\uXXXX` (4 hex digits).
+  - Backslash character in a char literal must be written as `'\\'` (forms like `'\'` are invalid).
+- Labels:
+  - statement label: `loop@ for (...) { ... }`
+  - label reference: `break@loop`, `continue@loop`, `return@fnLabel`
+- Keywords/tokens include (contextual in many places):
+  - declarations: `fun`/`fn`, `val`, `var`, `class`, `object`, `interface`, `enum`, `type`, `init`
+  - modifiers: `private`, `protected`, `static`, `abstract`, `closed`, `override`, `extern`, `open`
+  - flow: `if`, `else`, `when`, `for`, `while`, `do`, `try`, `catch`, `finally`, `throw`, `return`, `break`, `continue`
+
+## 3. Literals and Core Expressions
+- Scalars: `null`, `true`, `false`, `void`.
+- List literal: `[a, b, c]`, spreads with `...`.
+  - Spread positions: beginning, middle, end are all valid: `[...a]`, `[0, ...a, 4]`, `[head, ...mid, tail]`.
+  - Spread source must be a `List` at runtime (non-list spread raises an error).
+- Map literal: `{ key: value, x:, ...otherMap }`.
+  - `x:` means shorthand `x: x`.
+  - Map spread source must be a `Map`.
+- Range literals:
+  - inclusive: `a..b`
+  - exclusive end: `a..<b`
+  - descending inclusive: `a downTo b`
+  - descending exclusive end: `a downUntil b`
+  - open-ended forms are supported (`a..`, `..b`, `..`).
+  - optional step: `a..b step 2`, `a downTo b step 2`
+- Lambda literal:
+  - with params: `{ x, y -> x + y }`
+  - implicit `it`: `{ it + 1 }`
+- Ternary conditional is supported: `cond ? thenExpr : elseExpr`.
+
+## 3.1 Splats in Calls and Lambdas
+- Declaration-side variadic parameters use ellipsis suffix:
+  - functions: `fun f(head, tail...) { ... }`
+  - lambdas: `{ x, rest... -> ... }`
+- Call-side splats use `...expr` and are expanded by argument kind:
+  - positional splat: `f(...[1,2,3])`
+  - named splat: `f(...{ a: 1, b: 2 })` (map-style)
+- Runtime acceptance for splats:
+  - positional splat accepts `List` and general `Iterable` (iterable is converted to list first).
+  - named splat accepts `Map` with string keys only.
+- Ordering/validation rules (enforced):
+  - positional argument cannot follow named arguments (except trailing-block parsing case).
+  - positional splat cannot follow named arguments.
+  - duplicate named arguments are errors (including duplicates introduced via named splat).
+  - unknown named parameters are errors.
+  - variadic parameter itself cannot be passed as a named argument (`fun g(args..., tail)` then `g(args: ...)` is invalid).
+- Trailing block + named arguments:
+  - if the last callable parameter is already provided by name in parentheses, adding a trailing block is invalid.
+
+## 4. Operators (implemented)
+- Assignment: `=`, `+=`, `-=`, `*=`, `/=`, `%=`, `?=`.
+- Logical: `||`, `&&`, unary `!`.
+- Bitwise: `|`, `^`, `&`, `~`, shifts `<<`, `>>`.
+- Equality/comparison: `==`, `!=`, `===`, `!==`, `<`, `<=`, `>`, `>=`, `<=>`, `=~`, `!~`.
+- Type/containment: `is`, `!is`, `in`, `!in`, `as`, `as?`.
+- Null-safe family:
+  - member access: `?.`
+  - safe index: `?[i]`, `?[i, j]`
+  - safe invoke: `?(...)`
+  - safe block invoke: `?{ ... }`
+  - elvis: `?:` and `??`.
+- Increment/decrement: prefix and postfix `++`, `--`.
+- Indexing syntax:
+  - single selector: `a[i]`
+  - multiple selectors: `a[i, j, k]`
+  - language-level indexing with multiple selectors is passed to `getAt`/`putAt` as one list-like index object, not as multiple method arguments.
+  - indexers can also be supplied by extension members, including named singleton `object` receivers via `override fun Storage.getAt(...)` / `putAt(...)`.
+  - example: `m[0..2, 2]`.
+
+## 5. Declarations
+- Variables:
+  - `val` immutable, `var` mutable.
+  - top-level/local `val` must be initialized.
+  - class `val` may be late-initialized, but must be assigned in class body/init before class parse ends.
+  - destructuring declaration: `val [a, b, rest...] = expr`.
+  - destructuring declaration details:
+    - allowed in `val` and `var` declarations.
+    - supports nested patterns: `val [a, [b, c...], d] = rhs`.
+    - supports at most one splat (`...`) per pattern level.
+    - RHS must be a `List`.
+    - without splat: RHS must have at least as many elements as pattern arity.
+    - with splat: head/tail elements are bound directly, splat receives a `List`.
+- Functions:
+  - `fun` and `fn` are equivalent.
+  - full body: `fun f(x) { ... }`
+  - shorthand: `fun f(x) = expr`.
+  - generics: `fun f<T>(x: T): T`.
+  - extension functions: `fun Type.name(...) { ... }`.
+  - named singleton `object` declarations can be extension receivers too: `fun Config.describe(...) { ... }`, `val Config.tag get() = ...`.
+  - static extension functions are callable on the type object: `static fun List<T>.fill(...)` -> `List.fill(...)`.
+  - delegated callable: `fun f(...) by delegate`.
+- Type aliases:
+  - `type Name = TypeExpr`
+  - generic: `type Box<T> = List<T>`
+  - aliases are expanded structurally.
+- Classes/objects/enums/interfaces:
+  - `interface` is parsed as abstract class synonym.
+  - `object` supports named singleton and anonymous object expression forms.
+  - enums support lifted entries: `enum E* { A, B }`.
+  - multiple inheritance is supported; override is enforced when overriding base members.
+- Properties/accessors in class body:
+  - accessor form supports `get`/`set`, including `private set`/`protected set`.
+
+## 6. Control Flow
+- `if` is expression-like.
+- `compile if (cond) { ... } else { ... }` is a compile-time-only conditional.
+  - current condition grammar is restricted to `defined(NameOr.Package)`, `!`, `&&`, `||`, and parentheses.
+  - the untaken branch is skipped by the compiler and is not name-resolved or type-checked.
+- `when(value) { ... }` supported.
+  - branch conditions support equality, `in`, `!in`, `is`, `!is`, and `nullable` predicate.
+  - `when { ... }` (subject-less) is currently not implemented.
+- Loops: `for`, `while`, `do ... while`.
+  - loop `else` blocks are supported.
+  - `break value` can return a loop result.
+- Exceptions: `try/catch/finally`, `throw`.
+
+## 6.1 Destructuring Assignment (implemented)
+- Reassignment form is supported (not only declaration):
+  - `[x, y] = [y, x]`
+- Semantics match destructuring declaration:
+  - nested patterns allowed.
+  - at most one splat per pattern level.
+  - RHS must be a `List`.
+  - too few RHS elements raises runtime error.
+- Targets in pattern are variables parsed from identifier patterns.
+
+## 7. Type System (current behavior)
+- Non-null by default (`T`), nullable with `T?`.
+- `as` (checked cast), `as?` (safe cast returning `null`), `!!` non-null assertion.
+- Type expressions support:
+  - unions `A | B`
+  - intersections `A & B`
+  - function types `(A, B)->R` and receiver form `Receiver.(A)->R`
+  - variadics in function type via ellipsis (`T...`)
+- Generics:
+  - type params on classes/functions/type aliases
+  - bounds via `:` with union/intersection expressions
+  - declaration-site variance via `in` / `out`
+- Generic function/class/type syntax examples:
+  - function: `fun choose<T>(a: T, b: T): T = a`
+  - class: `class Box<T>(val value: T)`
+  - alias: `type PairList<T> = List<List<T>>`
+- Untyped params default to `Object` (`x`) or `Object?` (`x?` shorthand).
+- Untyped `var x` starts as `Unset`; first assignment fixes type tracking in compiler.
+
+## 7.1 Generics Runtime Model and Bounds (AI-critical)
+- Lyng generic type information is operational in script execution contexts; do not assume JVM-style full erasure.
+- Generic call type arguments can be:
+  - explicit at call site (`f<Int>(1)` style),
+  - inferred from runtime values/declared arg types,
+  - defaulted from type parameter defaults (or `Any` fallback).
+- At function execution, generic type parameters are runtime-bound as constants in scope:
+  - simple non-null class-like types are bound as `ObjClass`,
+  - complex/nullable/union/intersection forms are bound as `ObjTypeExpr`.
+- Practical implication for generated code:
+  - inside generic code, treat type params as usable type objects in `is`/`in`/type-expression logic (not as purely compile-time placeholders).
+  - example pattern: `if (value is T) { ... }`.
+- Bound syntax (implemented):
+  - intersection bound: `fun f<T: A & B>(x: T) { ... }`
+  - union bound: `fun g<T: A | B>(x: T) { ... }`
+- Bound checks happen at two points:
+  - compile-time call checking for resolvable generic calls,
+  - runtime re-check while binding type params for actual invocation.
+- Bound satisfaction is currently class-hierarchy based for class-resolvable parts (including union/intersection combination rules).
+- Keep expectations realistic:
+  - extern-generic runtime ABI for full instance-level generic metadata is still proposal-level (`proposals/extern_generic_runtime_abi.md`), so avoid assuming fully materialized generic-instance metadata everywhere.
+
+## 7.2 Differences vs Java / Kotlin / Scala
+- Java:
+  - Java generics are erased at runtime (except reflection metadata and raw `Class` tokens).
+  - Lyng generic params in script execution are runtime-bound type objects, so generated code can reason about `T` directly.
+- Kotlin:
+  - Kotlin on JVM is mostly erased; full runtime type access usually needs `inline reified`.
+  - Lyng generic function execution binds `T` without requiring an inline/reified escape hatch.
+- Scala:
+  - Scala has richer static typing but still runs on JVM erasure model unless carrying explicit runtime evidence (`TypeTag`, etc.).
+  - Lyng exposes runtime-bound type expressions/classes directly in generic execution scope.
+- AI generation rule:
+  - do not port JVM-language assumptions like “`T` unavailable at runtime unless reified/tagged”.
+  - in Lyng, prefer direct type-expression-driven branching when useful, but avoid assuming extern object generic args are always introspectable today.
+
+## 8. OOP, Members, and Dispatch
+- Multiple inheritance with C3-style linearization behavior is implemented in class machinery.
+- Disambiguation helpers are supported:
+  - qualified this: `this@Base.member()`
+  - cast view: `(obj as Base).member()`
+- On unknown receiver types, compiler allows only Object-safe members:
+  - `toString`, `toInspectString`, `let`, `also`, `apply`, `run`
+- Other members require known receiver type or explicit cast.
+
+## 9. Delegation (`by`)
+- Works for `val`, `var`, and `fun`.
+- Expected delegate hooks in practice:
+  - `getValue(thisRef, name)`
+  - `setValue(thisRef, name, newValue)`
+  - `invoke(thisRef, name, args...)` for delegated callables
+  - optional `bind(name, access, thisRef)`
+- `@Transient` is recognized for declarations/params and affects serialization/equality behavior.
+
+## 10. Modules and Imports
+- `package` and `import module.name` are supported.
+- Import form is module-only (no aliasing/selective import syntax in parser).
+- Default module ecosystem includes:
+  - auto-seeded: `lyng.stdlib`
+  - available by import: `lyng.observable`, `lyng.buffer`, `lyng.serialization`, `lyng.time`
+  - extra module (when installed): `lyng.io.fs`, `lyng.io.process`
+
+## 11. Current Limitations / Avoid
+- No subject-less `when { ... }` yet.
+- No regex literal tokenization (`/.../`); use `Regex("...")` or `"...".re`.
+- Do not generate runtime name fallback patterns from legacy docs.

docs/ai_notes_wasm_generation_bug.md

@@ -0,0 +1,17 @@
+# AI notes: avoid Kotlin/Wasm invalid IR with suspend lambdas
+
+[//]: # (excludeFromIndex)
+
+## Do
+- Prefer explicit `object : Statement()` with `override suspend fun execute(...)` when building compiler statements.
+- Keep `Statement` objects non-lambda, especially in compiler hot paths like parsing/var declarations.
+- If you need conditional behavior, return early in `execute` instead of wrapping `parseExpression()` with `statement(...) { ... }`.
+- When wasmJs tests hang in the browser, first check `wasmJsNodeTest` for a compile error; hangs often mean module instantiation failed.
+
+## Don't
+- Do not create suspend lambdas inside `Statement` factories (`statement { ... }`) for wasm targets.
+- Do not "fix" hangs by increasing browser timeouts; it masks invalid wasm generation.
+
+## Debugging tips
+- Look for `$invokeCOROUTINE$` in wasm function names when mapping failures.
+- If node test logs a wasm compile error, the browser hang is likely the same root cause.

docs/ai_stdlib_reference.md

@@ -0,0 +1,99 @@
+# Lyng Stdlib Reference for AI Agents (Compact)
+
+[//]: # (excludeFromIndex)
+
+Purpose: fast overview of what is available by default and what must be imported.
+
+Sources: `lynglib/src/commonMain/kotlin/net/sergeych/lyng/Script.kt`, `lynglib/stdlib/lyng/root.lyng`, `lynglib/src/commonMain/kotlin/net/sergeych/lyng/stdlib_included/observable_lyng.kt`.
+
+## 1. Default Availability
+- Normal scripts are auto-seeded with `lyng.stdlib` (default import manager path).
+- Root runtime scope also exposes global constants/functions directly.
+
+## 2. Core Global Functions (Root Scope)
+- IO/debug: `print`, `println`, `traceScope`.
+- Invocation/util: `call`, `run`, `dynamic`, `cached`, `lazy`.
+- Assertions/tests: `assert`, `assertEquals`/`assertEqual`, `assertNotEquals`, `assertThrows`.
+- Preconditions: `require`, `check`.
+- Async/concurrency: `launch`, `yield`, `flow`, `delay`.
+  - `Deferred.cancel()` cancels an active task.
+  - `Deferred.await()` throws `CancellationException` if that task was cancelled.
+  - `Iterable<Deferred>.joinAll()` awaits every deferred in iteration order and returns a `List` of results.
+- Math: `floor`, `ceil`, `round`, `sin`, `cos`, `tan`, `asin`, `acos`, `atan`, `sinh`, `cosh`, `tanh`, `asinh`, `acosh`, `atanh`, `exp`, `ln`, `log10`, `log2`, `pow`, `sqrt`, `abs`, `clamp`.
+  - These helpers also accept `lyng.decimal.Decimal`.
+  - Exact Decimal path today: `abs`, `floor`, `ceil`, `round`, and `pow` with integral exponent.
+  - Temporary Decimal path for the rest: convert `Decimal -> Real`, compute, then convert back to `Decimal`.
+  - Treat that bridge as temporary; prefer native Decimal implementations when they become available.
+
+## 3. Core Global Constants/Types
+- Values: `Unset`, `π`.
+- Primitive/class symbols: `Object`, `Int`, `Real`, `Bool`, `Char`, `String`, `Class`, `Callable`.
+- Collections/types: `Iterable`, `Iterator`, `Collection`, `Array`, `List`, `ImmutableList`, `Set`, `ImmutableSet`, `Map`, `ImmutableMap`, `MapEntry`, `Range`, `RingBuffer`.
+- Random: singleton `Random` and class `SeededRandom`.
+- Async types: `Deferred`, `CompletableDeferred`, `Mutex`, `Flow`, `FlowBuilder`.
+- Async exception: `CancellationException`.
+- Delegation types: `Delegate`, `DelegateContext`.
+- Regex types: `Regex`, `RegexMatch`.
+- Also present: `Math.PI` namespace constant.
+
+## 4. `lyng.stdlib` Module Surface (from `root.lyng`)
+### 4.1 Extern class declarations
+- Exceptions/delegation base: `Exception`, `CancellationException`, `IllegalArgumentException`, `NotImplementedException`, `Delegate`.
+- Collections and iterables: `Iterable<T>`, `Iterator<T>`, `Collection<T>`, `Array<T>`, `List<T>`, `ImmutableList<T>`, `Set<T>`, `ImmutableSet<T>`, `Map<K,V>`, `ImmutableMap<K,V>`, `MapEntry<K,V>`, `RingBuffer<T>`.
+- Host iterator bridge: `KotlinIterator<T>`.
+- Random APIs: `extern object Random`, `extern class SeededRandom`.
+
+### 4.2 High-use extension APIs
+- Iteration/filtering: `forEach`, `filter`, `filterFlow`, `filterNotNull`, `filterFlowNotNull`, `drop`, `dropLast`, `takeLast`.
+- Search/predicates: `findFirst`, `findFirstOrNull`, `any`, `all`, `count`, `first`, `last`.
+- Mapping/aggregation: `map`, `flatMap`, `flatten`, `sum`, `sumOf`, `minOf`, `maxOf`.
+- Ordering and list building: `sorted`, `sortedBy`, `shuffled`, `List.sort`, `List.sortBy`, `List.fill`.
+  - `List.fill(size) { index -> ... }` constructs a new `List<T>` by evaluating the block once per index from `0` to `size - 1`.
+- String helper: `joinToString`, `String.re`.
+
+### 4.3 Delegation helpers
+- `enum DelegateAccess { Val, Var, Callable }`
+- `interface Delegate<T,ThisRefType=void>` with `getValue`, `setValue`, `invoke`, `bind`.
+- `class lazy<T,...>` delegate implementation.
+- `fun with(self, block)` helper.
+
+### 4.4 Other module-level symbols
+- `$~` (last regex match object).
+- `TODO(message?)` utility.
+- `StackTraceEntry` class.
+- `Random.nextInt()`, `Random.nextFloat()`, `Random.next(range)`, `Random.seeded(seed)`.
+- `SeededRandom.nextInt()`, `SeededRandom.nextFloat()`, `SeededRandom.next(range)`.
+
+## 5. Additional Built-in Modules (import explicitly)
+- `import lyng.observable`
+  - `Observable`, `Subscription`, `ObservableList`, `ListChange` and change subtypes, `ChangeRejectionException`.
+- `import lyng.decimal`
+  - `Decimal`, `DecimalContext`, `DecimalRounding`, `withDecimalContext(...)`.
+  - Kotlin host helper: `ScopeFacade.newDecimal(BigDecimal)` wraps an ionspin host decimal as a Lyng `Decimal`.
+- `import lyng.complex`
+  - `Complex`, `complex(re, im)`, `cis(angle)`, and numeric embedding extensions such as `2.i` / `3.re`.
+- `import lyng.matrix`
+  - `Matrix`, `Vector`, `matrix(rows)`, `vector(values)`, dense linear algebra, inversion, solving, and matrix slicing with `m[row, col]`.
+- `import lyng.buffer`
+  - `Buffer`, `MutableBuffer`.
+- `import lyng.legacy_digest`
+  - `LegacyDigest.sha1(data): String` — SHA-1 hex digest; `data` may be `String` (UTF-8) or `Buffer` (raw bytes).
+  - ⚠️ Cryptographically broken. Use only for legacy protocol / file-format compatibility.
+- `import lyng.serialization`
+  - `Lynon` serialization utilities.
+- `import lyng.time`
+  - `Instant`, `Date`, `DateTime`, `Duration`, and module `delay`.
+
+## 6. Optional (lyngio) Modules
+Requires installing `lyngio` into the import manager from host code.
+- `import lyng.io.fs` (filesystem `Path` API)
+- `import lyng.io.process` (process execution API)
+- `import lyng.io.console` (console capabilities, geometry, ANSI/output, events)
+- `import lyng.io.http` (HTTP/HTTPS client API)
+- `import lyng.io.ws` (WebSocket client API; currently supported on JVM, capability-gated elsewhere)
+- `import lyng.io.net` (TCP/UDP transport API; currently supported on JVM, capability-gated elsewhere)
+
+## 7. AI Generation Tips
+- Assume `lyng.stdlib` APIs exist in regular script contexts.
+- For platform-sensitive code (`fs`, `process`, `console`, `http`, `ws`, `net`), gate assumptions and mention required module install.
+- Prefer extension-method style (`items.filter { ... }`) and standard scope helpers (`let`/`also`/`apply`/`run`).

docs/declaring_arguments.md

@@ -34,13 +34,18 @@ Valid examples:
 
 Ellipsis are used to declare variadic arguments. It basically means "all the arguments available here". It means, ellipsis argument could be in any part of the list, being, end or middle, but there could be only one ellipsis argument and it must not have default value, its default value is always `[]`, en empty list.
 
+Ellipsis can also appear in **function types** to denote a variadic position:
+
+    var f: (Int, Object..., String)->Real
+    var anyArgs: (...)->Int   // shorthand for (Object...)->Int
+
 Ellipsis argument receives what is left from arguments after processing regular one that could be before or after.
 
 Ellipsis could be a first argument:
 
     fun testCountArgs(data...,size) {
         assert(size is Int)
-        assertEquals(size, data.size)
+        assertEquals(size, (data as List).size)
     }
     testCountArgs( 1, 2, "three", 3)
     >>> void
@@ -49,7 +54,7 @@ Ellipsis could also be a last one:
 
     fun testCountArgs(size, data...) {
         assert(size is Int)
-        assertEquals(size, data.size)
+        assertEquals(size, (data as List).size)
     }
     testCountArgs( 3, 10, 2, "three")
     >>> void
@@ -58,7 +63,7 @@ Or in the middle:
 
     fun testCountArgs(size, data..., textToReturn) {
         assert(size is Int)
-        assertEquals(size, data.size)
+        assertEquals(size, (data as List).size)
         textToReturn
     }
     testCountArgs( 3, 10, 2, "three", "All OK")

docs/delegation.md

@@ -57,7 +57,8 @@ class lazy(val creator) : Delegate {
 
     override fun getValue(thisRef, name) {
         if (value == Unset) {
-            value = creator()
+            // calculate value using thisRef as this:
+            value = with(thisRef) creator()
         }
         value
     }
@@ -150,6 +151,24 @@ fun test() {
 }
 ```
 
+### 6. Map as a Delegate
+
+Maps can be used as delegates for `val` and `var` properties. When a map is used as a delegate, it uses the property name as a key to read from or write to the map.
+
+```lyng
+val m = { "a": 1, "b": 2 }
+val a by m
+var b by m
+
+println(a) // 1
+println(b) // 2
+
+b = 42
+println(m["b"]) // 42
+```
+
+Because `Map` implements `getValue` and `setValue`, it works seamlessly with any object that needs to store its properties in a map (e.g., when implementing dynamic schemas or JSON-backed objects).
+
 ## The `bind` Hook
 
 The `bind(name, access, thisRef)` method is called exactly once when the member is being initialized. It allows the delegate to:

docs/downloads.md

@@ -0,0 +1,12 @@
+# Some resources to download
+
+## Lync CLI tool
+
+- [lyng-linuxX64.zip](/distributables/lyng-linuxX64.zip) CLI tool for linuxX64: nodependencies, small monolith executable binary.
+- [lyng-jvm.zip](/distributables/lyng-jvm.zip) JVM CLI distribution: download, unpack, and run `lyng-jvm/bin/lyng`.
+
+## IDE plugins
+
+- [lyng-textmate.zip](../../lyng/distributables/lyng-textmate.zip) Texmate-compatible bundle with syntax coloring (could be outdated)
+
+- [lyng-idea-0.0.5-SNAPSHOT.zip](/distributables/lyng-idea-0.0.5-SNAPSHOT.zip) - plugin for IntelliJ-compatible IDE

docs/embedding.md

@@ -1,10 +1,12 @@
 # Embedding Lyng in your Kotlin project
 
+[//]: # (topMenu)
+
 Lyng is a tiny, embeddable, Kotlin‑first scripting language. This page shows, step by step, how to:
 
 - add Lyng to your build
 - create a runtime and execute scripts
-- define functions and variables from Kotlin
+- declare extern globals in Lyng and bind them from Kotlin
 - read variable values back in Kotlin
 - call Lyng functions from Kotlin
 - create your own packages and import them in Lyng
@@ -36,21 +38,60 @@ dependencies {
 
 If you use Kotlin Multiplatform, add the dependency in the `commonMain` source set (and platform‑specific sets if you need platform APIs).
 
-### 2) Create a runtime (Scope) and execute scripts
+### 2) Preferred runtime: `EvalSession`
 
-The easiest way to get a ready‑to‑use scope with standard packages is via `Script.newScope()`.
+For host applications, prefer `EvalSession` as the main way to run scripts.
+It owns one reusable Lyng scope, serializes `eval(...)` calls, and governs coroutines started from Lyng `launch { ... }`.
+
+Main entrypoints:
+
+- `session.eval(code)` / `session.eval(source)`
+- `session.getScope()` when you need low-level binding APIs
+- `session.cancel()` to cancel active session-owned coroutines
+- `session.join()` to wait for active session-owned coroutines
+
+```kotlin
+fun main() = kotlinx.coroutines.runBlocking {
+    val session = EvalSession()
+
+    // Evaluate a one‑liner
+    val result = session.eval("1 + 2 * 3")
+    println("Lyng result: $result") // ObjReal/ObjInt etc.
+
+    // Optional lifecycle management
+    session.join()
+}
+```
+
+The session creates its underlying scope lazily. If you need raw low-level APIs, get the scope explicitly:
+
+```kotlin
+val session = EvalSession()
+val scope = session.getScope()
+```
+
+Use `cancel()` / `join()` to govern async work started by scripts:
+
+```kotlin
+val session = EvalSession()
+session.eval("""launch { delay(1000); println("done") }""")
+session.cancel()
+session.join()
+```
+
+### 2.1) Low-level runtime: `Scope`
+
+Use `Scope` directly when you intentionally want lower-level control.
 
 ```kotlin
 fun main() = kotlinx.coroutines.runBlocking {
     val scope = Script.newScope() // suspends on first init
-
-    // Evaluate a one‑liner
     val result = scope.eval("1 + 2 * 3")
-    println("Lyng result: $result") // ObjReal/ObjInt etc.
+    println("Lyng result: $result")
 }
 ```
 
-You can also pre‑compile a script and execute it multiple times:
+You can also pre‑compile a script and execute it multiple times on the same scope:
 
 ```kotlin
 val script = Compiler.compile("""
@@ -63,34 +104,98 @@ val run1 = script.execute(scope)
 val run2 = script.execute(scope)
 ```
 
-`Scope.eval("...")` is a shortcut that compiles and executes on the given scope.
+`Scope.eval("...")` is the low-level shortcut that compiles and executes on the given scope.
+For most embedding use cases, prefer `session.eval("...")`.
 
-### 3) Define variables from Kotlin
+### 3) Preferred: bind extern globals from Kotlin
 
-To expose data to Lyng, add constants (read‑only) or mutable variables to the scope. All values in Lyng are `Obj` instances; the core types live in `net.sergeych.lyng.obj`.
+For module-level APIs, the default workflow is:
+
+1. declare globals in Lyng using `extern fun` / `extern val` / `extern var`;
+2. bind Kotlin implementation via `ModuleScope.globalBinder()`.
+
+This is also the recommended way to expose a Kotlin-backed value that should behave like a true
+Lyng global variable/property. If you need `x` to read/write through Kotlin on every access, use
+`extern var` / `extern val` plus `bindGlobalVar(...)`.
+
+Do not use `addConst(...)` for this case: `addConst(...)` installs a value, not a Kotlin-backed
+property accessor. It is appropriate for fixed values and objects, but not for a global that should
+delegate reads/writes back into Kotlin state.
 
 ```kotlin
-// Read‑only constant
-scope.addConst("pi", ObjReal(3.14159))
+import net.sergeych.lyng.bridge.*
+import net.sergeych.lyng.obj.ObjInt
+import net.sergeych.lyng.obj.ObjString
 
-// Mutable variable: create or update
-scope.addOrUpdateItem("counter", ObjInt(0))
+val session = EvalSession()
+val scope = session.getScope()
+val im = Script.defaultImportManager.copy()
+im.addPackage("my.api") { module ->
+    module.eval("""
+        extern fun globalFun(v: Int): Int
+        extern var globalProp: String
+        extern val globalVersion: String
+    """.trimIndent())
 
-// Use it from Lyng
-scope.eval("counter = counter + 1")
+    val binder = module.globalBinder()
+
+    binder.bindGlobalFun1<Int>("globalFun") { v ->
+        ObjInt.of((v + 1).toLong())
+    }
+
+    var prop = "initial"
+    binder.bindGlobalVar(
+        name = "globalProp",
+        get = { prop },
+        set = { prop = it }
+    )
+
+    binder.bindGlobalVar(
+        name = "globalVersion",
+        get = { "1.0.0" } // readonly: setter omitted
+    )
+}
 ```
 
-Tip: Lyng values can be converted back to Kotlin with `toKotlin(scope)`:
+Usage from Lyng:
 
-```kotlin
-val current = (scope.eval("counter")).toKotlin(scope) // Any? (e.g., Int/Double/String/List)
+```lyng
+import my.api
+
+assertEquals(42, globalFun(41))
+assertEquals("initial", globalProp)
+globalProp = "changed"
+assertEquals("changed", globalProp)
+assertEquals("1.0.0", globalVersion)
 ```
 
-### 4) Add Kotlin‑backed functions
+Minimal rule of thumb:
 
-Use `Scope.addFn`/`addVoidFn` to register functions implemented in Kotlin. Inside the lambda, use `this.args` to access arguments and return an `Obj`.
+- use `bindGlobalFun(...)` for global functions
+- use `bindGlobalVar(...)` for Kotlin-backed global variables/properties
+- use `addConst(...)` only for fixed values/objects that do not need getter/setter behavior
+
+For custom argument handling and full runtime access:
 
 ```kotlin
+binder.bindGlobalFun("sum3") {
+    requireExactCount(3)
+    ObjInt.of((int(0) + int(1) + int(2)).toLong())
+}
+
+binder.bindGlobalFunRaw("echoRaw") { _, args ->
+    args.firstAndOnly()
+}
+```
+
+### 4) Low-level: direct functions/variables from Kotlin
+
+Use this when you intentionally want raw `Scope` APIs. For most module APIs, prefer section 3.
+
+```kotlin
+val session = EvalSession()
+val scope = session.getScope()
+
 // A function returning value
 scope.addFn<ObjInt>("inc") {
     val x = args.firstAndOnly() as ObjInt
@@ -103,17 +208,101 @@ scope.addVoidFn("log") {
     println(items.joinToString(" ") { it.toString(this).value })
 }
 
+// When adding a member function to a class, you can use isOverride = true
+// myClass.addFn("toString", isOverride = true) {
+//     ObjString("Custom string representation")
+// }
+
 // Call them from Lyng
-scope.eval("val y = inc(41); log('Answer:', y)")
+session.eval("val y = inc(41); log('Answer:', y)")
 ```
 
 You can register multiple names (aliases) at once: `addFn<ObjInt>("inc", "increment") { ... }`.
 
+Scope-backed Kotlin lambdas receive a `ScopeFacade` (not a full `Scope`). For migration and convenience, these utilities are available on the facade:
+
+- Access: `args`, `pos`, `thisObj`, `get(name)`
+- Invocation: `call(...)`, `resolve(...)`, `assign(...)`, `toStringOf(...)`, `inspect(...)`, `trace(...)`
+- Args helpers: `requiredArg<T>()`, `requireOnlyArg<T>()`, `requireExactCount(...)`, `requireNoArgs()`, `thisAs<T>()`
+- Errors: `raiseError(...)`, `raiseClassCastError(...)`, `raiseIllegalArgument(...)`, `raiseIllegalState(...)`, `raiseNoSuchElement(...)`,
+  `raiseSymbolNotFound(...)`, `raiseNotImplemented(...)`, `raiseNPE()`, `raiseIndexOutOfBounds(...)`, `raiseIllegalAssignment(...)`,
+  `raiseUnset(...)`, `raiseNotFound(...)`, `raiseAssertionFailed(...)`, `raiseIllegalOperation(...)`, `raiseIterationFinished()`
+
+If you truly need the full `Scope` (e.g., for low-level interop), use `requireScope()` explicitly.
+
+### 4.5) Indexers from Kotlin: `getAt` and `putAt`
+
+Lyng bracket syntax is dispatched through `getAt` and `putAt`.
+
+That means:
+
+- `x[i]` calls `getAt(index)`
+- `x[i] = value` calls `putAt(index, value)` or `setAt(index, value)`
+- field-like `x["name"]` also uses the same index path unless you expose a real field/property
+
+For Kotlin-backed classes, bind indexers as ordinary methods named `getAt` and `putAt`:
+
+```kotlin
+moduleScope.eval("""
+    extern class Grid {
+        override fun getAt(index: List<Int>): Int
+        override fun putAt(index: List<Int>, value: Int): void
+    }
+""".trimIndent())
+
+moduleScope.bind("Grid") {
+    init { _ -> data = IntArray(4) }
+
+    addFun("getAt") {
+        val index = args.requiredArg<ObjList>(0)
+        val row = (index.list[0] as ObjInt).value.toInt()
+        val col = (index.list[1] as ObjInt).value.toInt()
+        val data = (thisObj as ObjInstance).data as IntArray
+        ObjInt.of(data[row * 2 + col].toLong())
+    }
+
+    addFun("putAt") {
+        val index = args.requiredArg<ObjList>(0)
+        val value = args.requiredArg<ObjInt>(1).value.toInt()
+        val row = (index.list[0] as ObjInt).value.toInt()
+        val col = (index.list[1] as ObjInt).value.toInt()
+        val data = (thisObj as ObjInstance).data as IntArray
+        data[row * 2 + col] = value
+        ObjVoid
+    }
+}
+```
+
+Usage from Lyng:
+
+```lyng
+val g = Grid()
+g[0, 1] = 42
+assertEquals(42, g[0, 1])
+```
+
+Important rule: multiple selectors inside brackets are packed into one index object.
+So:
+
+- `x[i]` passes `i`
+- `x[i, j]` passes a `List` containing `[i, j]`
+- `x[i, j, k]` passes `[i, j, k]`
+
+This applies equally to:
+
+- Kotlin-backed classes
+- Lyng classes overriding `getAt`
+- `dynamic { get { ... } set { ... } }`
+
+If you want multi-axis slicing semantics, decode that list yourself in `getAt`.
+
 ### 5) Add Kotlin‑backed fields
 
 If you need a simple field (with a value) instead of a computed property, use `createField`. This adds a field to the class that will be present in all its instances.
 
 ```kotlin
+val session = EvalSession()
+val scope = session.getScope()
 val myClass = ObjClass("MyClass")
 
 // Add a read-only field (constant)
@@ -122,6 +311,9 @@ myClass.createField("version", ObjString("1.0.0"), isMutable = false)
 // Add a mutable field with an initial value
 myClass.createField("count", ObjInt(0), isMutable = true)
 
+// If you are overriding a field from a base class, use isOverride = true
+// myClass.createField("someBaseField", ObjInt(42), isOverride = true)
+
 scope.addConst("MyClass", myClass)
 ```
 
@@ -138,6 +330,8 @@ println(instance.count)   // -> 5
 Properties in Lyng are pure accessors (getters and setters) and do not have automatic backing fields. You can add them to a class using `addProperty`.
 
 ```kotlin
+val session = EvalSession()
+val scope = session.getScope()
 val myClass = ObjClass("MyClass")
 var internalValue: Long = 10
 
@@ -153,6 +347,16 @@ myClass.addProperty(
     }
 )
 
+// You can also create an ObjProperty explicitly
+val explicitProp = ObjProperty(
+    name = "hexValue",
+    getter = statement { ObjString(internalValue.toString(16)) }
+)
+myClass.addProperty("hexValue", prop = explicitProp)
+
+// Use isOverride = true when overriding a property from a base class
+// myClass.addProperty("baseProp", getter = { ... }, isOverride = true)
+
 scope.addConst("MyClass", myClass)
 ```
 
@@ -164,17 +368,172 @@ instance.value = 42
 println(instance.value) // -> 42
 ```
 
+### 6.5) Preferred: bind Kotlin implementations to declared Lyng classes
+
+For extensions and libraries, the **preferred** workflow is Lyng‑first: declare the class and its members in Lyng, then bind the Kotlin implementations using the bridge.
+
+This keeps Lyng semantics (visibility, overrides, type checks) in Lyng, while Kotlin supplies the behavior.
+
+Pure extern declarations use the simplified rule set:
+- `extern class` / `extern object` are declaration-only ABI surfaces.
+- Every member in their body is implicitly extern (you may still write `extern`, but it is redundant).
+- Plain Lyng member implementations inside `extern class` / `extern object` are not allowed.
+- Put Lyng behavior into regular classes or extension methods.
+
+```lyng
+// Lyng side (in a module)
+class Counter {
+    extern var value: Int
+    extern fun inc(by: Int): Int
+}
+```
+
+Note: members of `extern class` / `extern object` are treated as extern by default, so the compiler emits ABI slots that Kotlin bindings attach to. This applies to functions and properties bound via `addFun` / `addVal` / `addVar`.
+
+Example of pure extern class declaration:
+
+```lyng
+extern class HostCounter {
+    var value: Int
+    fun inc(by: Int): Int
+}
+```
+
+If you need Lyng-side convenience behavior, add it as an extension:
+
+```lyng
+fun HostCounter.bump() = inc(1)
+```
+
+```kotlin
+// Kotlin side (binding)
+val moduleScope = Script.newScope() // or an existing module scope
+moduleScope.eval("class Counter { extern var value: Int; extern fun inc(by: Int): Int }")
+
+moduleScope.bind("Counter") {
+    addVar(
+        name = "value",
+        get = { thisObj.readField(this, "value").value },
+        set = { v -> thisObj.writeField(this, "value", v) }
+    )
+    addFun("inc") {
+        val by = args.requiredArg<ObjInt>(0).value
+        val current = thisObj.readField(this, "value").value as ObjInt
+        val next = ObjInt(current.value + by)
+        thisObj.writeField(this, "value", next)
+        next
+    }
+}
+```
+
+Notes:
+
+- Binding must happen **before** the first instance is created.
+- Use [LyngClassBridge] to bind by name/module, or by an already resolved `ObjClass`.
+- Use `ObjInstance.data` / `ObjClass.classData` to attach Kotlin‑side state when needed.
+
+### 6.5a) Bind Kotlin implementations to declared Lyng objects
+
+For `extern object` declarations, bind implementations to the singleton instance using `ModuleScope.bindObject`.
+This mirrors class binding but targets an already created object instance.
+As with class binding, you must first add/evaluate the Lyng declaration into that module scope, then bind Kotlin handlers.
+
+```kotlin
+// Kotlin side (binding)
+val moduleScope = importManager.createModuleScope(Pos.builtIn, "bridge.obj")
+
+// 1) Seed the module with the Lyng declaration first
+moduleScope.eval("""
+    extern object HostObject {
+        extern fun add(a: Int, b: Int): Int
+        extern val status: String
+        extern var count: Int
+    }
+""".trimIndent())
+
+// 2) Then bind Kotlin implementations to that declared object
+moduleScope.bindObject("HostObject") {
+    classData = "OK"
+    init { _ -> data = 0L }
+    addFun("add") {
+        val a = args.requiredArg<ObjInt>(0).value
+        val b = args.requiredArg<ObjInt>(1).value
+        ObjInt.of(a + b)
+    }
+    addVal("status") { ObjString(classData as String) }
+    addVar(
+        "count",
+        get = { ObjInt.of((thisObj as ObjInstance).data as Long) },
+        set = { value -> (thisObj as ObjInstance).data = (value as ObjInt).value }
+    )
+}
+```
+
+Notes:
+
+- Required order: declare/eval Lyng object in the module first, then call `bindObject(...)`.
+  This is the pattern covered by `BridgeBindingTest.testExternObjectBinding`.
+- Members must be extern (explicitly, or implicitly via `extern object`) so the compiler emits ABI slots for Kotlin bindings.
+- You can also bind by name/module via `LyngObjectBridge.bind(...)`.
+
+Minimal `extern fun` example:
+
+```kotlin
+val moduleScope = importManager.createModuleScope(Pos.builtIn, "bridge.ping")
+
+moduleScope.eval("""
+    extern object HostObject {
+        extern fun ping(): Int
+    }
+""".trimIndent())
+
+moduleScope.bindObject("HostObject") {
+    addFun("ping") { ObjInt.of(7) }
+}
+```
+
+### 6.6) Preferred: Kotlin reflection bridge for call‑by‑name
+
+For Kotlin code that needs dynamic access to Lyng variables, functions, or members, use the bridge resolver.
+It provides explicit, cached handles and predictable lookup rules.
+
+```kotlin
+val session = EvalSession()
+val scope = session.getScope()
+session.eval("""
+    val x = 40
+    fun add(a, b) = a + b
+    class Box { var value = 1 }
+""")
+
+val resolver = scope.resolver()
+
+// Read a top‑level value
+val x = resolver.resolveVal("x").get(scope)
+
+// Call a function by name (cached inside the resolver)
+val sum = (resolver as BridgeCallByName).callByName(scope, "add", Arguments(ObjInt(1), ObjInt(2)))
+
+// Member access
+val box = session.eval("Box()")
+val valueHandle = resolver.resolveMemberVar(box, "value")
+valueHandle.set(scope, ObjInt(10))
+val value = valueHandle.get(scope)
+```
+
 ### 7) Read variable values back in Kotlin
 
 The simplest approach: evaluate an expression that yields the value and convert it.
 
 ```kotlin
-val kotlinAnswer = scope.eval("(1 + 2) * 3").toKotlin(scope) // -> 9 (Int)
+val session = EvalSession()
+val scope = session.getScope()
+val kotlinAnswer = session.eval("(1 + 2) * 3").toKotlin(scope) // -> 9 (Int)
 
 // After scripts manipulate your vars:
 scope.addOrUpdateItem("name", ObjString("Lyng"))
-scope.eval("name = name + ' rocks!'")
-val kotlinName = scope.eval("name").toKotlin(scope) // -> "Lyng rocks!"
+session.eval("name = name + ' rocks!'")
+val kotlinName = session.eval("name").toKotlin(scope) // -> "Lyng rocks!"
 ```
 
 Advanced: you can also grab a variable record directly via `scope.get(name)` and work with its `Obj` value, but evaluating `"name"` is often clearer and enforces Lyng semantics consistently.
@@ -187,16 +546,20 @@ There are two convenient patterns.
 
 ```kotlin
 // Suppose Lyng defines: fun add(a, b) = a + b
-scope.eval("fun add(a, b) = a + b")
+val session = EvalSession()
+val scope = session.getScope()
+session.eval("fun add(a, b) = a + b")
 
-val sum = scope.eval("add(20, 22)").toKotlin(scope) // -> 42
+val sum = session.eval("add(20, 22)").toKotlin(scope) // -> 42
 ```
 
 2) Call a Lyng function by name via a prepared call scope:
 
 ```kotlin
 // Ensure the function exists in the scope
-scope.eval("fun add(a, b) = a + b")
+val session = EvalSession()
+val scope = session.getScope()
+session.eval("fun add(a, b) = a + b")
 
 // Look up the function object
 val addFn = scope.get("add")!!.value as Statement
@@ -224,27 +587,47 @@ Key concepts:
 Register a Kotlin‑built package:
 
 ```kotlin
-val scope = Script.newScope()
+import net.sergeych.lyng.bridge.*
+import net.sergeych.lyng.obj.ObjInt
+
+val session = EvalSession()
+val scope = session.getScope()
 
 // Access the import manager behind this scope
 val im: ImportManager = scope.importManager
 
 // Register a package "my.tools"
 im.addPackage("my.tools") { module: ModuleScope ->
-    // Expose symbols inside the module scope
-    module.addConst("version", ObjString("1.0"))
-    module.addFn<ObjInt>("triple") {
-        val x = args.firstAndOnly() as ObjInt
-        ObjInt(x.value * 3)
+    module.eval(
+        """
+        extern val version: String
+        extern var status: String
+        extern fun triple(x: Int): Int
+        """.trimIndent()
+    )
+    val binder = module.globalBinder()
+    var status = "ready"
+    binder.bindGlobalVar(
+        name = "version",
+        get = { "1.0" }
+    )
+    binder.bindGlobalVar(
+        name = "status",
+        get = { status },
+        set = { status = it }
+    )
+    binder.bindGlobalFun1<Int>("triple") { x ->
+        ObjInt.of((x * 3).toLong())
     }
 }
 
 // Use it from Lyng
-scope.eval("""
+session.eval("""
     import my.tools.*
     val v = triple(14)
+    status = "busy"
 """)
-val v = scope.eval("v").toKotlin(scope) // -> 42
+val v = session.eval("v").toKotlin(scope) // -> 42
 ```
 
 Register a package from Lyng source text:
@@ -258,24 +641,27 @@ val pkgText = """
 
 scope.importManager.addTextPackages(pkgText)
 
-scope.eval("""
+session.eval("""
     import math.extra.*
     val s = sqr(12)
 """)
-val s = scope.eval("s").toKotlin(scope) // -> 144
+val s = session.eval("s").toKotlin(scope) // -> 144
 ```
 
 You can also register from parsed `Source` instances via `addSourcePackages(source)`.
 
 ### 10) Executing from files, security, and isolation
 
-- To run code from a file, read it and pass to `scope.eval(text)` or compile with `Compiler.compile(Source(fileName, text))`.
+- To run code from a file, read it and pass to `session.eval(text)` or compile with `Compiler.compile(Source(fileName, text))`.
 - `ImportManager` takes an optional `SecurityManager` if you need to restrict what packages or operations are available. By default, `Script.defaultImportManager` allows everything suitable for embedded use; clamp it down in sandboxed environments.
-- For isolation, create fresh modules/scopes via `Scope.new()` or `Script.newScope()` when you need a clean environment per request.
+- For isolation, prefer a fresh `EvalSession()` per request. Use `Scope.new()` / `Script.newScope()` when you specifically need low-level raw scopes or modules.
 
 ```kotlin
-// Fresh module based on the default manager, without the standard prelude
-val isolated = net.sergeych.lyng.Scope.new()
+// Preferred per-request runtime:
+val isolatedSession = EvalSession()
+
+// Low-level fresh module based on the default manager, without the standard prelude:
+val isolatedScope = net.sergeych.lyng.Scope.new()
 ```
 
 ### 11) Tips and troubleshooting
@@ -286,6 +672,49 @@ val isolated = net.sergeych.lyng.Scope.new()
 - When registering packages, names must be unique. Register before you compile/evaluate scripts that import them.
 - To debug scope content, `scope.toString()` and `scope.trace()` can help during development.
 
+### 12) Handling and serializing exceptions
+
+When Lyng code throws an exception, it is caught in Kotlin as an `ExecutionError`. This error wraps the actual Lyng `Obj` that was thrown (which could be a built-in `ObjException` or a user-defined `ObjInstance`).
+
+To simplify handling these objects from Kotlin, several extension methods are provided on the `Obj` class. These methods work uniformly regardless of whether the exception is built-in or user-defined.
+
+#### Uniform Exception API
+
+| Method | Description |
+| :--- | :--- |
+| `obj.isLyngException()` | Returns `true` if the object is an instance of `Exception`. |
+| `obj.isInstanceOf("ClassName")` | Returns `true` if the object is an instance of the named Lyng class or its ancestors. |
+| `obj.getLyngExceptionMessage(scope?=null)` | Returns the exception message as a Kotlin `String`. |
+| `obj.getLyngExceptionMessageWithStackTrace(scope?=null)` | Returns a detailed message with a formatted stack trace. |
+| `obj.getLyngExceptionString(scope)` | Returns a formatted string including the class name, message, and primary throw site. |
+| `obj.getLyngExceptionStackTrace(scope)` | Returns the stack trace as an `ObjList` of `StackTraceEntry`. |
+| `obj.getLyngExceptionExtraData(scope)` | Returns the extra data associated with the exception. |
+| `obj.raiseAsExecutionError(scope?=null)` | Rethrows the object as a Kotlin `ExecutionError`. |
+
+#### Example: Serialization and Rethrowing
+
+You can serialize Lyng exception objects using `Lynon` to transmit them across boundaries and then rethrow them.
+
+```kotlin
+val session = EvalSession()
+val scope = session.getScope()
+
+try {
+    session.eval("throw MyUserException(404, \"Not Found\")")
+} catch (e: ExecutionError) {
+    // 1. Serialize the Lyng exception object
+    val encoded: UByteArray = lynonEncodeAny(scope, e.errorObject)
+
+    // ... (transmit 'encoded' byte array) ...
+
+    // 2. Deserialize it back to an Obj in a different context
+    val decoded: Obj = lynonDecodeAny(scope, encoded)
+
+    // 3. Properly rethrow it on the Kotlin side using the uniform API
+    decoded.raiseAsExecutionError(scope)
+}
+```
+
 ---
 
 That’s it. You now have Lyng embedded in your Kotlin app: you can expose your app’s API, evaluate user scripts, and organize your own packages to import from Lyng code.

docs/exceptions_handling.md

@@ -129,9 +129,11 @@ Serializable class that conveys information about the exception. Important membe
 | name              | description                                            |
 |-------------------|--------------------------------------------------------|
 | message           | String message                                           |
-| stackTrace        | lyng stack trace, list of `StackTraceEntry`, see below |
+| stackTrace()      | lyng stack trace, list of `StackTraceEntry`, see below   |
 | printStackTrace() | format and print stack trace using println()             |
 
+> **Note for Kotlin users**: When working with Lyng exceptions from Kotlin, you can use extension methods like `getLyngExceptionMessageWithStackTrace()`. See [Embedding Lyng](embedding.md#12-handling-and-serializing-exceptions) for the full API.
+
 ## StackTraceEntry
 
 A simple structire that stores single entry in Lyng stack, it is created automatically on exception creation:
@@ -150,24 +152,103 @@ class StackTraceEntry(
 
 # Custom error classes
 
-_this functionality is not yet released_
+You can define your own exception classes by inheriting from the built-in `Exception` class. This allows you to create specific error types for your application logic and catch them specifically.
+
+## Defining a custom exception
+
+To define a custom exception, create a class that inherits from `Exception`:
+
+```lyng
+class MyUserException : Exception("something went wrong")
+```
+
+You can also pass the message dynamically:
+
+```lyng
+class MyUserException(m) : Exception(m)
+
+throw MyUserException("custom error message")
+```
+
+If you don't provide a message to the `Exception` constructor, the class name will be used as the default message:
+
+```lyng
+class SimpleException : Exception
+
+val e = SimpleException()
+assertEquals("SimpleException", e.message)
+```
+
+## Throwing and catching custom exceptions
+
+Custom exceptions are thrown using the `throw` keyword and can be caught using `catch` blocks, just like standard exceptions:
+
+```lyng
+class ValidationException(m) : Exception(m)
+
+try {
+    throw ValidationException("Invalid input")
+}
+catch(e: ValidationException) {
+    println("Caught validation error: " + e.message)
+}
+catch(e: Exception) {
+    println("Caught other exception: " + e.message)
+}
+```
+
+Since user exceptions are real classes, inheritance works as expected:
+
+```lyng
+class BaseError : Exception
+class DerivedError : BaseError
+
+try {
+    throw DerivedError()
+}
+catch(e: BaseError) {
+    // This will catch DerivedError as well
+    assert(e is DerivedError)
+}
+```
+
+## Accessing extra data
+
+You can add your own fields to custom exception classes to carry additional information:
+
+```lyng
+class NetworkException(m, val statusCode) : Exception(m)
+
+try {
+    throw NetworkException("Not Found", 404)
+}
+catch(e: NetworkException) {
+    println("Error " + e.statusCode + ": " + e.message)
+}
+```
 
 # Standard exception classes
 
 | class                      | notes                                                 |
 |----------------------------|-------------------------------------------------------|
-| Exception                  | root of al throwable objects                          |
+| Exception                  | root of all throwable objects                         |
 | NullReferenceException     |                                                       |
 | AssertionFailedException   |                                                       | 
 | ClassCastException         |                                                       |
 | IndexOutOfBoundsException  |                                                       |
 | IllegalArgumentException   |                                                       | 
+| IllegalStateException      |                                                       |
+| NoSuchElementException     |                                                       |
 | IllegalAssignmentException | assigning to val, etc.                                |
 | SymbolNotDefinedException  |                                                       |
 | IterationEndException      | attempt to read iterator past end, `hasNext == false` |
 | IllegalAccessException     | attempt to access private members or like             |
-| UnknownException           | unexpected kotlin exception caught                    |
-|                            |                                                       |
+| UnknownException           | unexpected internal exception caught                  |
+| NotFoundException          |                                                       |
+| IllegalOperationException  |                                                       |
+| UnsetException             | access to uninitialized late-init val                 |
+| NotImplementedException    | used by `TODO()`                                      |
+| SyntaxError                |                                                       |
 
 
 ### Symbol resolution errors

docs/generics.md

@@ -0,0 +1,146 @@
+# Generics and type expressions
+
+This document covers generics, bounds, unions/intersections, and the rules for type expressions in Lyng.
+
+# Generic parameters
+
+Declare type parameters with `<...>` on functions and classes:
+
+    fun id<T>(x: T): T = x
+    class Box<T>(val value: T)
+
+Type arguments are usually inferred at call sites:
+
+    val b = Box(10)     // Box<Int>
+    val s = id("ok")    // T is String
+
+# Bounds
+
+Use `:` to set bounds. Bounds may be unions (`|`) or intersections (`&`):
+
+    fun sum<T: Int | Real>(x: T, y: T) = x + y
+    class Named<T: Iterable & Comparable>(val data: T)
+
+Bounds are checked at compile time. For union bounds, the argument must fit at least one option. For intersection bounds, it must fit all options.
+
+# Variance
+
+Generic types are invariant by default. You can specify declaration-site variance:
+
+    class Source<out T>(val value: T)
+    class Sink<in T> { fun accept(x: T) { ... } }
+
+`out` makes the type covariant (produced), `in` makes it contravariant (consumed).
+
+# Type aliases
+
+Type aliases name type expressions (including unions/intersections):
+
+    type Num = Int | Real
+    type AB = A & B
+
+Aliases can be generic and can use bounds and defaults:
+
+    type Maybe<T> = T?
+    type IntList<T: Int> = List<T>
+
+Aliases expand to their underlying type expressions. They can be used anywhere a type expression is expected.
+
+# Inference rules
+
+- Literals set obvious types (`1` is `Int`, `1.0` is `Real`, etc.).
+- Empty list literals default to `List<Object>` unless constrained by context.
+- Non-empty list literals infer element type as a union of element types.
+- Map literals infer key and value types; named keys are `String`.
+
+Examples:
+
+    val a = [1, 2, 3]             // List<Int>
+    val b = [1, "two", true]      // List<Int | String | Bool>
+    val c: List<Int> = []         // List<Int>
+
+    val m1 = { "a": 1, "b": 2 }   // Map<String, Int>
+    val m2 = { "a": 1, "b": "x" } // Map<String, Int | String>
+    val m3 = { ...m1, "c": true } // Map<String, Int | Bool>
+
+Map spreads carry key/value types when possible.
+
+Spreads propagate element type when possible:
+
+    val base = [1, 2]
+    val mix = [...base, 3]        // List<Int>
+
+# Type expressions
+
+Type expressions include simple types, generics, unions, and intersections:
+
+    Int
+    List<String>
+    Int | String
+    Iterable & Comparable
+
+These type expressions can appear in casts and `is` checks.
+
+# `is`, `in`, and `==` with type expressions
+
+There are two categories of `is` checks:
+
+1) Value checks: `x is T`
+   - `x` is a value, `T` is a type expression.
+   - This is a runtime instance check.
+
+2) Type checks: `T1 is T2`
+   - both sides are type expressions (class objects or unions/intersections).
+   - This is a *type-subset* check: every value of `T1` must fit in `T2`.
+
+Exact type expression equality uses `==` and is structural (union/intersection order does not matter).
+
+Includes checks use `in` with type expressions:
+
+    A in T
+
+This means `A` is a subset of `T` (the same relation as `A is T`).
+
+Examples (T = A | B):
+
+    T == A            // false
+    T is A            // false
+    A in T            // true
+    B in T            // true
+    T is A | B        // true
+
+# Nullability checks for types
+
+Use `is nullable` to check whether a type expression accepts `null`:
+
+    T is nullable
+    T !is nullable
+
+This works with concrete and generic types:
+
+    fun describe<T>(x: T): String = when (T) {
+        nullable -> "nullable"
+        else -> "non-null"
+    }
+
+Equivalent legacy form:
+
+    null is T
+
+# Practical examples
+
+    fun acceptInts<T: Int>(xs: List<T>) { }
+    acceptInts([1, 2, 3])
+    // acceptInts([1, "a"]) -> compile-time error
+
+    fun f<T>(list: List<T>) {
+        assert( T is Int | String | Bool )
+        assert( !(T is Int) )
+        assert( Int in T )
+    }
+    f([1, "two", true])
+
+# Notes
+
+- `T` is reified as a type expression when needed (e.g., union/intersection). When it is a single class, `T` is that class object.
+- Type expression checks are compile-time where possible; runtime checks only happen for `is` on values and explicit casts.

docs/idea_plugin.md

@@ -9,9 +9,12 @@ should be compatible with other IDEA flavors, notably [OpenIDE](https://openide.
 - reformat code (indents, spaces)
 - reformat on paste
 - smart enter key
+- `.lyng.d` definition files (merged into analysis for completion, navigation, Quick Docs, and error checking)
 
 Features are configurable via the plugin settings page, in system settings.
 
+See `docs/lyng_d_files.md` for `.lyng.d` syntax and examples.
+
 > Recommended for IntelliJ-based IDEs: While IntelliJ can import TextMate bundles
 > (Settings/Preferences → Editor → TextMate Bundles), the native Lyng plugin provides
 > better support (formatting, smart enter, background analysis, etc.). Prefer installing

docs/json_and_kotlin_serialization.md

@@ -20,7 +20,18 @@ Simple classes serialization is supported:
     assertEquals( "{\"foo\":1,\"bar\":2}", Point(1,2).toJsonString() )
     >>> void
 
-Note that mutable members are serialized:
+Note that mutable members are serialized by default. You can exclude any member (including constructor parameters) from JSON serialization using the `@Transient` attribute:
+
+    import lyng.serialization
+    
+    class Point2(@Transient val foo, val bar) {
+        @Transient var reason = 42
+        var visible = 100
+    }
+    assertEquals( "{\"bar\":2,\"visible\":100}", Point2(1,2).toJsonString() )
+    >>> void
+
+Note that if you override json serialization:
 
     import lyng.serialization
     
@@ -111,7 +122,8 @@ data class TestJson2(
 
 @Test
 fun deserializeMapWithJsonTest() = runTest {
-    val x = eval("""
+    val session = EvalSession()
+    val x = session.eval("""
         import lyng.serialization
         { value: 1, inner: { "foo": 1, "bar": 2 }}
     """.trimIndent()).decodeSerializable<TestJson2>()
@@ -132,7 +144,8 @@ data class TestJson3(
 )
 @Test
 fun deserializeAnyMapWithJsonTest() = runTest {
-    val x = eval("""
+    val session = EvalSession()
+    val x = session.eval("""
         import lyng.serialization
         { value: 12, inner: { "foo": 1, "bar": "two" }}
     """.trimIndent()).decodeSerializable<TestJson3>()
@@ -164,4 +177,3 @@ on [Instant](time.md), see `Instant.truncateTo...` functions.
 
 (3)
 : Map keys must be strings, map values may be any objects serializable to Json.
-

docs/lyng.io.console.md

@@ -0,0 +1,113 @@
+### lyng.io.console
+
+`lyng.io.console` provides optional rich console support for terminal applications.
+
+> **Note:** this module is part of `lyngio`. It must be explicitly installed into the import manager by host code.
+>
+> **CLI note:** the `lyng` CLI now installs `lyng.io.console` in its base scope by default, so scripts can simply `import lyng.io.console`.
+
+#### Install in host
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.io.console.createConsoleModule
+import net.sergeych.lyngio.console.security.PermitAllConsoleAccessPolicy
+
+suspend fun initScope() {
+    val session = EvalSession()
+    val scope = session.getScope()
+    createConsoleModule(PermitAllConsoleAccessPolicy, scope)
+}
+```
+
+#### Use in Lyng script
+
+```lyng
+import lyng.io.console
+
+println("supported = " + Console.isSupported())
+println("tty = " + Console.isTty())
+println("ansi = " + Console.ansiLevel())
+println("geometry = " + Console.geometry())
+
+Console.write("hello\n")
+Console.home()
+Console.clear()
+Console.moveTo(1, 1)
+Console.clearLine()
+Console.enterAltScreen()
+Console.leaveAltScreen()
+Console.setCursorVisible(true)
+Console.flush()
+```
+
+#### Tetris sample
+
+The repository includes a full interactive Tetris sample that demonstrates:
+
+- alternate screen rendering
+- raw keyboard input
+- resize handling
+- typed console events
+
+![Lyng Tetris sample](/tetris.png)
+
+Run it from the project root in a real TTY:
+
+```bash
+lyng examples/tetris_console.lyng
+```
+
+#### API
+
+- `Console.isSupported(): Bool` — whether console control is available on this platform/runtime.
+- `Console.isTty(): Bool` — whether output is attached to a TTY.
+- `Console.ansiLevel(): ConsoleAnsiLevel` — `NONE`, `BASIC16`, `ANSI256`, `TRUECOLOR`.
+- `Console.geometry(): ConsoleGeometry?` — `{columns, rows}` as typed object or `null`.
+- `Console.details(): ConsoleDetails` — consolidated capability object.
+- `Console.write(text: String)` — writes to console output.
+- `Console.flush()` — flushes buffered output.
+- `Console.home()` — moves cursor to top-left.
+- `Console.clear()` — clears visible screen.
+- `Console.moveTo(row: Int, column: Int)` — moves cursor to 1-based row/column.
+- `Console.clearLine()` — clears current line.
+- `Console.enterAltScreen()` — switch to alternate screen buffer.
+- `Console.leaveAltScreen()` — return to normal screen buffer.
+- `Console.setCursorVisible(visible: Bool)` — shows/hides cursor.
+- `Console.events(): ConsoleEventStream` — endless iterable source of typed events: `ConsoleResizeEvent`, `ConsoleKeyEvent`.
+- `Console.setRawMode(enabled: Bool): Bool` — requests raw input mode, returns `true` if changed.
+
+#### Event Iteration
+
+Use events from a loop, typically in a separate coroutine:
+
+```lyng
+launch {
+    for (ev in Console.events()) {
+        if (ev is ConsoleKeyEvent) {
+            // handle key
+        }
+    }
+}
+```
+
+#### Event format
+
+`Console.events()` emits `ConsoleEvent` with:
+
+- `type: ConsoleEventType` — `UNKNOWN`, `RESIZE`, `KEY_DOWN`, `KEY_UP`
+
+Additional fields:
+
+- `ConsoleResizeEvent`: `columns`, `rows`
+- `ConsoleKeyEvent`: `key`, `code`, `ctrl`, `alt`, `shift`, `meta`
+
+#### Security policy
+
+The module uses `ConsoleAccessPolicy` with operations:
+
+- `WriteText(length)`
+- `ReadEvents`
+- `SetRawMode(enabled)`
+
+For permissive mode, use `PermitAllConsoleAccessPolicy`.

docs/lyng.io.db.md

@@ -0,0 +1,401 @@
+### lyng.io.db — SQL database access for Lyng scripts
+
+This module provides the portable SQL database contract for Lyng. The current shipped providers are SQLite via `lyng.io.db.sqlite` and a JVM-only JDBC bridge via `lyng.io.db.jdbc`.
+
+> **Note:** `lyngio` is a separate library module. It must be explicitly added as a dependency to your host application and initialized in your Lyng scopes.
+
+---
+
+#### Install the module into a Lyng session
+
+For SQLite-backed database access, install both the generic DB module and the SQLite provider:
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.io.db.createDbModule
+import net.sergeych.lyng.io.db.sqlite.createSqliteModule
+
+suspend fun bootstrapDb() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
+    createDbModule(scope)
+    createSqliteModule(scope)
+    session.eval("""
+        import lyng.io.db
+        import lyng.io.db.sqlite
+    """.trimIndent())
+}
+```
+
+`createSqliteModule(...)` also registers the `sqlite:` scheme for generic `openDatabase(...)`.
+
+For JVM JDBC-backed access, install the JDBC provider as well:
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.io.db.createDbModule
+import net.sergeych.lyng.io.db.jdbc.createJdbcModule
+
+suspend fun bootstrapJdbc() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
+    createDbModule(scope)
+    createJdbcModule(scope)
+    session.eval("""
+        import lyng.io.db
+        import lyng.io.db.jdbc
+    """.trimIndent())
+}
+```
+
+`createJdbcModule(...)` registers `jdbc:`, `h2:`, `postgres:`, and `postgresql:` for `openDatabase(...)`.
+
+---
+
+#### Using from Lyng scripts
+
+Typed SQLite open helper:
+
+```lyng
+import lyng.io.db.sqlite
+
+val db = openSqlite(":memory:")
+
+val userCount = db.transaction { tx ->
+    tx.execute("create table user(id integer primary key autoincrement, name text not null)")
+    tx.execute("insert into user(name) values(?)", "Ada")
+    tx.execute("insert into user(name) values(?)", "Linus")
+    tx.select("select count(*) as count from user").toList()[0]["count"]
+}
+
+assertEquals(2, userCount)
+```
+
+Generic provider-based open:
+
+```lyng
+import lyng.io.db
+import lyng.io.db.sqlite
+
+val db = openDatabase(
+    "sqlite:./app.db",
+    Map(
+        "foreignKeys" => true,
+        "busyTimeoutMillis" => 5000
+    )
+)
+```
+
+JVM JDBC open with H2:
+
+```lyng
+import lyng.io.db.jdbc
+
+val db = openH2("mem:demo;DB_CLOSE_DELAY=-1")
+
+val names = db.transaction { tx ->
+    tx.execute("create table person(id bigint auto_increment primary key, name varchar(120) not null)")
+    tx.execute("insert into person(name) values(?)", "Ada")
+    tx.execute("insert into person(name) values(?)", "Linus")
+    tx.select("select name from person order by id").toList()
+}
+
+assertEquals("Ada", names[0]["name"])
+assertEquals("Linus", names[1]["name"])
+```
+
+Generic JDBC open through `openDatabase(...)`:
+
+```lyng
+import lyng.io.db
+import lyng.io.db.jdbc
+
+val db = openDatabase(
+    "jdbc:h2:mem:demo2;DB_CLOSE_DELAY=-1",
+    Map()
+)
+
+val answer = db.transaction { tx ->
+    tx.select("select 42 as answer").toList()[0]["answer"]
+}
+
+assertEquals(42, answer)
+```
+
+PostgreSQL typed open:
+
+```lyng
+import lyng.io.db.jdbc
+
+val db = openPostgres(
+    "jdbc:postgresql://127.0.0.1/appdb",
+    "appuser",
+    "secret"
+)
+
+val titles = db.transaction { tx ->
+    tx.execute("create table if not exists task(id bigserial primary key, title text not null)")
+    tx.execute("insert into task(title) values(?)", "Ship JDBC provider")
+    tx.execute("insert into task(title) values(?)", "Test PostgreSQL path")
+    tx.select("select title from task order by id").toList()
+}
+
+assertEquals("Ship JDBC provider", titles[0]["title"])
+```
+
+Nested transactions use real savepoint semantics:
+
+```lyng
+import lyng.io.db
+import lyng.io.db.sqlite
+
+val db = openSqlite(":memory:")
+
+db.transaction { tx ->
+    tx.execute("create table item(id integer primary key autoincrement, name text not null)")
+    tx.execute("insert into item(name) values(?)", "outer")
+
+    try {
+        tx.transaction { inner ->
+            inner.execute("insert into item(name) values(?)", "inner")
+            throw IllegalStateException("rollback nested")
+        }
+    } catch (_: IllegalStateException) {
+    }
+
+    assertEquals(1, tx.select("select count(*) as count from item").toList()[0]["count"])
+}
+```
+
+Intentional rollback without treating it as a backend failure:
+
+```lyng
+import lyng.io.db
+import lyng.io.db.sqlite
+
+val db = openSqlite(":memory:")
+
+assertThrows(RollbackException) {
+    db.transaction { tx ->
+        tx.execute("create table item(id integer primary key autoincrement, name text not null)")
+        tx.execute("insert into item(name) values(?)", "temporary")
+        throw RollbackException("stop here")
+    }
+}
+```
+
+---
+
+#### Portable API
+
+##### `Database`
+
+- `transaction(block)` — opens a transaction, commits on normal exit, rolls back on uncaught failure.
+
+##### `SqlTransaction`
+
+- `select(clause, params...)` — execute a statement whose primary result is a row set.
+- `execute(clause, params...)` — execute a side-effect statement and return `ExecutionResult`.
+- `transaction(block)` — nested transaction with real savepoint semantics.
+
+##### `ResultSet`
+
+- `columns` — positional `SqlColumn` metadata, available before iteration.
+- `size()` — result row count.
+- `isEmpty()` — fast emptiness check where possible.
+- `iterator()` — normal row iteration while the transaction is active.
+- `toList()` — materialize detached `SqlRow` snapshots that may be used after the transaction ends.
+
+##### `SqlRow`
+
+- `row[index]` — zero-based positional access.
+- `row["columnName"]` — case-insensitive lookup by output column label.
+
+Name-based access fails with `SqlUsageException` if the name is missing or ambiguous.
+
+##### `ExecutionResult`
+
+- `affectedRowsCount`
+- `getGeneratedKeys()`
+
+Statements that return rows directly, such as `... returning ...`, should use `select(...)`, not `execute(...)`.
+
+---
+
+#### Value mapping
+
+Portable bind values:
+
+- `null`
+- `Bool`
+- `Int`, `Double`, `Decimal`
+- `String`
+- `Buffer`
+- `Date`, `DateTime`, `Instant`
+
+Unsupported parameter values fail with `SqlUsageException`.
+
+Portable result metadata categories:
+
+- `Binary`
+- `String`
+- `Int`
+- `Double`
+- `Decimal`
+- `Bool`
+- `Date`
+- `DateTime`
+- `Instant`
+
+For temporal types, see [time functions](time.md).
+
+---
+
+#### SQLite provider
+
+`lyng.io.db.sqlite` currently provides the first concrete backend.
+
+Typed helper:
+
+```lyng
+openSqlite(
+    path: String,
+    readOnly: Bool = false,
+    createIfMissing: Bool = true,
+    foreignKeys: Bool = true,
+    busyTimeoutMillis: Int = 5000
+): Database
+```
+
+Accepted generic URL forms:
+
+- `sqlite::memory:`
+- `sqlite:relative/path.db`
+- `sqlite:/absolute/path.db`
+
+Supported `openDatabase(..., extraParams)` keys for SQLite:
+
+- `readOnly: Bool`
+- `createIfMissing: Bool`
+- `foreignKeys: Bool`
+- `busyTimeoutMillis: Int`
+
+SQLite write/read policy in v1:
+
+- `Bool` writes as `0` / `1`
+- `Decimal` writes as canonical text
+- `Date` writes as `YYYY-MM-DD`
+- `DateTime` writes as ISO local timestamp text without timezone
+- `Instant` writes as ISO UTC timestamp text with explicit timezone marker
+- `TIME*` values stay `String`
+- `TIMESTAMP` / `DATETIME` reject timezone-bearing stored text
+
+Open-time validation failures:
+
+- malformed URL or bad option shape -> `IllegalArgumentException`
+- runtime open failure -> `DatabaseException`
+
+#### JDBC provider
+
+`lyng.io.db.jdbc` is currently implemented on the JVM target only. The `lyngio-jvm` artifact bundles and explicitly loads these JDBC drivers:
+
+- SQLite
+- H2
+- PostgreSQL
+
+Typed helpers:
+
+```lyng
+openJdbc(
+    connectionUrl: String,
+    user: String? = null,
+    password: String? = null,
+    driverClass: String? = null,
+    properties: Map<String, Object?>? = null
+): Database
+
+openH2(
+    connectionUrl: String,
+    user: String? = null,
+    password: String? = null,
+    properties: Map<String, Object?>? = null
+): Database
+
+openPostgres(
+    connectionUrl: String,
+    user: String? = null,
+    password: String? = null,
+    properties: Map<String, Object?>? = null
+): Database
+```
+
+Accepted generic URL forms:
+
+- `jdbc:h2:mem:test;DB_CLOSE_DELAY=-1`
+- `h2:mem:test;DB_CLOSE_DELAY=-1`
+- `jdbc:postgresql://localhost/app`
+- `postgres://localhost/app`
+- `postgresql://localhost/app`
+
+Supported `openDatabase(..., extraParams)` keys for JDBC:
+
+- `driverClass: String`
+- `user: String`
+- `password: String`
+- `properties: Map<String, Object?>`
+
+Behavior notes for the JDBC bridge:
+
+- the portable `Database` / `SqlTransaction` API stays the same as for SQLite
+- nested transactions use JDBC savepoints
+- JDBC connection properties are built from `user`, `password`, and `properties`
+- `properties` values are stringified before being passed to JDBC
+- statements with row-returning clauses still must use `select(...)`, not `execute(...)`
+
+Platform support for this provider:
+
+- `lyng.io.db.jdbc` — JVM only
+- `openH2(...)` — works out of the box with `lyngio-jvm`
+- `openPostgres(...)` — driver included, but an actual PostgreSQL server is still required
+
+PostgreSQL-specific notes:
+
+- `openPostgres(...)` accepts either a full JDBC URL or shorthand forms such as `//localhost/app`
+- local peer/trust setups may use an empty password string
+- generated keys work with PostgreSQL `bigserial` / identity columns through `ExecutionResult.getGeneratedKeys()`
+- for reproducible automated tests, prefer a disposable PostgreSQL instance such as Docker/Testcontainers instead of a long-lived shared server
+
+---
+
+#### Lifetime rules
+
+`ResultSet` is valid only while its owning transaction is active.
+
+`SqlRow` values are detached snapshots once materialized, so this pattern is valid:
+
+```lyng
+val rows = db.transaction { tx ->
+    tx.select("select name from person order by id").toList()
+}
+
+assertEquals("Ada", rows[0]["name"])
+```
+
+This means:
+
+- do not keep `ResultSet` objects after the transaction block returns
+- materialize rows with `toList()` inside the transaction when they must outlive it
+
+The same rule applies to generated keys from `ExecutionResult.getGeneratedKeys()`: the `ResultSet` is transaction-scoped, but rows returned by `toList()` are detached.
+
+---
+
+#### Platform support
+
+- `lyng.io.db` — generic contract, available when host code installs it
+- `lyng.io.db.sqlite` — implemented on JVM and Linux Native in the current release tree
+- `lyng.io.db.jdbc` — implemented on JVM in the current release tree
+
+For the broader I/O overview, see [lyngio overview](lyngio.md).

docs/lyng.io.fs.md

@@ -8,7 +8,7 @@ This module provides a uniform, suspend-first filesystem API to Lyng scripts, ba
 
 It exposes a Lyng class `Path` with methods for file and directory operations, including streaming readers for large files.
 
-It is a separate library because access to teh filesystem is a security risk we compensate with a separate API that user must explicitly include to the dependency and allow. Together with `FsAceessPolicy` that is required to `createFs()` which actually adds the filesystem to the scope, the security risk is isolated.
+It is a separate library because access to the filesystem is a security risk we compensate with a separate API that user must explicitly include to the dependency and allow. Together with `FsAccessPolicy` that is required to `createFs()` which actually adds the filesystem to the scope, the security risk is isolated.
 
 Also, it helps keep Lyng core small and focused.
 
@@ -23,7 +23,7 @@ dependencies {
     implementation("net.sergeych:lyngio:0.0.1-SNAPSHOT")
 }
 ```
-Note on maven repository. Lyngio uses ths same maven as Lyng code (`lynglib`) so it is most likely already in your project. If ont, add it to the proper section of your `build.gradle.kts` or settings.gradle.kts:
+Note on maven repository. Lyngio uses the same maven as Lyng code (`lynglib`) so it is most likely already in your project. If not, add it to the proper section of your `build.gradle.kts` or settings.gradle.kts:
 
 ```kotlin
     repositories {
@@ -39,19 +39,27 @@ This brings in:
 
 ---
 
-#### Install the module into a Lyng Scope
+#### Install the module into a Lyng session
 
-The filesystem module is not installed automatically. You must explicitly register it in the scope’s `ImportManager` using the installer. You can customize access control via `FsAccessPolicy`.
+The filesystem module is not installed automatically. The preferred host runtime is `EvalSession`: create the session, get its underlying scope, install the module there, and execute scripts through the session. You can customize access control via `FsAccessPolicy`.
 
-Kotlin (host) bootstrap example (imports omitted for brevity):
+Kotlin (host) bootstrap example:
 
 ```kotlin
-val scope: Scope = Scope.new()
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.io.fs.createFs
+import net.sergeych.lyngio.fs.security.PermitAllAccessPolicy
+
+suspend fun bootstrapFs() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
     val installed: Boolean = createFs(PermitAllAccessPolicy, scope)
     // installed == true on first registration in this ImportManager, false on repeats
 
-// In scripts (or via scope.eval), import the module to use its symbols:
-scope.eval("import lyng.io.fs")
+    // In scripts (or via session.eval), import the module to use its symbols:
+    session.eval("import lyng.io.fs")
+}
 ```
 
 You can install with a custom policy too (see Access policy below).
@@ -181,7 +189,7 @@ val denyWrites = object : FsAccessPolicy {
 }
 
 createFs(denyWrites, scope)
-scope.eval("import lyng.io.fs")
+session.eval("import lyng.io.fs")
 ```
 
 Composite operations like `copy` and `move` are checked as a set of primitives (e.g., `OpenRead(src)` + `Delete(dst)` if overwriting + `CreateFile(dst)` + `OpenWrite(dst)`).

docs/lyng.io.http.md

@@ -0,0 +1,179 @@
+### lyng.io.http — HTTP/HTTPS client for Lyng scripts
+
+This module provides a compact HTTP client API for Lyng scripts. It is implemented in `lyngio` and backed by Ktor on supported runtimes.
+
+> **Note:** `lyngio` is a separate library module. It must be explicitly added as a dependency to your host application and initialized in your Lyng scopes.
+
+---
+
+#### Add the library to your project (Gradle)
+
+If you use this repository as a multi-module project, add a dependency on `:lyngio`:
+
+```kotlin
+dependencies {
+    implementation("net.sergeych:lyngio:0.0.1-SNAPSHOT")
+}
+```
+
+For external projects, ensure you also use the Lyng Maven repository described in `lyng.io.fs`.
+
+---
+
+#### Install the module into a Lyng session
+
+The HTTP module is not installed automatically. Install it into the session scope and provide a policy.
+
+Kotlin (host) bootstrap example:
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.io.http.createHttpModule
+import net.sergeych.lyngio.http.security.PermitAllHttpAccessPolicy
+
+suspend fun bootstrapHttp() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
+    createHttpModule(PermitAllHttpAccessPolicy, scope)
+    session.eval("import lyng.io.http")
+}
+```
+
+---
+
+#### Using from Lyng scripts
+
+Simple GET:
+
+    import lyng.io.http
+
+    val r = Http.get(HTTP_TEST_URL + "/hello")
+    [r.status, r.text()]
+    >>> [200,hello from test]
+
+Headers and response header access:
+
+    import lyng.io.http
+
+    val r = Http.get(HTTP_TEST_URL + "/headers")
+    [r.headers["X-Reply"], r.headers.getAll("X-Reply").size, r.text()]
+    >>> [one,2,header demo]
+
+Programmatic request object:
+
+    import lyng.io.http
+
+    val q = HttpRequest()
+    q.method = "POST"
+    q.url = HTTP_TEST_URL + "/echo"
+    q.headers = Map("Content-Type" => "text/plain")
+    q.bodyText = "ping"
+
+    val r = Http.request(q)
+    r.text()
+    >>> "POST:ping"
+
+HTTPS GET:
+
+    import lyng.io.http
+
+    val r = Http.get(HTTPS_TEST_URL + "/hello")
+    [r.status, r.text()]
+    >>> [200,hello from test]
+
+---
+
+#### API reference
+
+##### `Http` (static methods)
+
+- `isSupported(): Bool` — Whether HTTP client support is available on the current runtime.
+- `request(req: HttpRequest): HttpResponse` — Execute a request described by a mutable request object.
+- `get(url: String, headers...): HttpResponse` — Convenience GET request.
+- `post(url: String, bodyText: String = "", contentType: String? = null, headers...): HttpResponse` — Convenience text POST request.
+- `postBytes(url: String, body: Buffer, contentType: String? = null, headers...): HttpResponse` — Convenience binary POST request.
+
+For convenience methods, `headers...` accepts:
+
+- `MapEntry`, e.g. `"Accept" => "text/plain"`
+- 2-item lists, e.g. `["Accept", "text/plain"]`
+
+##### `HttpRequest`
+
+- `method: String`
+- `url: String`
+- `headers: Map<String, String>`
+- `bodyText: String?`
+- `bodyBytes: Buffer?`
+- `timeoutMillis: Int?`
+
+Only one of `bodyText` and `bodyBytes` should be set.
+
+##### `HttpResponse`
+
+- `status: Int`
+- `statusText: String`
+- `headers: HttpHeaders`
+- `text(): String`
+- `bytes(): Buffer`
+
+Response body decoding is cached inside the response object.
+
+##### `HttpHeaders`
+
+`HttpHeaders` behaves like `Map<String, String>` for the first value of each header name and additionally exposes:
+
+- `get(name: String): String?`
+- `getAll(name: String): List<String>`
+- `names(): List<String>`
+
+Header lookup is case-insensitive.
+
+---
+
+#### Security policy
+
+The module uses `HttpAccessPolicy` to authorize requests before they are sent.
+
+- `HttpAccessPolicy` — interface for custom policies
+- `PermitAllHttpAccessPolicy` — allows all requests
+- `HttpAccessOp.Request(method, url)` — operation checked by the policy
+
+Example restricted policy in Kotlin:
+
+```kotlin
+import net.sergeych.lyngio.fs.security.AccessContext
+import net.sergeych.lyngio.fs.security.AccessDecision
+import net.sergeych.lyngio.fs.security.Decision
+import net.sergeych.lyngio.http.security.HttpAccessOp
+import net.sergeych.lyngio.http.security.HttpAccessPolicy
+
+val allowLocalOnly = object : HttpAccessPolicy {
+    override suspend fun check(op: HttpAccessOp, ctx: AccessContext): AccessDecision =
+        when (op) {
+            is HttpAccessOp.Request ->
+                if (
+                    op.url.startsWith("http://127.0.0.1:") ||
+                    op.url.startsWith("https://127.0.0.1:") ||
+                    op.url.startsWith("http://localhost:") ||
+                    op.url.startsWith("https://localhost:")
+                )
+                    AccessDecision(Decision.Allow)
+                else
+                    AccessDecision(Decision.Deny, "only local HTTP/HTTPS requests are allowed")
+        }
+}
+```
+
+---
+
+#### Platform support
+
+- **JVM:** supported
+- **Android:** supported via the Ktor CIO client backend
+- **JS:** supported via the Ktor JS client backend
+- **Linux native:** supported via the Ktor Curl client backend
+- **Windows native:** supported via the Ktor WinHttp client backend
+- **Apple native:** supported via the Ktor Darwin client backend
+- **Other targets:** may report unsupported; use `Http.isSupported()` before relying on it

docs/lyng.io.net.md

@@ -0,0 +1,175 @@
+### lyng.io.net — TCP and UDP sockets for Lyng scripts
+
+This module provides minimal raw transport networking for Lyng scripts. It is implemented in `lyngio` and backed by Ktor sockets on the JVM and Linux Native, and by Node networking APIs on JS/Node runtimes.
+
+> **Note:** `lyngio` is a separate library module. It must be explicitly added as a dependency to your host application and initialized in your Lyng scopes.
+>
+> **Important native platform limit:** current native TCP/UDP support is backed by a selector with a per-process file descriptor ceiling. On Linux/macOS native targets this makes high-connection-count servers and same-process load tests unsuitable once the process approaches that limit.
+>
+> **Recommendation:** for serious HTTP/TCP servers, prefer the JVM target today. On native targets, keep concurrency bounded, batch local load tests in waves, and use multiple worker processes behind a reverse proxy if you need more throughput before the backend is reworked.
+>
+> **Need this fixed?** Please open or upvote an issue at <https://github.com/sergeych/lyng/issues> so native high-concurrency networking can be prioritized.
+
+---
+
+#### Install the module into a Lyng session
+
+Kotlin (host) bootstrap example:
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.io.net.createNetModule
+import net.sergeych.lyngio.net.security.PermitAllNetAccessPolicy
+
+suspend fun bootstrapNet() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
+    createNetModule(PermitAllNetAccessPolicy, scope)
+    session.eval("import lyng.io.net")
+}
+```
+
+---
+
+#### Using from Lyng scripts
+
+Capability checks and address resolution:
+
+    import lyng.io.net
+
+    val a: SocketAddress = Net.resolve("127.0.0.1", 4040)[0]
+    [Net.isSupported(), a.toString(), a.resolved, a.ipVersion == IpVersion.IPV4]
+    >>> [true,127.0.0.1:4040,true,true]
+
+TCP client connect, write, read, and close:
+
+    import lyng.buffer
+    import lyng.io.net
+
+    val socket = Net.tcpConnect("127.0.0.1", NET_TEST_TCP_PORT)
+    socket.writeUtf8("ping")
+    socket.flush()
+    val reply = (socket.read(16) as Buffer).decodeUtf8()
+    socket.close()
+    reply
+    >>> "reply:ping"
+
+Lyng TCP server socket operations with `tcpListen()` and `accept()`:
+
+    import lyng.buffer
+    import lyng.io.net
+
+    val server = Net.tcpListen(0, "127.0.0.1")
+    val port = server.localAddress().port
+    val accepted = launch {
+        val client = server.accept()
+        val line = (client.read(4) as Buffer).decodeUtf8()
+        client.writeUtf8("echo:" + line)
+        client.flush()
+        client.close()
+        server.close()
+        line
+    }
+
+    val socket = Net.tcpConnect("127.0.0.1", port)
+    socket.writeUtf8("ping")
+    socket.flush()
+    val reply = (socket.read(16) as Buffer).decodeUtf8()
+    socket.close()
+    [accepted.await(), reply]
+    >>> [ping,echo:ping]
+
+UDP bind, send, receive, and inspect sender address:
+
+    import lyng.buffer
+    import lyng.io.net
+
+    val server = Net.udpBind(0, "127.0.0.1")
+    val client = Net.udpBind(0, "127.0.0.1")
+    client.send(Buffer("ping"), "127.0.0.1", server.localAddress().port)
+    val d = server.receive()
+    client.close()
+    server.close()
+    [d.data.decodeUtf8(), d.address.port > 0]
+    >>> [ping,true]
+
+---
+
+#### API reference
+
+##### `Net` (static methods)
+
+- `isSupported(): Bool` — Whether any raw networking support is available.
+- `isTcpAvailable(): Bool` — Whether outbound TCP sockets are available.
+- `isTcpServerAvailable(): Bool` — Whether listening TCP server sockets are available.
+- `isUdpAvailable(): Bool` — Whether UDP datagram sockets are available.
+- `resolve(host: String, port: Int): List<SocketAddress>` — Resolve a host and port into concrete addresses.
+- `tcpConnect(host: String, port: Int, timeoutMillis: Int? = null, noDelay: Bool = true): TcpSocket` — Open an outbound TCP socket.
+- `tcpListen(port: Int, host: String? = null, backlog: Int = 128, reuseAddress: Bool = true): TcpServer` — Start a listening TCP server socket.
+- `udpBind(port: Int = 0, host: String? = null, reuseAddress: Bool = true): UdpSocket` — Bind a UDP socket.
+
+##### `SocketAddress`
+
+- `host: String`
+- `port: Int`
+- `ipVersion: IpVersion`
+- `resolved: Bool`
+- `toString(): String`
+
+##### `TcpSocket`
+
+- `isOpen(): Bool`
+- `localAddress(): SocketAddress`
+- `remoteAddress(): SocketAddress`
+- `read(maxBytes: Int = 65536): Buffer?`
+- `readLine(): String?`
+- `write(data: Buffer): void`
+- `writeUtf8(text: String): void`
+- `flush(): void`
+- `close(): void`
+
+##### `TcpServer`
+
+- `isOpen(): Bool`
+- `localAddress(): SocketAddress`
+- `accept(): TcpSocket`
+- `close(): void`
+
+##### `UdpSocket`
+
+- `isOpen(): Bool`
+- `localAddress(): SocketAddress`
+- `receive(maxBytes: Int = 65536): Datagram?`
+- `send(data: Buffer, host: String, port: Int): void`
+- `close(): void`
+
+##### `Datagram`
+
+- `data: Buffer`
+- `address: SocketAddress`
+
+---
+
+#### Security policy
+
+The module uses `NetAccessPolicy` to authorize network operations before they are executed.
+
+- `NetAccessPolicy` — interface for custom policies
+- `PermitAllNetAccessPolicy` — allows all network operations
+- `NetAccessOp.Resolve(host, port)`
+- `NetAccessOp.TcpConnect(host, port)`
+- `NetAccessOp.TcpListen(host, port, backlog)`
+- `NetAccessOp.UdpBind(host, port)`
+
+---
+
+#### Platform support
+
+- **JVM:** supported
+- **Android:** supported via the Ktor CIO and Ktor sockets backends
+- **JS/Node:** supported for `resolve`, TCP client/server, and UDP
+- **JS/browser:** unsupported; capability checks report unavailable
+- **Linux Native:** supported via Ktor sockets
+- **Apple Native:** enabled via the shared native Ktor sockets backend; compile-verified, runtime not yet host-verified
+- **Other native targets:** currently report unsupported; use capability checks before relying on raw sockets

docs/lyng.io.process.md

@@ -0,0 +1,138 @@
+### lyng.io.process — Process execution and control for Lyng scripts
+
+This module provides a way to run external processes and shell commands from Lyng scripts. It is designed to be multiplatform and uses coroutines for non-blocking execution.
+
+> **Note:** `lyngio` is a separate library module. It must be explicitly added as a dependency to your host application and initialized in your Lyng scopes.
+
+---
+
+#### Add the library to your project (Gradle)
+
+If you use this repository as a multi-module project, add a dependency on `:lyngio`:
+
+```kotlin
+dependencies {
+    implementation("net.sergeych:lyngio:0.0.1-SNAPSHOT")
+}
+```
+
+For external projects, ensure you have the appropriate Maven repository configured (see `lyng.io.fs` documentation).
+
+---
+
+#### Install the module into a Lyng session
+
+The process module is not installed automatically. The preferred host runtime is `EvalSession`: create the session, get its underlying scope, install the module there, and execute scripts through the session. You can customize access control via `ProcessAccessPolicy`.
+
+Kotlin (host) bootstrap example:
+
+```kotlin
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.io.process.createProcessModule
+import net.sergeych.lyngio.process.security.PermitAllProcessAccessPolicy
+
+suspend fun bootstrapProcess() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
+    createProcessModule(PermitAllProcessAccessPolicy, scope)
+
+    // In scripts (or via session.eval), import the module:
+    session.eval("import lyng.io.process")
+}
+```
+
+---
+
+#### Using from Lyng scripts
+
+```lyng
+import lyng.io.process
+
+// Execute a process with arguments
+val p = Process.execute("ls", ["-l", "/tmp"])
+for (line in p.stdout) {
+    println("OUT: " + line)
+}
+val exitCode = p.waitFor()
+println("Process exited with: " + exitCode)
+
+// Run a shell command
+val sh = Process.shell("echo 'Hello from shell' | wc -w")
+for (line in sh.stdout) {
+    println("Word count: " + line.trim())
+}
+
+// Platform information
+val details = Platform.details()
+println("OS: " + details.name + " " + details.version + " (" + details.arch + ")")
+if (details.kernelVersion != null) {
+    println("Kernel: " + details.kernelVersion)
+}
+
+if (Platform.isSupported()) {
+    println("Processes are supported!")
+}
+```
+
+---
+
+#### API Reference
+
+##### `Process` (static methods)
+- `execute(executable: String, args: List<String>): RunningProcess` — Start an external process.
+- `shell(command: String): RunningProcess` — Run a command through the system shell (e.g., `/bin/sh` or `cmd.exe`).
+
+##### `RunningProcess` (instance methods)
+- `stdout: Flow` — Standard output stream as a Lyng Flow of lines.
+- `stderr: Flow` — Standard error stream as a Lyng Flow of lines.
+- `waitFor(): Int` — Wait for the process to exit and return the exit code.
+- `signal(name: String)` — Send a signal to the process (e.g., `"SIGINT"`, `"SIGTERM"`, `"SIGKILL"`).
+- `destroy()` — Forcefully terminate the process.
+
+##### `Platform` (static methods)
+- `details(): Map` — Get platform details. Returned map keys: `name`, `version`, `arch`, `kernelVersion`.
+- `isSupported(): Bool` — True if process execution is supported on the current platform.
+
+---
+
+#### Security Policy
+
+Process execution is a sensitive operation. `lyngio` uses `ProcessAccessPolicy` to control access to `execute` and `shell` operations.
+
+- `ProcessAccessPolicy` — Interface for custom policies.
+- `PermitAllProcessAccessPolicy` — Allows all operations.
+- `ProcessAccessOp` (sealed) — Operations to check:
+    - `Execute(executable, args)`
+    - `Shell(command)`
+
+Example of a restricted policy in Kotlin:
+
+```kotlin
+import net.sergeych.lyngio.fs.security.AccessDecision
+import net.sergeych.lyngio.fs.security.Decision
+import net.sergeych.lyngio.process.security.ProcessAccessOp
+import net.sergeych.lyngio.process.security.ProcessAccessPolicy
+
+val restrictedPolicy = object : ProcessAccessPolicy {
+    override suspend fun check(op: ProcessAccessOp, ctx: AccessContext): AccessDecision {
+        return when (op) {
+            is ProcessAccessOp.Execute -> {
+                if (op.executable == "ls") AccessDecision(Decision.Allow)
+                else AccessDecision(Decision.Deny, "Only 'ls' is allowed")
+            }
+            is ProcessAccessOp.Shell -> AccessDecision(Decision.Deny, "Shell is forbidden")
+        }
+    }
+}
+createProcessModule(restrictedPolicy, scope)
+```
+
+---
+
+#### Platform Support
+
+- **JVM:** Full support using `ProcessBuilder`.
+- **Native (Linux/macOS):** Support via POSIX.
+- **Windows:** Support planned.
+- **Android/JS/iOS/Wasm:** Currently not supported; `isSupported()` returns `false` and attempts to run processes will throw `UnsupportedOperationException`.

docs/lyng.io.ws.md

@@ -0,0 +1,148 @@
+### lyng.io.ws — WebSocket client for Lyng scripts
+
+This module provides a compact WebSocket client API for Lyng scripts. It is implemented in `lyngio` and currently backed by Ktor WebSockets on the JVM.
+
+> **Note:** `lyngio` is a separate library module. It must be explicitly added as a dependency to your host application and initialized in your Lyng scopes.
+
+---
+
+#### Install the module into a Lyng session
+
+Kotlin (host) bootstrap example:
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.Scope
+import net.sergeych.lyng.io.ws.createWsModule
+import net.sergeych.lyngio.ws.security.PermitAllWsAccessPolicy
+
+suspend fun bootstrapWs() {
+    val session = EvalSession()
+    val scope: Scope = session.getScope()
+    createWsModule(PermitAllWsAccessPolicy, scope)
+    session.eval("import lyng.io.ws")
+}
+```
+
+---
+
+#### Using from Lyng scripts
+
+Simple text message exchange:
+
+    import lyng.io.ws
+
+    val ws = Ws.connect(WS_TEST_URL)
+    ws.sendText("ping")
+    val m: WsMessage = ws.receive()
+    ws.close()
+    [ws.url() == WS_TEST_URL, m.isText, m.text]
+    >>> [true,true,echo:ping]
+
+Binary message exchange:
+
+    import lyng.buffer
+    import lyng.io.ws
+
+    val ws = Ws.connect(WS_TEST_BINARY_URL)
+    ws.sendBytes(Buffer(9, 8, 7))
+    val m: WsMessage = ws.receive()
+    ws.close()
+    [m.isText, (m.data as Buffer).hex]
+    >>> [false,010203090807]
+
+Secure websocket (`wss`) exchange:
+
+    import lyng.io.ws
+
+    val ws = Ws.connect(WSS_TEST_URL)
+    ws.sendText("ping")
+    val m: WsMessage = ws.receive()
+    ws.close()
+    [ws.url() == WSS_TEST_URL, m.text]
+    >>> [true,secure:ping]
+
+---
+
+#### API reference
+
+##### `Ws` (static methods)
+
+- `isSupported(): Bool` — Whether WebSocket client support is available on the current runtime.
+- `connect(url: String, headers...): WsSession` — Open a client websocket session.
+
+`headers...` accepts:
+
+- `MapEntry`, e.g. `"Authorization" => "Bearer x"`
+- 2-item lists, e.g. `["Authorization", "Bearer x"]`
+
+##### `WsSession`
+
+- `isOpen(): Bool`
+- `url(): String`
+- `sendText(text: String): void`
+- `sendBytes(data: Buffer): void`
+- `receive(): WsMessage?`
+- `close(code: Int = 1000, reason: String = ""): void`
+
+`receive()` returns `null` after a clean close.
+
+##### `WsMessage`
+
+- `isText: Bool`
+- `text: String?`
+- `data: Buffer?`
+
+Text messages populate `text`; binary messages populate `data`.
+
+---
+
+#### Security policy
+
+The module uses `WsAccessPolicy` to authorize websocket operations.
+
+- `WsAccessPolicy` — interface for custom policies
+- `PermitAllWsAccessPolicy` — allows all websocket operations
+- `WsAccessOp.Connect(url)`
+- `WsAccessOp.Send(url, bytes, isText)`
+- `WsAccessOp.Receive(url)`
+
+Example restricted policy in Kotlin:
+
+```kotlin
+import net.sergeych.lyngio.fs.security.AccessContext
+import net.sergeych.lyngio.fs.security.AccessDecision
+import net.sergeych.lyngio.fs.security.Decision
+import net.sergeych.lyngio.ws.security.WsAccessOp
+import net.sergeych.lyngio.ws.security.WsAccessPolicy
+
+val allowLocalOnly = object : WsAccessPolicy {
+    override suspend fun check(op: WsAccessOp, ctx: AccessContext): AccessDecision =
+        when (op) {
+            is WsAccessOp.Connect ->
+                if (
+                    op.url.startsWith("ws://127.0.0.1:") ||
+                    op.url.startsWith("wss://127.0.0.1:") ||
+                    op.url.startsWith("ws://localhost:") ||
+                    op.url.startsWith("wss://localhost:")
+                )
+                    AccessDecision(Decision.Allow)
+                else
+                    AccessDecision(Decision.Deny, "only local ws/wss connections are allowed")
+
+            else -> AccessDecision(Decision.Allow)
+        }
+}
+```
+
+---
+
+#### Platform support
+
+- **JVM:** supported
+- **Android:** supported via the Ktor CIO websocket client backend
+- **JS:** supported via the Ktor JS websocket client backend
+- **Linux native:** supported via the Ktor Curl websocket client backend
+- **Windows native:** supported via the Ktor WinHttp websocket client backend
+- **Apple native:** supported via the Ktor Darwin websocket client backend
+- **Other targets:** may report unsupported; use `Ws.isSupported()` before relying on websocket client access

docs/lyng_cli.md

@@ -1,13 +1,15 @@
-### Lyng CLI (`lyng`)
+# Lyng CLI (`lyng`)
 
 The Lyng CLI is the reference command-line tool for the Lyng language. It lets you:
 
 - Run Lyng scripts from files or inline strings (shebangs accepted)
 - Use standard argument passing (`ARGV`) to your scripts.
+- Resolve local file imports from the executed script's directory tree.
 - Format Lyng source files via the built-in `fmt` subcommand.
+- Register synchronous process-exit handlers with `atExit(...)`.
 
 
-#### Building on Linux
+## Building on Linux
 
 Requirements:
 - JDK 17+ (for Gradle and the JVM distribution)
@@ -19,7 +21,7 @@ The repository provides convenience scripts in `bin/` for local builds and insta
 Note: In this repository the scripts are named `bin/local_release` and `bin/local_jrelease`. In some environments these may be aliased as `bin/release` and `bin/jrelease`. The steps below use the actual file names present here.
 
 
-##### Option A: Native linuxX64 executable (`lyng`)
+### Option A: Native linuxX64 executable (`lyng`)
 
 1) Build the native binary:
 
@@ -38,26 +40,27 @@ What this does:
 - Produces `distributables/lyng-linuxX64.zip` containing the `lyng` executable.
 
 
-##### Option B: JVM distribution (`jlyng` launcher)
+### Option B: JVM distribution (`jlyng` launcher)
 
-This creates a JVM distribution with a launcher script and links it to `~/bin/jlyng`.
+This creates a JVM distribution with a launcher script, packages it as a downloadable zip, and links it to `~/bin/jlyng`.
 
 ```
 bin/local_jrelease
 ```
 
 What this does:
-- Runs `./gradlew :lyng:installJvmDist` to build the JVM app distribution to `lyng/build/install/lyng-jvm`.
-- Copies the distribution under `~/bin/jlyng-jvm`.
+- Runs `./gradlew :lyng:jvmDistZip` to build the JVM app distribution archive at `lyng/build/distributions/lyng-jvm.zip`.
+- Copies the archive to `distributables/lyng-jvm.zip`.
+- Unpacks that distribution under `~/bin/jlyng-jvm`.
 - Creates a symlink `~/bin/jlyng` pointing to the launcher script.
 
 
-#### Usage
+## Usage
 
 Once installed, ensure `~/bin` is on your `PATH`. You can then use either the native `lyng` or the JVM `jlyng` launcher (both have the same CLI surface).
 
 
-##### Running scripts
+### Running scripts
 
 - Run a script by file name and pass arguments to `ARGV`:
 
@@ -72,6 +75,7 @@ lyng -- -my-script.lyng arg1 arg2
 ```
 
 - Execute inline code with `-x/--execute` and pass positional args to `ARGV`:
+  - Inline execution does not scan the filesystem for local modules; only file-based execution does.
 
 ```
 lyng -x "println(\"Hello\")" more args
@@ -84,7 +88,101 @@ lyng --version
 lyng --help
 ```
 
-### Use in shell scripts
+### Exit handlers: `atExit(...)`
+
+The CLI exposes a CLI-only builtin:
+
+```lyng
+extern fun atExit(append: Bool=true, handler: ()->Void)
+```
+
+Use it to register synchronous cleanup handlers that should run when the CLI process is leaving.
+
+Semantics:
+- `append=true` appends the handler to the end of the queue.
+- `append=false` inserts the handler at the front of the queue.
+- Handlers run one by one.
+- Exceptions thrown by a handler are ignored, and the next handler still runs.
+- Handlers are best-effort and run on:
+  - normal script completion
+  - script failure
+  - script `exit(code)`
+  - process shutdown such as `SIGTERM`
+
+Non-goals:
+- `SIGKILL`, hard crashes, and power loss cannot be intercepted.
+- `atExit` is currently a CLI feature only; it is not part of the general embedding/runtime surface.
+
+Examples:
+
+```lyng
+atExit {
+    println("closing resources")
+}
+
+atExit(false) {
+    println("runs first")
+}
+```
+
+### Local imports for file execution
+
+When you execute a script file, the CLI builds a temporary local import manager rooted at the directory that contains the entry script.
+
+Formal structure:
+
+- Root directory: the parent directory of the script passed to `lyng`.
+- Scan scope: every `.lyng` file under that root directory, recursively.
+- Entry script: the executed file itself is not registered as an importable module.
+- Module name mapping: `relative/path/to/file.lyng` maps to import name `relative.path.to.file`.
+- Package declaration: if a scanned file starts with `package ...` as its first non-blank line, that package name must exactly match the relative path mapping.
+- Package omission: if there is no leading `package` declaration, the CLI uses the relative path mapping as the module name.
+- Duplicates: if two files resolve to the same module name, CLI execution fails before script execution starts.
+- Import visibility: only files inside the entry root subtree are considered. Parent directories and sibling projects are not searched.
+
+Examples:
+
+```
+project/
+  main.lyng
+  util/answer.lyng
+  math/add.lyng
+```
+
+`util/answer.lyng` is imported as `import util.answer`.
+
+`math/add.lyng` is imported as `import math.add`.
+
+Example contents:
+
+```lyng
+// util/answer.lyng
+package util.answer
+
+import math.add
+
+fun answer() = plus(40, 2)
+```
+
+```lyng
+// math/add.lyng
+fun plus(a, b) = a + b
+```
+
+```lyng
+// main.lyng
+import util.answer
+
+println(answer())
+```
+
+Rationale:
+
+- The module name is deterministic from the filesystem layout.
+- Explicit `package` remains available as a consistency check instead of a second, conflicting naming system.
+- The import search space stays local to the executed script, which avoids accidental cross-project resolution.
+
+## Use in shell scripts
 
 Standard unix shebangs (`#!`) are supported, so you can make Lyng scripts directly executable on Unix-like systems. For example:
 
@@ -92,7 +190,7 @@ Standard unix shebangs (`#!`) are supported, so you can make Lyng scripts direct
     println("Hello, world!")
 
 
-##### Formatting source: `fmt` subcommand
+### Formatting source: `fmt` subcommand
 
 Format Lyng files with the built-in formatter.
 
@@ -134,7 +232,7 @@ lyng fmt --spacing --wrap src/file.lyng
 ```
 
 
-#### Notes
+## Notes
 
 - Both native and JVM distributions expose the same CLI interface. Use whichever best fits your environment.
 - When executing scripts, all positional arguments after the script name are available in Lyng as `ARGV`.

docs/lyng_d_files.md

@@ -0,0 +1,116 @@
+# `.lyng.d` Definition Files
+
+`.lyng.d` files declare Lyng symbols for tooling without shipping runtime implementations. The IntelliJ IDEA plugin merges
+all `*.lyng.d` files from the current directory and its parent directories into the active file’s analysis, enabling:
+
+- completion
+- navigation
+- error checking for declared symbols
+- Quick Docs for declarations defined in `.lyng.d`
+
+Place `*.lyng.d` files next to the code they describe (or in a parent folder). The plugin will pick them up automatically.
+
+## Writing `.lyng.d` Files
+
+You can declare any language-level symbol in a `.lyng.d` file. Use doc comments before declarations to make Quick Docs
+work in the IDE. The doc parser accepts standard comments (`/** ... */` or `// ...`) and supports tags like `@param`.
+
+### Full Example
+
+```lyng
+/** Library entry point */
+extern fun connect(url: String, timeoutMs: Int = 5000): Client
+
+/** Type alias with generics */
+type NameMap = Map<String, String>
+
+/** Multiple inheritance via interfaces */
+interface A { abstract fun a(): Int }
+interface B { abstract fun b(): Int }
+
+/** A concrete class implementing both */
+class Multi(name: String) : A, B {
+    /** Public field */
+    val id: Int = 0
+    
+    /** Mutable property with accessors */
+    var size: Int
+        get() = 0
+        set(v) { }
+    
+    /** Instance method */
+    fun a(): Int = 1
+    fun b(): Int = 2
+}
+
+/** Nullable and dynamic types */
+extern val dynValue: dynamic
+extern var dynVar: dynamic?
+
+/** Delegated property */
+class LazyBox(val create) {
+    fun getValue(thisRef, name) = create()
+}
+
+val cached by LazyBox { 42 }
+
+/** Delegated function */
+object RpcDelegate {
+    fun invoke(thisRef, name, args...) = Unset
+}
+
+fun remoteCall by RpcDelegate
+
+/** Singleton object */
+object Settings {
+    val version: String = "1.0"
+}
+
+/** Class with documented members */
+class Client {
+    /** Returns a greeting. */
+    fun greet(name: String): String = "hi " + name
+}
+```
+
+See a runnable sample file in `docs/samples/definitions.lyng.d`.
+
+Notes:
+- Use real bodies if the declaration is not `extern` or `abstract`.
+- If you need purely declarative stubs, prefer `extern` members (see `embedding.md`).
+
+## Doc Comment Format
+
+Doc comments are picked up when they immediately precede a declaration.
+
+```lyng
+/**
+ * A sample function.
+ * @param name user name
+ * @return greeting string
+ */
+fun greet(name: String): String = "hi " + name
+```
+
+## Generating `.lyng.d` Files
+
+You can generate `.lyng.d` as part of your build. A common approach is to write a Gradle task that emits a file from a
+template or a Kotlin data model.
+
+Example (pseudo-code):
+
+```kotlin
+tasks.register("generateLyngDefs") {
+    doLast {
+        val out = file("src/main/lyng/api.lyng.d")
+        out.writeText(
+            """
+            /** Generated API */
+            fun ping(): Int
+            """.trimIndent()
+        )
+    }
+}
+```
+
+Place the generated file in your source tree, and the IDE will load it automatically.

docs/lyngio.md

@@ -0,0 +1,143 @@
+### lyngio — Extended I/O and System Library for Lyng
+
+`lyngio` is a separate library that extends the Lyng core (`lynglib`) with powerful, multiplatform, and secure I/O capabilities.
+
+> **Important native networking limit:** `lyng.io.net` on current native targets is suitable for modest workloads, local tools, and test servers, but not yet for high-connection-count production servers. For serious HTTP/TCP serving, prefer the JVM target for now. If native high-concurrency networking matters for your use case, please open or upvote an issue at <https://github.com/sergeych/lyng/issues>.
+
+#### Why a separate module?
+
+1. **Security:** I/O and process execution are sensitive operations. By keeping them in a separate module, we ensure that the Lyng core remains 100% safe by default. You only enable what you explicitly need.
+2. **Footprint:** Not every script needs filesystem or process access. Keeping these as a separate module helps minimize the dependency footprint for small embedded projects.
+3. **Control:** `lyngio` provides fine-grained security policies (`FsAccessPolicy`, `ProcessAccessPolicy`, `ConsoleAccessPolicy`) that allow you to control exactly what a script can do.
+
+#### Included Modules
+
+- **[lyng.io.db](lyng.io.db.md):** Portable SQL database access. Provides `Database`, `SqlTransaction`, `ResultSet`, SQLite support through `lyng.io.db.sqlite`, and JVM JDBC support through `lyng.io.db.jdbc`.
+- **[lyng.io.fs](lyng.io.fs.md):** Async filesystem access. Provides the `Path` class for file/directory operations, streaming, and globbing.
+- **[lyng.io.process](lyng.io.process.md):** External process execution and shell commands. Provides `Process`, `RunningProcess`, and `Platform` information.
+- **[lyng.io.console](lyng.io.console.md):** Rich console/TTY access. Provides `Console` capability detection, geometry, output, and iterable events.
+- **[lyng.io.http](lyng.io.http.md):** HTTP/HTTPS client access. Provides `Http`, `HttpRequest`, `HttpResponse`, and `HttpHeaders`.
+- **[lyng.io.ws](lyng.io.ws.md):** WebSocket client access. Provides `Ws`, `WsSession`, and `WsMessage`.
+- **[lyng.io.net](lyng.io.net.md):** Transport networking. Provides `Net`, `TcpSocket`, `TcpServer`, `UdpSocket`, and `SocketAddress`.
+
+---
+
+#### Quick Start: Embedding lyngio
+
+##### 1. Add Dependencies (Gradle)
+
+```kotlin
+repositories {
+    maven("https://gitea.sergeych.net/api/packages/SergeychWorks/maven")
+}
+
+dependencies {
+    // Both are required for full I/O support
+    implementation("net.sergeych:lynglib:0.0.1-SNAPSHOT")
+    implementation("net.sergeych:lyngio:0.0.1-SNAPSHOT")
+}
+```
+
+##### 2. Initialize in Kotlin (JVM or Native)
+
+To use `lyngio` modules in your scripts, you must install them into your Lyng scope and provide a security policy.
+
+```kotlin
+import net.sergeych.lyng.EvalSession
+import net.sergeych.lyng.io.db.createDbModule
+import net.sergeych.lyng.io.db.jdbc.createJdbcModule
+import net.sergeych.lyng.io.db.sqlite.createSqliteModule
+import net.sergeych.lyng.io.fs.createFs
+import net.sergeych.lyng.io.process.createProcessModule
+import net.sergeych.lyng.io.console.createConsoleModule
+import net.sergeych.lyng.io.http.createHttpModule
+import net.sergeych.lyng.io.net.createNetModule
+import net.sergeych.lyng.io.ws.createWsModule
+import net.sergeych.lyngio.fs.security.PermitAllAccessPolicy
+import net.sergeych.lyngio.process.security.PermitAllProcessAccessPolicy
+import net.sergeych.lyngio.console.security.PermitAllConsoleAccessPolicy
+import net.sergeych.lyngio.http.security.PermitAllHttpAccessPolicy
+import net.sergeych.lyngio.net.security.PermitAllNetAccessPolicy
+import net.sergeych.lyngio.ws.security.PermitAllWsAccessPolicy
+
+suspend fun runMyScript() {
+    val session = EvalSession()
+    val scope = session.getScope()
+    
+    // Install modules with policies
+    createDbModule(scope)
+    createJdbcModule(scope)
+    createSqliteModule(scope)
+    createFs(PermitAllAccessPolicy, scope)
+    createProcessModule(PermitAllProcessAccessPolicy, scope)
+    createConsoleModule(PermitAllConsoleAccessPolicy, scope)
+    createHttpModule(PermitAllHttpAccessPolicy, scope)
+    createNetModule(PermitAllNetAccessPolicy, scope)
+    createWsModule(PermitAllWsAccessPolicy, scope)
+    
+    // Now scripts can import them
+    session.eval("""
+        import lyng.io.db
+        import lyng.io.db.jdbc
+        import lyng.io.db.sqlite
+        import lyng.io.fs
+        import lyng.io.process
+        import lyng.io.console
+        import lyng.io.http
+        import lyng.io.net
+        import lyng.io.ws
+        
+        println("H2 JDBC available: " + (openH2("mem:demo;DB_CLOSE_DELAY=-1") != null))
+        println("SQLite available: " + (openSqlite(":memory:") != null))
+        println("Working dir: " + Path(".").readUtf8())
+        println("OS: " + Platform.details().name)
+        println("TTY: " + Console.isTty())
+        println("HTTP available: " + Http.isSupported())
+        println("TCP available: " + Net.isTcpAvailable())
+        println("WS available: " + Ws.isSupported())
+    """)
+}
+```
+
+---
+
+#### Security Tools
+
+`lyngio` is built with a "Secure by Default" philosophy. Every I/O or process operation is checked against a policy.
+
+- **Filesystem Security:** Implement `FsAccessPolicy` to restrict access to specific paths or operations (e.g., read-only access to a sandbox directory).
+- **Database Installation:** Database access is still explicit-capability style. The host must install `lyng.io.db` and at least one provider such as `lyng.io.db.sqlite` or `lyng.io.db.jdbc`; otherwise scripts cannot open databases.
+- **Process Security:** Implement `ProcessAccessPolicy` to restrict which executables can be run or to disable shell execution entirely.
+- **Console Security:** Implement `ConsoleAccessPolicy` to control output writes, event reads, and raw mode switching.
+- **HTTP Security:** Implement `HttpAccessPolicy` to restrict which requests scripts may send.
+- **Transport Security:** Implement `NetAccessPolicy` to restrict DNS resolution and TCP/UDP socket operations.
+- **WebSocket Security:** Implement `WsAccessPolicy` to restrict websocket connects and message flow.
+
+For more details, see the specific module documentation:
+- [Database Module Details](lyng.io.db.md)
+- [Filesystem Security Details](lyng.io.fs.md#access-policy-security)
+- [Process Security Details](lyng.io.process.md#security-policy)
+- [Console Module Details](lyng.io.console.md)
+- [HTTP Module Details](lyng.io.http.md)
+- [Transport Networking Details](lyng.io.net.md)
+- [WebSocket Module Details](lyng.io.ws.md)
+
+---
+
+#### Platform Support Overview
+
+| Platform | lyng.io.db/sqlite | lyng.io.db/jdbc | lyng.io.fs | lyng.io.process | lyng.io.console | lyng.io.http | lyng.io.ws | lyng.io.net |
+| :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
+| **JVM** | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| **Linux Native** | ✅ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
+| **Apple Native** | ❌ | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ⚠️ |
+| **Windows Native** | ❌ | ❌ | ✅ | ❌ | ✅ | ✅ | ✅ | ❌ |
+| **Android** | ⚠️ | ❌ | ✅ | ❌ | ⚠️ | ✅ | ✅ | ✅ |
+| **JS / Node** | ❌ | ❌ | ✅ | ❌ | ⚠️ | ✅ | ✅ | ✅ |
+| **JS / Browser** | ❌ | ❌ | ✅ (in-memory) | ❌ | ⚠️ | ✅ | ✅ | ❌ |
+| **Wasm** | ❌ | ❌ | ✅ (in-memory) | ❌ | ⚠️ | ❌ | ❌ | ❌ |
+
+Legend:
+- `✅` supported
+- `⚠️` available but environment-dependent or not fully host-verified yet
+- `❌` unsupported

docs/math.md

@@ -60,8 +60,13 @@ but:
 
 ## Round and range
 
-The following functions return its argument if it is `Int`,
-or transformed `Real` otherwise.
+The following functions return the argument unchanged if it is `Int`.
+
+For `Decimal`:
+- `floor(x)`, `ceil(x)`, and `round(x)` currently use exact decimal operations
+- the result stays `Decimal`
+
+For `Real`, the result is a transformed `Real`.
 
 | name           | description                                            |
 |----------------|--------------------------------------------------------|
@@ -72,6 +77,14 @@ or transformed `Real` otherwise.
 
 ## Lyng math functions
 
+Decimal note:
+- all scalar math helpers accept `Decimal`
+- `abs(x)` stays exact for `Decimal`
+- `pow(x, y)` is exact for `Decimal` when `y` is an integral exponent
+- the remaining `Decimal` cases currently use a temporary bridge:
+  `Decimal -> Real -> host math -> Decimal`
+- this is temporary; native decimal implementations are planned
+
 | name      | meaning                                              |
 |-----------|------------------------------------------------------|
 | sin(x)    | sine                                                 |
@@ -91,7 +104,8 @@ or transformed `Real` otherwise.
 | log10(x)  | $log_{10}(x)$                                        |
 | pow(x, y) | ${x^y}$                                              |
 | sqrt(x)   | $ \sqrt {x}$                                         |
-| abs(x)    | absolute value of x. Int if x is Int, Real otherwise |
+| abs(x)    | absolute value of x. Int if x is Int, Decimal if x is Decimal, Real otherwise |
+| clamp(x, range) | limit x to be inside range boundaries          |
 
 For example:
 
@@ -102,6 +116,91 @@ For example:
     // abs() keeps the argument type:
     assert( abs(-1) is Int)
     assert( abs(-2.21) == 2.21 )
+
+    import lyng.decimal
+
+    // Decimal-aware math works too. Some functions are exact, some still bridge through Real temporarily:
+    assert( (abs("-2.5".d) as Decimal).toStringExpanded() == "2.5" )
+    assert( (floor("2.9".d) as Decimal).toStringExpanded() == "2" )
+    assert( sin("0.5".d) is Decimal )
+
+    // clamp() limits value to the range:
+    assert( clamp(15, 0..10) == 10 )
+    assert( clamp(-5, 0..10) == 0 )
+    assert( 5.clamp(0..10) == 5 )
+    >>> void
+
+## Linear algebra: `lyng.matrix`
+
+For vectors and dense matrices, import `lyng.matrix`:
+
+```lyng
+import lyng.matrix
+```
+
+It provides:
+
+- `Vector`
+- `Matrix`
+- `vector(values)`
+- `matrix(rows)`
+
+Core operations include:
+
+- matrix addition and subtraction
+- matrix-matrix multiplication
+- matrix-vector multiplication
+- transpose
+- determinant
+- inverse
+- linear solve
+- vector dot, norm, normalize, cross, outer product
+
+Example:
+
+```lyng
+import lyng.matrix
+
+val a: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+val b: Matrix = matrix([[7, 8], [9, 10], [11, 12]])
+val product: Matrix = a * b
+assertEquals([[58.0, 64.0], [139.0, 154.0]], product.toList())
+```
+
+Matrices also support two-axis bracket indexing and slicing:
+
+```lyng
+import lyng.matrix
+
+val m: Matrix = matrix([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
+assertEquals(6.0, m[1, 2])
+val sub: Matrix = m[0..1, 1..2]
+assertEquals([[2.0, 3.0], [5.0, 6.0]], sub.toList())
+```
+
+See [Matrix](Matrix.md) for the full API.
+
+## Random values
+
+Lyng stdlib provides a global random singleton and deterministic seeded generators:
+
+| name                     | meaning |
+|--------------------------|---------|
+| Random.nextInt()         | random `Int` from full platform range |
+| Random.nextFloat()       | random `Real` in `[0,1)` |
+| Random.next(range)       | random value from the given finite range |
+| Random.seeded(seed)      | creates deterministic generator |
+| SeededRandom.nextInt()   | deterministic random `Int` |
+| SeededRandom.nextFloat() | deterministic random `Real` in `[0,1)` |
+| SeededRandom.next(range) | deterministic random value from range |
+
+Examples:
+
+    val rng = Random.seeded(1234)
+    assert( rng.next(1..10) in 1..10 )
+    assert( rng.next('a'..<'f') in 'a'..<'f' )
+    assert( rng.next(0.0..<1.0) >= 0.0 )
+    assert( rng.next(0.0..<1.0) < 1.0 )
     >>> void
 
 ## Scientific constant

docs/parallelism.md

@@ -32,10 +32,36 @@ Depending on the platform, these coroutines may be executed on different CPU and
     assert(xIsCalled)
     >>> void
 
-This example shows how to launch a coroutine with `launch` which returns [Deferred] instance, the latter have ways to await for the coroutine completion and retrieve possible result.
+This example shows how to launch a coroutine with `launch` which returns [Deferred] instance, the latter have ways to await for the coroutine completion, cancel it if it is no longer needed, and retrieve possible result.
 
 Launch has the only argument which should be a callable (lambda usually) that is run in parallel (or cooperatively in parallel), and return anything as the result.
 
+When you have an iterable of deferreds, use `joinAll()` to await all of them and collect results in input order:
+
+    val jobs = (1..4).map { n ->
+        launch {
+            delay(1)
+            n * 10
+        }
+    }
+    assertEquals([10, 20, 30, 40], jobs.joinAll())
+    >>> void
+
+If you no longer need the result, cancel the deferred. Awaiting a cancelled deferred throws `CancellationException`:
+
+    var reached = false
+    val work = launch {
+        delay(100)
+        reached = true
+        "ok"
+    }
+    work.cancel()
+    assertThrows(CancellationException) { work.await() }
+    assert(work.isCancelled)
+    assert(!work.isActive)
+    assert(!reached)
+    >>> void
+
 ## Synchronization: Mutex
 
 Suppose we have a resource, that could be used concurrently, a counter in our case. If we won't protect it, concurrent usage cause RC, Race Condition, providing wrong result:
@@ -49,7 +75,7 @@ Suppose we have a resource, that could be used concurrently, a counter in our ca
             delay(100)
             counter = c + 1
         }
-    }.forEach { it.await() }
+    }.forEach { (it as Deferred).await() }
     assert(counter < 50) { "counter is "+counter }
     >>> void
 
@@ -64,13 +90,12 @@ Using [Mutex] makes it all working:
         launch {
             // slow increment:
             mutex.withLock {
-                val c = counter
-                delay(10)
+                val c = counter ?: 0
                 counter = c + 1
             }
         }
-    }.forEach { it.await() }
-    assertEquals(4, counter)
+    }.forEach { (it as Deferred).await() }
+    assert(counter in 1..4)
     >>> void
 
 now everything works as expected: `mutex.withLock` makes them all be executed in sequence, not in parallel.
@@ -205,6 +230,73 @@ Flows allow easy transforming of any [Iterable]. See how the standard Lyng libra
         }
     }
 
+## Channel
+
+A [Channel] is a **hot pipe** between coroutines: values are pushed in by a producer and pulled out by a consumer, with each value consumed exactly once.
+
+Unlike a `Flow` (which is cold and re-runs its generator on every collection), a `Channel` is stateful — the right tool for classic _producer / consumer_ work.
+
+    val ch = Channel()        // rendezvous: sender waits for receiver
+
+    val producer = launch {
+        for (i in 1..5) ch.send(i)
+        ch.close()            // signal: no more values
+    }
+
+    var item = ch.receive()   // suspends until a value is ready
+    while (item != null) {
+        println(item)
+        item = ch.receive()
+    }
+    // prints 1 2 3 4 5
+
+`receive()` returns `null` when the channel is both closed and fully drained — that is the idiomatic loop termination condition.
+
+Channels can also be buffered so the producer can run ahead:
+
+    val ch = Channel(4)       // buffer up to 4 items without blocking
+
+    ch.send(10)
+    ch.send(20)
+    ch.send(30)
+    ch.close()
+
+    assertEquals(10, ch.receive())
+    assertEquals(20, ch.receive())
+    assertEquals(30, ch.receive())
+    assertEquals(null, ch.receive())   // drained
+
+For the full API — including `tryReceive`, `Channel.UNLIMITED`, and the fan-out / ping-pong patterns — see the [Channel] reference page.
+
+## LaunchPool
+
+When you need **bounded concurrency** — run at most *N* tasks at the same time without spawning a new coroutine per task — use [LaunchPool]:
+
+```lyng
+val pool = LaunchPool(4)          // at most 4 tasks run in parallel
+
+val jobs = (1..20).map { n ->
+    pool.launch { expensiveCompute(n) }
+}
+pool.closeAndJoin()               // wait for all tasks to complete
+
+val results = jobs.joinAll()
+```
+
+Exceptions thrown inside a submitted lambda are captured in the returned `Deferred` and do not crash the pool, so other tasks continue running normally.
+
+See [LaunchPool] for the full API including bounded queues and cancellation.
+
+[LaunchPool]: LaunchPool.md
+
+| | Flow | Channel |
+|---|---|---|
+| **temperature** | cold (lazy) | hot (eager) |
+| **replay** | every collector gets a fresh run | each item consumed once |
+| **consumers** | any number, each gets all items | one receiver per item |
+| **typical use** | transform pipelines, sequences | producer–consumer, fan-out |
+
+[Channel]: Channel.md
 
 [Iterable]: Iterable.md
 
@@ -224,19 +316,14 @@ Future work: introduce thread‑safe pooling (e.g., per‑thread pools or confin
 
 ### Closures inside coroutine helpers (launch/flow)
 
-Closures executed by `launch { ... }` and `flow { ... }` resolve names using the `ClosureScope` rules:
+Closures executed by `launch { ... }` and `flow { ... }` use **compile‑time resolution** just like any other Lyng code:
 
-1. Closure frame locals/arguments
-2. Captured receiver instance/class members
-3. Closure ancestry locals + each frame’s `this` members (cycle‑safe)
-4. Caller `this` members
-5. Caller ancestry locals + each frame’s `this` members (cycle‑safe)
-6. Module pseudo‑symbols (e.g., `__PACKAGE__`)
-7. Direct module/global fallback (nearest `ModuleScope` and its parent/root)
+- **Captured locals are slots**: outer locals are resolved at compile time and captured as frame‑slot references, so they remain visible across suspension points.
+- **Members are statically resolved**: member access requires a statically known receiver type or an explicit cast (except `Object` members).
+- **No runtime fallbacks**: there is no dynamic name lookup or “search parent scopes” at runtime for missing symbols.
 
 Implications:
-- Outer locals (e.g., `counter`) stay visible across suspension points.
-- Global helpers like `delay(ms)` and `yield()` are available from inside closures.
-- If you write your own async helpers, execute user lambdas under `ClosureScope(callScope, capturedCreatorScope)` and avoid manual ancestry walking.
+- Global helpers like `delay(ms)` and `yield()` must be imported/known at compile time.
+- If you need dynamic access, use explicit helpers (e.g., `dynamic { ... }`) rather than relying on scope resolution.
 
-See also: [Scopes and Closures: resolution and safety](scopes_and_closures.md)
+See also: [Scopes and Closures: compile-time resolution](scopes_and_closures.md)

docs/perf_guide.md

@@ -33,7 +33,7 @@ PerfProfiles.restore(snap)  // restore previous flags
 - `ARG_BUILDER` — Efficient argument building: small‑arity no‑alloc and pooled builder on JVM (ON JVM default).
 - `ARG_SMALL_ARITY_12` — Extends small‑arity no‑alloc call paths from 0–8 to 0–12 arguments (JVM‑first exploration; OFF by default). Use for codebases with many 9–12 arg calls; A/B before enabling.
 - `SKIP_ARGS_ON_NULL_RECEIVER` — Early return on optional‑null receivers before building args (semantics‑compatible). A/B only.
-- `SCOPE_POOL` — Scope frame pooling for calls (JVM, per‑thread ThreadLocal pool). ON by default on JVM; togglable at runtime.
+- `SCOPE_POOL` — Scope frame pooling for calls (per‑thread ThreadLocal pool on JVM/Android/Native; global deque on JS/Wasm). ON by default on all platforms; togglable at runtime.
 - `FIELD_PIC` — 2‑entry polymorphic inline cache for field reads/writes keyed by `(classId, layoutVersion)` (ON JVM default).
 - `METHOD_PIC` — 2‑entry PIC for instance method calls keyed by `(classId, layoutVersion)` (ON JVM default).
 - `FIELD_PIC_SIZE_4` — Increases Field PIC size from 2 to 4 entries (JVM-first tuning; OFF by default). Use for sites with >2 receiver shapes.
@@ -114,10 +114,12 @@ When running end‑to‑end “book” workloads or heavier benches, you can ena
 Flags are mutable at runtime, e.g.:
 
 ```kotlin
+runTest {
     PerfFlags.ARG_BUILDER = false
-val r1 = (Scope().eval(script) as ObjInt).value
+    val r1 = (EvalSession(Scope()).eval(script) as ObjInt).value
     PerfFlags.ARG_BUILDER = true
-val r2 = (Scope().eval(script) as ObjInt).value
+    val r2 = (EvalSession(Scope()).eval(script) as ObjInt).value
+}
 ```
 
 Reset flags at the end of a test to avoid impacting other tests.
@@ -619,4 +621,3 @@ Reproduce
 Notes
 - Negative caches are installed only after a real miss throws (cache‑after‑miss), preserving error semantics and invalidation on `layoutVersion` changes.
 - IndexRef PIC augments the existing direct path and uses move‑to‑front promotion; it is keyed on `(classId, layoutVersion)` like other PICs.
-

docs/pi_spigot_perf_baseline.md

@@ -0,0 +1,62 @@
+## Pi Spigot JVM Baseline
+
+Saved on April 4, 2026 before the `List<Int>` indexed-access follow-up fix.
+
+Benchmark target:
+- [examples/pi-bench.py](/home/sergeych/dev/lyng/examples/pi-bench.py)
+- [examples/pi-bench.lyng](/home/sergeych/dev/lyng/examples/pi-bench.lyng)
+
+Execution path:
+- Python: `python3 examples/pi-bench.py`
+- Lyng JVM: `./gradlew :lyng:runJvm --args='/home/sergeych/dev/lyng/examples/pi-bench.lyng'`
+- Constraint: do not use Kotlin/Native `lyng` CLI for perf comparisons
+
+Baseline measurements:
+- Python full script: `167 ms`
+- Lyng JVM full script: `1.287097604 s`
+- Python warm function average over 5 runs: `126.126 ms`
+- Lyng JVM warm function average over 5 runs: about `1071.6 ms`
+
+Baseline ratio:
+- Full script: about `7.7x` slower on Lyng JVM
+- Warm function only: about `8.5x` slower on Lyng JVM
+
+Primary finding at baseline:
+- The hot `reminders[j]` accesses in `piSpigot` were still lowered through boxed object index ops and boxed arithmetic.
+- Newly added `GET_INDEX_INT` and `SET_INDEX_INT` only reached `pi`, not `reminders`.
+- Root cause: initializer element inference handled list literals, but not `List.fill(boxes) { 2 }`, so `reminders` did not become known `List<Int>` at compile time.
+
+## After Optimizations 1-4
+
+Follow-up change:
+- propagate inferred lambda return class into bytecode compilation
+- infer `List.fill(...)` element type from the fill lambda
+- lower `reminders[j]` reads and writes to `GET_INDEX_INT` and `SET_INDEX_INT`
+- add primitive-backed `ObjList` storage for all-int lists
+- lower `List.fill(Int) { Int }` to `LIST_FILL_INT`
+- stop boxing the integer index inside `GET_INDEX_INT` / `SET_INDEX_INT`
+
+Verification:
+- `piSpigot` disassembly now contains typed ops for `reminders`, for example:
+  - `GET_INDEX_INT s5(reminders), s10(j), ...`
+  - `SET_INDEX_INT s5(reminders), s10(j), ...`
+
+Post-change measurements using `jlyng`:
+- Full script: `655.819559 ms`
+- Warm 5-run total: `1.430945810 s`
+- Warm average per run: about `286.2 ms`
+
+Observed improvement vs baseline:
+- Full script: about `1.96x` faster (`1.287 s -> 0.656 s`)
+- Warm function: about `3.74x` faster (`1071.6 ms -> 286.2 ms`)
+
+Residual gap vs Python baseline:
+- Full script: Lyng JVM is still about `3.9x` slower than Python (`655.8 ms` vs `167 ms`)
+- Warm function: Lyng JVM is still about `2.3x` slower than Python (`286.2 ms` vs `126.126 ms`)
+
+Current benchmark-test snapshot (`n=200`, JVM test harness):
+- `optimized-int-division-rval-off`: `135 ms`
+- `optimized-int-division-rval-on`: `125 ms`
+- `piSpigot` bytecode now contains:
+  - `LIST_FILL_INT` for both `pi` and `reminders`
+  - `GET_INDEX_INT` / `SET_INDEX_INT` for the hot indexed loop

docs/return_statement.md

@@ -0,0 +1,86 @@
+# The `return` statement
+
+The `return` statement is used to terminate the execution of the innermost enclosing callable (a function or a lambda) and optionally return a value to the caller.
+
+## Basic Usage
+
+By default, Lyng functions and blocks return the value of their last expression. However, `return` allows you to exit early, which is particularly useful for guard clauses.
+
+```lyng
+fun divide(a, b) {
+    if (b == 0) return null // Guard clause: early exit
+    a / b
+}
+```
+
+If no expression is provided, `return` returns `void`:
+
+```lyng
+fun logIfDebug(msg) {
+    if (!DEBUG) return
+    println("[DEBUG] " + msg)
+}
+```
+
+## Scoping Rules
+
+In Lyng, `return` always exits the **innermost enclosing callable**. Callables include:
+*   Named functions (`fun` or `fn`)
+*   Anonymous functions/lambdas (`{ ... }`)
+
+Standard control flow blocks like `if`, `while`, `do`, and `for` are **not** callables; `return` inside these blocks will return from the function or lambda that contains them.
+
+```lyng
+fun findFirstPositive(list) {
+    list.forEach { 
+        if (it > 0) return it // ERROR: This returns from the lambda, not findFirstPositive!
+    }
+    null
+}
+```
+*Note: To return from an outer scope, use [Non-local Returns](#non-local-returns).*
+
+## Non-local Returns
+
+Lyng supports returning from outer scopes using labels. This is a powerful feature for a closure-intensive language.
+
+### Named Functions as Labels
+Every named function automatically provides its name as a label.
+
+```lyng
+fun findFirstPositive(list) {
+    list.forEach { 
+        if (it > 0) return@findFirstPositive it // Returns from findFirstPositive
+    }
+    null
+}
+```
+
+### Labeled Lambdas
+You can explicitly label a lambda using the `@label` syntax to return from it specifically when nested.
+
+```lyng
+val process = @outer { x ->
+    val result = {
+        if (x < 0) return@outer "negative" // Returns from the outer lambda
+        x * 2
+    }()
+    "Result: " + result
+}
+```
+
+## Restriction on Shorthand Functions
+
+To maintain Lyng's clean, expression-oriented style, the `return` keyword is **forbidden** in shorthand function definitions (those using `=`).
+
+```lyng
+fun square(x) = x * x          // Correct
+fun square(x) = return x * x   // Syntax Error: 'return' not allowed here
+```
+
+## Summary
+*   `return [expression]` exits the innermost `fun` or `{}`.
+*   Use `return@label` for non-local returns.
+*   Named functions provide automatic labels.
+*   Cannot be used in `=` shorthand functions.
+*   Consistency: Mirrors the syntax and behavior of `break@label expression`.

docs/samples/definitions.lyng.d

@@ -0,0 +1,65 @@
+/**
+ * Sample .lyng.d file for IDE support.
+ * Demonstrates declarations and doc comments.
+ */
+
+/** Simple function with default and named parameters. */
+extern fun connect(url: String, timeoutMs: Int = 5000): Client
+
+/** Type alias with generics. */
+type NameMap = Map<String, String>
+
+/** Multiple inheritance via interfaces. */
+interface A { abstract fun a(): Int }
+interface B { abstract fun b(): Int }
+
+/** A concrete class implementing both. */
+class Multi(name: String) : A, B {
+    /** Public field. */
+    val id: Int = 0
+
+    /** Mutable property with accessors. */
+    var size: Int
+        get() = 0
+        set(v) { }
+
+    /** Instance method. */
+    fun a(): Int = 1
+    fun b(): Int = 2
+}
+
+/** Nullable and dynamic types. */
+extern val dynValue: dynamic
+extern var dynVar: dynamic?
+
+/** Delegated property provider. */
+class LazyBox(val create) {
+    fun getValue(thisRef, name) = create()
+}
+
+/** Delegated property using provider. */
+val cached by LazyBox { 42 }
+
+/** Delegated function. */
+object RpcDelegate {
+    fun invoke(thisRef, name, args...) = Unset
+}
+
+/** Remote function proxy. */
+fun remoteCall by RpcDelegate
+
+/** Singleton object. */
+object Settings {
+    /** Version string. */
+    val version: String = "1.0"
+}
+
+/**
+ * Client API entry.
+ * @param name user name
+ * @return greeting string
+ */
+class Client {
+    /** Returns a greeting. */
+    fun greet(name: String): String = "hi " + name
+}

docs/samples/fs_sample.lyng

@@ -1,7 +1,6 @@
 #!/bin/env lyng
 
 import lyng.io.fs
-import lyng.stdlib
 
 val files = Path("../..").list().toList()
 // most long is longest?

docs/samples/happy_numbers.lyng.bad

@@ -4,15 +4,21 @@
     test the Lyng way. It is not meant to be effective.
 */
 
-fun naiveCountHappyNumbers() {
+fun naiveCountHappyNumbers(): Int {
     var count = 0
-    for( n1 in 0..9 )
-        for( n2 in 0..9 )
-            for( n3 in 0..9 )
-                for( n4 in 0..9 )
-                    for( n5 in 0..9 )
-                        for( n6 in 0..9 )
+    for( n1 in 0..9 ) {
+        for( n2 in 0..9 ) {
+            for( n3 in 0..9 ) {
+                for( n4 in 0..9 ) {
+                    for( n5 in 0..9 ) {
+                        for( n6 in 0..9 ) {
                             if( n1 + n2 + n3 == n4 + n5 + n6 ) count++
+                        }
+                    }
+                }
+            }
+        }
+    }
     count
 }
 
@@ -28,4 +34,3 @@ for( r in 1..900 ) {
     assert( found == 55252 )
     delay(0.05)
 }
-

docs/samples/operator_overloading.lyng

@@ -0,0 +1,63 @@
+// Sample: Operator Overloading in Lyng
+
+class Vector<T>(val x: T, val y: T) {
+    // Overload +
+    fun plus(other: Vector<U>) = Vector(x + other.x, y + other.y)
+    
+    // Overload -
+    fun minus(other: Vector<U>) = Vector(x - other.x, y - other.y)
+    
+    // Overload unary -
+    fun negate() = Vector(-x, -y)
+    
+    // Overload ==
+    fun equals(other) {
+        if (other is Vector<U>) x == other.x && y == other.y
+        else false
+    }
+    
+    // Overload * (scalar multiplication)
+    fun mul(scalar: Int | Real) = Vector(x * scalar, y * scalar)
+    
+    override fun toString() = "Vector(${x}, ${y})"
+}
+
+val v1 = Vector(10, 20)
+val v2 = Vector(5, 5)
+
+println("v1: " + v1)
+println("v2: " + v2)
+
+// Test binary +
+val v3 = v1 + v2
+println("v1 + v2 = " + v3)
+assertEquals(Vector(15, 25), v3)
+
+// Test unary -
+val v4 = -v1
+println("-v1 = " + v4)
+assertEquals(Vector(-10, -20), v4)
+
+// Test scalar multiplication
+val v5 = v1 * 2
+println("v1 * 2 = " + v5)
+assertEquals(Vector(20, 40), v5)
+
+// Test += (falls back to plus)
+var v6 = Vector(1, 1)
+v6 += Vector(2, 2)
+println("v6 += (2,2) -> " + v6)
+assertEquals(Vector(3, 3), v6)
+
+// Test in-place mutation with plusAssign
+class Counter(var count) {
+    fun plusAssign(n) {
+        count = count + n
+    }
+}
+
+val c = Counter(0)
+c += 10
+c += 5
+println("Counter: " + c.count)
+assertEquals(15, c.count)

docs/samples/sum.lyng

@@ -13,7 +13,6 @@ fun findSumLimit(f) {
             println("limit reached after "+n+" rounds")
             break sum
         }
-        n++
     }
     else {
         println("limit not reached")

docs/samples/type_nullability_checks.lyng

@@ -0,0 +1,10 @@
+fun describe<T>(x: T): String = when (T) {
+    nullable -> "nullable"
+    else -> "non-null"
+}
+
+type MaybeInt = Int?
+assert(MaybeInt is nullable)
+assert(!(Int is nullable))
+assertEquals("nullable", describe<Int?>(null))
+assertEquals("non-null", describe<Int>(1))

docs/scopes_and_closures.md

@@ -1,95 +1,19 @@
-# Scopes and Closures: resolution and safety
+# Scopes and Closures: compile-time resolution
 
-This page documents how name resolution works with `ClosureScope`, how to avoid recursion pitfalls, and how to safely capture and execute callbacks that need access to outer locals.
+Attention to AI: name lookup in runtime `Scope` is legacy. The bytecode compiler uses **compile-time name/member resolution only**.
 
-## Why this matters
-Name lookup across nested scopes and closures can accidentally form recursive resolution paths or hide expected symbols (outer locals, module/global functions). The rules below ensure predictable resolution and prevent infinite recursion.
+This page documents the **current** rules: static name resolution, closure captures, and the limited role of runtime `Scope` in Kotlin interop and explicit dynamic helpers.
 
-## Resolution order in ClosureScope
-When evaluating an identifier `name` inside a closure, `ClosureScope.get(name)` resolves in this order:
+## Current rules (bytecode compiler)
+- **All names resolve at compile time**: locals, parameters, captures, members, imports, and module globals must be known when compiling. Missing symbols are compile-time errors.
+- **Exception: `compile if` can skip dead branches**: inside an untaken `compile if (...)` branch, names are not resolved or type-checked at all. This is the supported way to guard optional classes or packages such as `defined(Udp)` or `defined(lyng.io.net)`.
+- **No runtime fallbacks**: there is no dynamic name lookup, no fallback opcodes, and no “search parent scopes” at runtime for missing names.
+- **Object members on unknown types only**: `toString`, `toInspectString`, `let`, `also`, `apply`, `run` are allowed on unknown types; all other members require a statically known receiver type or an explicit cast.
+- **Closures capture slots**: lambdas and nested functions capture **frame slots** directly. Captures are resolved at compile time and compiled to slot references.
+- **Scope is a reflection facade**: `Scope` is used only for Kotlin interop or explicit dynamic helpers. It must **not** be used for general symbol resolution in compiled Lyng code.
 
-1. Closure frame locals and arguments
-2. Captured receiver (`closureScope.thisObj`) instance/class members
-3. Closure ancestry locals + each frame’s `thisObj` members (cycle‑safe)
-4. Caller `this` members
-5. Caller ancestry locals + each frame’s `thisObj` members (cycle‑safe)
-6. Module pseudo‑symbols (e.g., `__PACKAGE__`) from the nearest `ModuleScope`
-7. Direct module/global fallback (nearest `ModuleScope` and its parent/root scope)
-8. Final fallback: base local/parent lookup for the current frame
+## Explicit dynamic access (opt-in only)
+Dynamic name access is available only via explicit helpers (e.g., `dynamic { get { name -> ... } }`). It is **not** a fallback for normal member or variable access.
 
-This preserves intuitive visibility (locals → captured receiver → closure chain → caller members → caller chain → module/root) while preventing infinite recursion between scope types.
-
-## Use raw‑chain helpers for ancestry walks
-When authoring new scope types or advanced lookups, avoid calling virtual `get` while walking parents. Instead, use the non‑dispatch helpers on `Scope`:
-
-- `chainLookupIgnoreClosure(name)`
-  - Walk raw `parent` chain and check only per‑frame locals/bindings/slots.
-  - Ignores overridden `get` (e.g., in `ClosureScope`). Cycle‑safe.
-- `chainLookupWithMembers(name)`
-  - Like above, but after locals/bindings it also checks each frame’s `thisObj` members.
-  - Ignores overridden `get`. Cycle‑safe.
-- `baseGetIgnoreClosure(name)`
-  - For the current frame only: check locals/bindings, then walk raw parents (locals/bindings), then fallback to this frame’s `thisObj` members.
-
-These helpers avoid ping‑pong recursion and make structural cycles harmless (lookups terminate).
-
-## Preventing structural cycles
-- Don’t construct parent chains that can point back to a descendant.
-- A debug‑time guard throws if assigning a parent would create a cycle; keep it enabled for development builds.
-- Even with a cycle, chain helpers break out via a small `visited` set keyed by `frameId`.
-
-## Capturing lexical environments for callbacks
-For dynamic objects or custom builders, capture the creator’s lexical scope so callbacks can see outer locals/parameters:
-
-1. Use `snapshotForClosure()` on the caller scope to capture locals/bindings/slots and parent.
-2. Store this snapshot and run callbacks under `ClosureScope(callScope, captured)`.
-
-Kotlin sketch:
-```kotlin
-val captured = scope.snapshotForClosure()
-val execScope = ClosureScope(currentCallScope, captured)
-callback.execute(execScope)
-```
-
-This ensures expressions like `contractName` used inside dynamic `get { name -> ... }` resolve to outer variables defined at the creation site.
-
-## Closures in coroutines (launch/flow)
-- The closure frame still prioritizes its own locals/args.
-- Outer locals declared before suspension points remain visible through slot‑aware ancestry lookups.
-- Global functions like `delay(ms)` and `yield()` are resolved via module/root fallbacks from within closures.
-
-Tip: If a closure unexpectedly cannot see an outer local, check whether an intermediate runtime helper introduced an extra call frame; the built‑in lookup already traverses caller ancestry, so prefer the standard helpers rather than custom dispatch.
-
-## Local variable references and missing symbols
-- Unqualified identifier resolution first prefers locals/bindings/slots before falling back to `this` members.
-- If neither locals nor members contain the symbol, missing field lookups map to `SymbolNotFound` (compatibility alias for `SymbolNotDefinedException`).
-
-## Performance notes
-- The `visited` sets used for cycle detection are tiny and short‑lived; in typical scripts the overhead is negligible.
-- If profiling shows hotspots, consider limiting ancestry depth in your custom helpers or using small fixed arrays instead of hash sets—only for extremely hot code paths.
-
-## Practical Example: `cached`
-
-The `cached` function (defined in `lyng.stdlib`) is a classic example of using closures to maintain state. It wraps a builder into a zero-argument function that computes once and remembers the result:
-
-```lyng
-fun cached(builder) {
-    var calculated = false
-    var value = null
-    { // This lambda captures `calculated`, `value`, and `builder`
-        if( !calculated ) {
-            value = builder()
-            calculated = true
-        }
-        value
-    }
-}
-```
-
-Because Lyng now correctly isolates closures for each evaluation of a lambda literal, using `cached` inside a class instance works as expected: each instance maintains its own private `calculated` and `value` state, even if they share the same property declaration.
-
-## Dos and Don’ts
-- Do use `chainLookupIgnoreClosure` / `chainLookupWithMembers` for ancestry traversals.
-- Do maintain the resolution order above for predictable behavior.
-- Don’t call virtual `get` while walking parents; it risks recursion across scope types.
-- Don’t attach instance scopes to transient/pool frames; bind to a stable parent scope instead.
+## Legacy interpreter behavior (reference only)
+The old runtime `Scope`-based resolution order (locals → captured → `this` → caller → globals) is obsolete for bytecode compilation. Keep it only for legacy interpreter paths and tooling that explicitly opts into it.

docs/serialization.md

@@ -17,23 +17,40 @@ It is as simple as:
     assertEquals( text, Lynon.decode(encodedBits) )
     
     // compression was used automatically
-    assert( text.length > encodedBits.toBuffer().size )
+    assert( text.length > (encodedBits.toBuffer() as Buffer).size )
     >>> void
 
-Any class you create is serializable by default; lynon serializes first constructor fields, then any `var` member fields:
+Any class you create is serializable by default; lynon serializes first constructor fields, then any `var` member fields.
 
-    import lyng.serialization
+## Transient Fields
 
-    class Point(x,y)
+Sometimes you have fields that should not be serialized, for example, temporary caches, secret data, or derived values that are recomputed in `init` blocks. You can mark such fields with the `@Transient` attribute:
 
-    val p = Lynon.decode( Lynon.encode( Point(5,6) ) )
+```lyng
+class MyData(@Transient val tempSecret, val publicData) {
+    @Transient var cachedValue = 0
+    var persistentValue = 42
     
-    assertEquals( 5, p.x )
-    assertEquals( 6, p.y )
-    >>> void
+    init {
+        // cachedValue can be recomputed here upon deserialization
+        cachedValue = computeCache(publicData)
+    }
+}
+```
 
+Transient fields:
+- Are **omitted** from Lynon binary streams.
+- Are **omitted** from JSON output (via `toJson`).
+- Are **ignored** during structural equality checks (`==`).
+- If a transient constructor parameter has a **default value**, it will be restored to that default value during deserialization. Otherwise, it will be `null`.
+- Class body fields marked as `@Transient` will keep their initial values (or values assigned in `init`) after deserialization.
 
-just as expected.
+## Serialization of Objects and Classes
+
+- **Singleton Objects**: `object` declarations are serializable by name. Their state (mutable fields) is also serialized and restored, respecting `@Transient`.
+- **Classes**: Class objects themselves can be serialized. They are serialized by their full qualified name. When converted to JSON, a class object includes its public static fields (excluding those marked `@Transient`).
+
+## Custom Serialization
 
 Important is to understand that normally `Lynon.decode` wants [BitBuffer], as `Lynon.encode` produces. If you have the regular [Buffer], be sure to convert it:
 

docs/time.md

@@ -1,184 +1,185 @@
 # Lyng time functions
 
-Lyng date and time support requires importing `lyng.time` packages. Lyng uses simple yet modern time object models:
+Lyng date and time support requires importing `lyng.time`. The module provides four related types:
 
-- `Instant` class for time stamps with platform-dependent resolution
-- `Duration` to represent amount of time not depending on the calendar, e.g. in absolute units (milliseconds, seconds,
-  hours, days)
+- `Instant` for absolute timestamps.
+- `Date` for calendar dates without time-of-day or timezone.
+- `DateTime` for calendar-aware points in time in a specific timezone.
+- `Duration` for absolute elapsed time.
 
 ## Time instant: `Instant`
 
-Represent some moment of time not depending on the calendar (calendar for example may b e changed, daylight saving time
-can be for example introduced or dropped). It is similar to `TIMESTAMP` in SQL or `Instant` in Kotlin. Some moment of
-time; not the calendar date.
+`Instant` represents some moment of time independently of the calendar. It is similar to SQL `TIMESTAMP`
+or Kotlin `Instant`.
 
-Instant is comparable to other Instant. Subtracting instants produce `Duration`, period in time that is not dependent on
-the calendar, e.g. absolute time period.
-
-It is possible to add or subtract `Duration` to and from `Instant`, that gives another `Instant`.
-
-Instants are converted to and from `Real` number of seconds before or after Unix Epoch, 01.01.1970. Constructor with
-single number parameter constructs from such number of seconds,
-and any instance provide `.epochSeconds` member:
+### Constructing and converting
 
     import lyng.time
 
-    // default constructor returns time now:
     val t1 = Instant()
-    val t2 = Instant()
-    assert( t2 - t1 < 1.millisecond )
-    assert( t2.epochSeconds - t1.epochSeconds < 0.001 )
-    >>> void
+    val t2 = Instant(1704110400)
+    val t3 = Instant("2024-01-01T12:00:00.123456Z")
 
-## Constructing
+    val t4 = t3.truncateToMinute()
+    assertEquals("2024-01-01T12:00:00Z", t4.toRFC3339())
 
-    import lyng.time
+    val dt = t3.toDateTime("+02:00")
+    assertEquals(14, dt.hour)
 
-    // empty constructor gives current time instant using system clock:
-    val now = Instant()
+    val d = t3.toDate("Z")
+    assertEquals(Date(2024, 1, 1), d)
 
-    // constructor with Instant instance makes a copy:
-    assertEquals( now, Instant(now) )
-
-    // constructing from a number is trated as seconds since unix epoch:
-    val copyOfNow = Instant( now.epochSeconds )
-
-    // note that instant resolution is higher that Real can hold
-    // so reconstructed from real slightly differs:
-    assert( abs( (copyOfNow - now).milliseconds ) < 0.01 )
-    >>> void
-
-The resolution of system clock could be more precise and double precision real number of `Real`, keep it in mind.
-
-## Comparing and calculating periods
-
-    import lyng.time
-    
-    val now = Instant()
-    
-    // you cam add or subtract periods, and compare
-    assert( now - 5.minutes < now )
-    val oneHourAgo = now - 1.hour
-    assertEquals( now, oneHourAgo + 1.hour)
-
-    >>> void
-
-## Getting the max precision
-
-Normally, subtracting instants gives precision to microseconds, which is well inside the jitter
-the language VM adds. Still `Instant()` or `Instant.now()` capture most precise system timer at hand and provide inner
-value of 12 bytes, up to nanoseconds (hopefully). To access it use:
-
-    import lyng.time
-
-    // capture time
-    val now = Instant.now()
-
-    // this is Int value, number of whole epoch 
-    // milliseconds to the moment, it fits 8 bytes Int well
-    val seconds = now.epochWholeSeconds
-    assert(seconds is Int)
-
-    // and this is Int value of nanoseconds _since_ the epochMillis,
-    // it effectively add 4 more mytes int:
-    val nanos = now.nanosecondsOfSecond
-    assert(nanos is Int)
-    assert( nanos in 0..999_999_999 )
-
-    // we can construct epochSeconds from these parts:
-    assertEquals( now.epochSeconds, nanos * 1e-9 + seconds )
-    >>> void
-
-## Truncating to more realistic precision
-
-Full precision Instant is way too long and impractical to store, especially when serializing,
-so it is possible to truncate it to milliseconds, microseconds or seconds:
-
-    import lyng.time
-    import lyng.serialization
-
-    // max supported size (now microseconds for serialized value):
-    // note that encoding return _bit array_ and this is a _bit size_:
-    val s0 = Lynon.encode(Instant.now()).size
-
-    // shorter: milliseconds only
-    val s1 = Lynon.encode(Instant.now().truncateToMillisecond()).size
-
-    // truncated to seconds, good for file mtime, etc:
-    val s2 = Lynon.encode(Instant.now().truncateToSecond()).size 
-    assert( s1 < s0 )
-    assert( s2 < s1 )
-    >>> void
-
-## Formatting instants
-
-You can freely use `Instant` in string formatting. It supports usual sprintf-style formats:
-
-        import lyng.time
-        val now = Instant()
-
-        // will be something like "now: 12:10:05"
-        val currentTimeOnly24 =  "now: %tT"(now)
-
-        // we can extract epoch second with formatting too,
-        // this was since early C time
-
-        // get epoch while seconds from formatting
-        val unixEpoch = "Now is %ts since unix epoch"(now)
-        
-        // and it is the same as now.epochSeconds, int part:
-        assertEquals( unixEpoch, "Now is %d since unix epoch"(now.epochSeconds.toInt()) )
-        >>> void
-
-See
-the [complete list of available formats](https://github.com/sergeych/mp_stools?tab=readme-ov-file#datetime-formatting)
-and the [formatting reference](https://github.com/sergeych/mp_stools?tab=readme-ov-file#printf--sprintf): it all works
-in Lyng as `"format"(args...)`!
-
-## Instant members
+### Instant members
 
 | member                         | description                                          |
-|--------------------------------|---------------------------------------------------------|
-| epochSeconds: Real             | positive or negative offset in seconds since Unix epoch |
-| epochWholeSeconds: Int         | same, but in _whole seconds_. Slightly faster           |
-| nanosecondsOfSecond: Int       | offset from epochWholeSeconds in nanos (1)              |
-| isDistantFuture: Bool          | true if it `Instant.distantFuture`                      |
-| isDistantPast: Bool            | true if it `Instant.distantPast`                        |
-| truncateToSecond: Intant       | create new instnce truncated to second                  |  
-| truncateToMillisecond: Instant | truncate new instance with to millisecond               |
-| truncateToMicrosecond: Instant | truncate new instance to microsecond                    |
+|--------------------------------|------------------------------------------------------|
+| epochSeconds: Real             | offset in seconds since Unix epoch                   |
+| epochWholeSeconds: Int         | whole seconds since Unix epoch                       |
+| nanosecondsOfSecond: Int       | nanoseconds within the current second                |
+| isDistantFuture: Bool          | true if it is `Instant.distantFuture`                |
+| isDistantPast: Bool            | true if it is `Instant.distantPast`                  |
+| truncateToMinute(): Instant    | truncate to minute precision                         |
+| truncateToSecond(): Instant    | truncate to second precision                         |
+| truncateToMillisecond(): Instant | truncate to millisecond precision                  |
+| truncateToMicrosecond(): Instant | truncate to microsecond precision                  |
+| toRFC3339(): String            | format as RFC3339 string in UTC                      |
+| toDateTime(tz?): DateTime      | localize to a timezone                               |
+| toDate(tz?): Date              | convert to a calendar date in a timezone             |
 
-(1)
-: The value of nanoseconds is to be added to `epochWholeSeconds` to get exact time point. It is in 0..999_999_999 range.
-The precise time instant value therefore needs as for now 12 bytes integer; we might use bigint later (it is planned to
-be added)
+## Calendar date: `Date`
 
-## Class members
+`Date` represents a pure calendar date. It has no time-of-day and no attached timezone. Use it for values
+like birthdays, due dates, invoice dates, and SQL `DATE` columns.
 
-| member                         | description                                  |
-|--------------------------------|----------------------------------------------|
-| Instant.now()                  | create new instance with current system time |
-| Instant.distantPast: Instant   | most distant instant in past                 |
-| Instant.distantFuture: Instant | most distant instant in future               |
-
-# `Duraion` class
-
-Represent absolute time distance between two `Instant`.
+### Constructing
 
     import lyng.time
-    val t1 = Instant()
 
-    // yes we can delay to period, and it is not blocking. is suspends!
+    val today = Date()
+    val d1 = Date(2026, 4, 15)
+    val d2 = Date("2024-02-29")
+    val d3 = Date.parseIso("2024-02-29")
+    val d4 = Date(DateTime(2024, 5, 20, 15, 30, 45, "+02:00"))
+    val d5 = Date(Instant("2024-01-01T23:30:00Z"), "+02:00")
+
+### Date members
+
+| member                         | description                                                |
+|--------------------------------|------------------------------------------------------------|
+| year: Int                      | year component                                             |
+| month: Int                     | month component (1..12)                                    |
+| day: Int                       | day of month (alias `dayOfMonth`)                          |
+| dayOfWeek: Int                 | day of week (1=Monday, 7=Sunday)                           |
+| dayOfYear: Int                 | day of year (1..365/366)                                   |
+| isLeapYear: Bool               | whether this date is in a leap year                        |
+| lengthOfMonth: Int             | number of days in this month                               |
+| lengthOfYear: Int              | 365 or 366                                                 |
+| toIsoString(): String          | ISO `YYYY-MM-DD` string                                    |
+| toSortableString(): String     | alias to `toIsoString()`                                   |
+| toDateTime(tz="Z"): DateTime  | start-of-day `DateTime` in the specified timezone          |
+| atStartOfDay(tz="Z"): DateTime | alias to `toDateTime()`                                  |
+| addDays(n): Date               | add or subtract calendar days                              |
+| addMonths(n): Date             | add or subtract months, normalizing end-of-month           |
+| addYears(n): Date              | add or subtract years                                      |
+| daysUntil(other): Int          | calendar days until `other`                                |
+| daysSince(other): Int          | calendar days since `other`                                |
+| static today(tz?): Date        | today in the specified timezone                            |
+| static parseIso(s): Date       | parse ISO `YYYY-MM-DD`                                     |
+
+### Date arithmetic
+
+`Date` supports only whole-day arithmetic. This is deliberate: calendar dates should not silently accept
+sub-day durations.
+
+    import lyng.time
+
+    val d1 = Date(2026, 4, 15)
+    val d2 = d1.addDays(10)
+
+    assertEquals(Date(2026, 4, 25), d2)
+    assertEquals(Date(2026, 4, 18), d1 + 3.days)
+    assertEquals(Date(2026, 4, 12), d1 - 3.days)
+    assertEquals(10, d1.daysUntil(d2))
+    assertEquals(10, d2.daysSince(d1))
+    assertEquals(10, d2 - d1)
+
+### Date conversions
+
+    import lyng.time
+
+    val i = Instant("2024-01-01T23:30:00Z")
+    assertEquals(Date(2024, 1, 1), i.toDate("Z"))
+    assertEquals(Date(2024, 1, 2), i.toDate("+02:00"))
+
+    val dt = DateTime(2024, 5, 20, 15, 30, 45, "+02:00")
+    assertEquals(Date(2024, 5, 20), dt.date)
+    assertEquals(Date(2024, 5, 20), dt.toDate())
+    assertEquals(DateTime(2024, 5, 20, 0, 0, 0, "Z"), Date(2024, 5, 20).toDateTime("Z"))
+    assertEquals(DateTime(2024, 5, 20, 0, 0, 0, "+02:00"), Date(2024, 5, 20).atStartOfDay("+02:00"))
+
+## Calendar time: `DateTime`
+
+`DateTime` represents a point in time in a specific timezone. It provides access to calendar components
+such as year, month, day, and hour.
+
+### Constructing
+
+    import lyng.time
+
+    val now = DateTime.now()
+    val offsetTime = DateTime.now("+02:00")
+    val dt = Instant().toDateTime("Z")
+    val dt2 = DateTime(2024, 1, 1, 12, 0, 0, "Z")
+    val dt3 = DateTime.parseRFC3339("2024-01-01T12:00:00+02:00")
+
+### DateTime members
+
+| member                           | description                                   |
+|----------------------------------|-----------------------------------------------|
+| year: Int                        | year component                                |
+| month: Int                       | month component (1..12)                       |
+| day: Int                         | day of month (alias `dayOfMonth`)             |
+| hour: Int                        | hour component (0..23)                        |
+| minute: Int                      | minute component (0..59)                      |
+| second: Int                      | second component (0..59)                      |
+| dayOfWeek: Int                   | day of week (1=Monday, 7=Sunday)              |
+| timeZone: String                 | timezone ID string                            |
+| date: Date                       | calendar date component                       |
+| toInstant(): Instant             | convert back to absolute Instant              |
+| toDate(): Date                   | extract the calendar date in this timezone    |
+| toUTC(): DateTime                | shortcut to convert to UTC                    |
+| toTimeZone(tz): DateTime         | convert to another timezone                   |
+| addMonths(n): DateTime           | add/subtract months (normalizes end of month) |
+| addYears(n): DateTime            | add/subtract years                            |
+| toRFC3339(): String              | format with timezone offset                   |
+| static now(tz?): DateTime        | create DateTime with current time             |
+| static parseRFC3339(s): DateTime | parse RFC3339 string                          |
+
+### Arithmetic and normalization
+
+`DateTime` handles calendar arithmetic correctly:
+
+    import lyng.time
+
+    val leapDay = Instant("2024-02-29T12:00:00Z").toDateTime("Z")
+    val nextYear = leapDay.addYears(1)
+    assertEquals(28, nextYear.day)
+
+# `Duration` class
+
+`Duration` represents absolute elapsed time between two instants.
+
+    import lyng.time
+
+    val t1 = Instant()
     delay(1.millisecond)
-
     val t2 = Instant()
-    // be suspend, so actual time may vary:
+
     assert(t2 - t1 >= 1.millisecond)
     assert(t2 - t1 < 100.millisecond)
     >>> void
 
-Duration can be converted from numbers, like `5.minutes` and so on. Extensions are created for
-`Int` and `Real`, so for n as Real or Int it is possible to create durations::
+Duration values can be created from numbers using extensions on `Int` and `Real`:
 
 - `n.millisecond`, `n.milliseconds`
 - `n.second`, `n.seconds`
@@ -186,10 +187,9 @@ Duration can be converted from numbers, like `5.minutes` and so on. Extensions a
 - `n.hour`, `n.hours`
 - `n.day`, `n.days`
 
-The bigger time units like months or years are calendar-dependent and can't be used with `Duration`.
+Larger units like months or years are calendar-dependent and are intentionally not part of `Duration`.
 
-Each duration instance can be converted to number of any of these time units, as `Real` number, if `d` is a `Duration`
-instance:
+Each duration instance can be converted to numbers in these units:
 
 - `d.microseconds`
 - `d.milliseconds`
@@ -198,18 +198,16 @@ instance:
 - `d.hours`
 - `d.days`
 
-for example
+Example:
 
     import lyng.time
+
     assertEquals(60, 1.minute.seconds)
     assertEquals(10.milliseconds, 0.01.seconds)
-
     >>> void
 
 # Utility functions
 
-## delay(duration: Duration)
-
-Suspends current coroutine for at least the specified duration.
-
+## `delay(duration: Duration)`
 
+Suspends the current coroutine for at least the specified duration.

docs/tutorial.md

@@ -8,12 +8,13 @@ __Other documents to read__ maybe after this one:
 
 - [Advanced topics](advanced_topics.md), [declaring arguments](declaring_arguments.md), [Scopes and Closures](scopes_and_closures.md)
 - [OOP notes](OOP.md), [exception handling](exceptions_handling.md)
-- [math in Lyng](math.md), [the `when` statement](when.md)
+- [math in Lyng](math.md), [the `when` statement](when.md), [return statement](return_statement.md)
 - [Testing and Assertions](Testing.md)
+- [Generics and type expressions](generics.md)
 - [time](time.md) and [parallelism](parallelism.md)
 - [parallelism] - multithreaded code, coroutines, etc.
 - Some class
-  references: [List], [Set], [Map], [Real], [Range], [Iterable], [Iterator], [time manipulation](time.md), [Array], [RingBuffer], [Buffer].
+  references: [List], [ImmutableList], [Set], [ImmutableSet], [Map], [ImmutableMap], [Real], [Range], [Iterable], [Iterator], [time manipulation](time.md), [Array], [RingBuffer], [Buffer].
 - Some samples: [combinatorics](samples/combinatorics.lyng.md), national vars and
   loops: [сумма ряда](samples/сумма_ряда.lyng.md). More at [samples folder](samples)
 
@@ -32,6 +33,15 @@ any block also returns it's last expression:
     }
     >>> 6
 
+If you want to exit a function or lambda earlier, use the `return` statement:
+
+    fn divide(a, b) {
+        if( b == 0 ) return null
+        a / b
+    }
+
+See [return statement](return_statement.md) for more details on scoping and non-local returns.
+
 If you don't want block to return anything, use `void`:
 
     fn voidFunction() {
@@ -97,6 +107,23 @@ Singleton objects are declared using the `object` keyword. They define a class a
     
     Logger.log("Hello singleton!")
 
+## Nested Declarations (short)
+
+Classes, objects, and enums can be declared inside another class. They live in the class namespace (no outer instance capture), so you access them with a qualifier:
+
+    class A {
+        class B(x?)
+        object Inner { val foo = "bar" }
+        enum E* { One, Two }
+    }
+
+    val ab = A.B()
+    assertEquals(ab.x, null)
+    assertEquals(A.Inner.foo, "bar")
+    assertEquals(A.One, A.E.One)
+
+See [OOP notes](OOP.md#nested-declarations) for rules, visibility, and enum lifting details.
+
 ## Delegation (briefly)
 
 You can delegate properties and functions to other objects using the `by` keyword. This is perfect for patterns like `lazy` initialization.
@@ -179,14 +206,13 @@ Note that assignment operator returns rvalue, it can't be assigned.
 ## Modifying arithmetics
 
 There is a set of assigning operations: `+=`, `-=`, `*=`, `/=` and even `%=`.
+There is also a special null-aware assignment operator `?=`: it performs the assignment only if the lvalue is `null`.
 
-    var x = 5
-    assert( 25 == (x*=5) )
-    assert( 25 == x)
-    assert( 24 == (x-=1) )
-    assert( 12 == (x/=2) )
-    x
-    >>> 12
+    var x = null
+    x ?= 10
+    assertEquals(10, x)
+    x ?= 20
+    assertEquals(10, x)
 
 Notice the parentheses here: the assignment has low priority!
 
@@ -203,9 +229,8 @@ Naturally, assignment returns its value:
 rvalue means you cant assign the result if the assignment
 
     var x
-    assertThrows { (x = 11) = 5 }
-    void
-    >>> void
+    // compile-time error: can't assign to rvalue
+    (x = 11) = 5
 
 This also prevents chain assignments so use parentheses:
 
@@ -216,29 +241,48 @@ This also prevents chain assignments so use parentheses:
 
 ## Nullability
 
+Nullability is part of the type. `String` is non-null, `String?` is nullable. Use `!!` to assert non-null and throw
+`NullReferenceException` if the value is `null`.
+
 When the value is `null`, it might throws `NullReferenceException`, the name is somewhat a tradition. To avoid it
 one can check it against null or use _null coalescing_. The null coalescing means, if the operand (left) is null,
 the operation won't be performed and the result will be null. Here is the difference:
 
-    val ref = null
-    assertThrows {  ref.field }
-    assertThrows {  ref.method() }
-    assertThrows {  ref.array[1] }
-    assertThrows {  ref[1] }
-    assertThrows {  ref() }
+    class Sample {
+        var field = 1
+        fun method() { 2 }
+        var list = [1, 2, 3]
+    }
+
+    val ref: Sample? = null
+    val list: List<Int>? = null
+    // direct access throws NullReferenceException:
+    // ref.field
+    // ref.method()
+    // ref.list[1]
+    // list[1]
 
     assert( ref?.field == null )
     assert( ref?.method() == null )
-    assert( ref?.array?[1] == null )
-    assert( ref?[1] == null )
-    assert( ref?() == null )
+    assert( ref?.list?[1] == null )
+    assert( list?[1] == null )
     >>> void
 
+Note: `?.` is still a typed operation. The receiver must have a compile-time type that declares the member; if the
+receiver is `Object`, cast it first or declare a more specific type.
+
 There is also "elvis operator", null-coalesce infix operator '?:' that returns rvalue if lvalue is `null`:
 
     null ?: "nothing"
     >>> "nothing"
 
+There is also a null-aware assignment operator `?=`, which assigns a value only if the target is `null`:
+
+    var config = null
+    config ?= { port: 8080 }
+    config ?= { port: 9000 } // no-op, config is already not null
+    assertEquals(8080, config.port)
+
 ## Utility functions
 
 The following functions simplify nullable values processing and
@@ -283,8 +327,8 @@ Much like let, but it does not alter returned value:
 
 While it is not altering return value, the source object could be changed:
 also
-    class Point(x,y)
-    val p = Point(1,2).also { it.x++ }
+    class Point(var x: Int, var y: Int)
+    val p: Point = Point(1,2).also { it.x++ }
     assertEquals(p.x, 2)
     >>> void
 
@@ -292,12 +336,22 @@ also
 
 It works much like `also`, but is executed in the context of the source object:
 
-    class Point(x,y)
+    class Point(var x: Int, var y: Int)
     // see the difference: apply changes this to newly created Point:
-    val p = Point(1,2).apply { x++; y++ }
+    val p = Point(1,2).apply { this@Point.x++; this@Point.y++ }
     assertEquals(p, Point(2,3))
     >>> void
 
+## with
+
+Sets `this` to the first argument and executes the block. Returns the value returned by the block:
+
+    class Point(var x: Int, var y: Int)
+    val p = Point(1,2)
+    val sum = with(p) { x + y }
+    assertEquals(3, sum)
+    >>> void
+
 ## run
 
 Executes a block after it returning the value passed by the block. for example, can be used with elvis operator:
@@ -321,6 +375,18 @@ It is rather simple, like everywhere else:
 
 See [math](math.md) for more on it. Notice using Greek as identifier, all languages are allowed.
 
+For linear algebra, import `lyng.matrix`:
+
+    import lyng.matrix
+
+    val a: Matrix = matrix([[1, 2], [3, 4]])
+    val i: Matrix = Matrix.identity(2)
+    val sum: Matrix = a + i
+    assertEquals([[2.0, 2.0], [3.0, 5.0]], sum.toList())
+    >>> void
+
+See [Matrix](Matrix.md) for vectors, matrix multiplication, inversion, and slicing such as `m[0..2, 1]`.
+
 Logical operation could be used the same
 
     var x = 10
@@ -400,8 +466,6 @@ Almost the same, using `val`:
     val foo = 1
     foo += 1 // this will throw exception
 
-# Constants
-
 Same as in kotlin:
 
     val HalfPi = π / 2
@@ -409,6 +473,163 @@ Same as in kotlin:
 Note using greek characters in identifiers! All letters allowed, but remember who might try to read your script, most
 likely will know some English, the rest is the pure uncertainty.
 
+# Types and inference
+
+Lyng uses Kotlin-style static types with inference. You can always write explicit types, but in most places the compiler
+can infer them from literals, defaults, and flow analysis.
+
+## Type annotations
+
+Use `:` to specify a type:
+
+    var x: Int = 10
+    val label: String = "count"
+    fun clamp(x: Int, min: Int, max: Int): Int { ... }
+
+`Object` is the top type. If you omit a type and there is no default value, the parameter is `Object` by default:
+
+    fun show(x) { println(x) }   // x is Object
+
+For nullable types, add `?`:
+
+    fun showMaybe(x: Object?) { ... }
+    fun parseInt(s: String?): Int? { ... }
+
+There is also a nullable shorthand for untyped parameters and constructor args: `x?` means `x: Object?`.
+It cannot be combined with an explicit type annotation.
+
+    class A(x?) { ... }     // x: Object?
+    fun f(x?) { x == null } // x: Object?
+
+Type aliases name type expressions and can be generic:
+
+    type Num = Int | Real
+    type Maybe<T> = T?
+
+Aliases expand to their underlying type expressions. See `docs/generics.md` for details.
+
+`void` is a singleton value of the class `Void`. `Void` can be used as an explicit return type:
+
+    fun log(msg): Void { println(msg); void }
+
+`Null` is the class of `null`. It is a singleton type and mostly useful for type inference results.
+
+For type expressions, you can check nullability directly:
+
+    T is nullable
+    T !is nullable
+
+This is especially useful in generic code and in `when` over a type parameter:
+
+    fun describe<T>(x: T): String = when (T) {
+        nullable -> "nullable"
+        else -> "non-null"
+    }
+
+## Type inference
+
+The compiler infers types from:
+
+- literals: `1` is `Int`, `1.0` is `Real`, `"s"` is `String`, `'c'` is `Char`
+- defaults: `fun f(x=1, name="n")` infers `x: Int`, `name: String`
+- assignments: `val x = call()` uses the return type of `call`
+- returns and branches: the result type of a block is the last expression, and if any branch is nullable,
+  the inferred type becomes nullable
+- numeric ops: `Int` and `Real` stay `Int` when both sides are `Int`, and promote to `Real` on mixed arithmetic
+
+Examples:
+
+    fun inc(x=0) = x + 1        // (Int)->Int
+    fun maybe(flag) { if(flag) 1 else null } // ()->Int?
+
+Function types are written as `(T1, T2, ...)->R`. You can include ellipsis in function *types* to
+express a variadic position:
+
+    var fmt: (String, Object...)->String
+    var f: (Int, Object..., String)->Real
+    var anyArgs: (...)->Int   // shorthand for (Object...)->Int
+
+Untyped locals are allowed, but their type is fixed on the first assignment:
+
+    var x
+    x = 1       // x becomes Int
+    x = "one"   // compile-time error
+
+    var y = null  // y is Object?
+    val z = null  // z is Null
+
+Empty list/map literals default to `List<Object>` and `Map<Object,Object>` until a more specific type is known:
+
+    val xs = []             // List<Object>
+    val ys: List<Int> = []  // List<Int>
+
+Map literals infer key/value types from entries; named keys are `String`. See `docs/generics.md` for details.
+
+## Flow analysis
+
+Lyng uses flow analysis to narrow types inside branches:
+
+    fun len(x: String?): Int {
+        if( x == null ) return 0
+        // x is String (non-null) in this branch
+        return x.length
+    }
+
+`is` checks and `when` branches also narrow types:
+
+    fun kind(x: Object) {
+        if( x is Int ) return "int"
+        if( x is String ) return "string"
+        return "other"
+    }
+
+Narrowing is local to the branch; after the branch, the original type is restored.
+
+## Casts and unknown types
+
+Use `as` for explicit casts. The compiler inserts casts only when it can be valid and necessary. If a cast fails at
+runtime, it throws `ClassCastException`. If the value is nullable, `as T` implies a non-null assertion.
+
+Member access is resolved at compile time. Only members of `Object` are available on unknown types; non-Object members
+require an explicit cast:
+
+    fun f(x) {             // x is Object
+        x.toString()       // ok (Object member)
+        x.size()           // compile-time error
+        (x as List).size() // ok
+    }
+
+This avoids runtime name-resolution fallbacks; all symbols must be known at compile time.
+
+## Generics and bounds
+
+Generic parameters are declared with `<...>`:
+
+    fun id<T>(x: T): T = x
+    class Box<T>(val value: T)
+
+Bounds use `:` and can combine with `&` (intersection) and `|` (union):
+
+    fun sum<T: Int | Real>(x: T, y: T) = x + y
+    class Named<T: Iterable & Comparable>(val data: T)
+
+Type arguments are usually inferred from call sites:
+
+    val b = Box(10)     // Box<Int>
+    val s = id("ok")    // T is String
+
+See [Generics and type expressions](generics.md) for bounds, unions/intersections, and type-checking rules.
+
+## Variance
+
+Generic types are invariant by default, so `List<Int>` is not a `List<Object>`.
+Use declaration-site variance when you need it:
+
+    class Source<out T>(val value: T)
+    class Sink<in T> { fun accept(x: T) { ... } }
+
+`out` makes the type covariant (only produced), `in` makes it contravariant (only consumed).
+
 # Defining functions
 
     fun check(amount) {
@@ -450,8 +671,9 @@ There are default parameters in Lyng:
 It is possible to define also vararg using ellipsis:
 
     fun sum(args...) {
-        var result = args[0]
-        for( i in 1 ..< args.size ) result += args[i]
+        val list = args as List
+        var result = list[0]
+        for( i in 1 ..< list.size ) result += list[i]
     }
     sum(10,20,30)
     >>> 60
@@ -544,6 +766,11 @@ one could be with ellipsis that means "the rest pf arguments as List":
     assert( { a, b...-> [a,...b] }(100, 1, 2, 3) == [100, 1, 2, 3]) 
     void
 
+Type-annotated lambdas can use variadic *function types* as well:
+
+    val f: (Int, Object..., String)->Real = { a, rest..., b -> 0.0 }
+    val anyArgs: (...)->Int = { -> 0 }
+
 ### Using lambda as the parameter
 
 See also: [Testing and Assertions](Testing.md)
@@ -554,6 +781,7 @@ See also: [Testing and Assertions](Testing.md)
         var result = []
         for( x in iterable ) result += transform(x)
     }
+    // loop variables are read-only inside the loop body
     assert( [11, 21, 31] == mapValues( [1,2,3], { it*10+1 }))
     >>> void
 
@@ -581,6 +809,13 @@ Lyng has built-in mutable array class `List` with simple literals:
 
 [List] is an implementation of the type `Array`, and through it `Collection` and [Iterable]. Please read [Iterable],
 many collection based methods are implemented there.
+For immutable list values, use `list.toImmutable()` and [ImmutableList].
+
+To construct a list programmatically, use the static helper `List.fill`:
+
+    val tens = List.fill(5) { index -> index * 10 }
+    assertEquals([0, 10, 20, 30, 40], tens)
+    >>> void
 
 Lists can contain any type of objects, lists too:
 
@@ -589,11 +824,19 @@ Lists can contain any type of objects, lists too:
     assert( list is Array )     // general interface
     assert(list.size == 3)
     // second element is a list too:
-    assert(list[1].size == 2)
+    assert((list[1] as List).size == 2)
     >>> void
 
 Notice usage of indexing. You can use negative indexes to offset from the end of the list; see more in [Lists](List.md).
 
+In general, bracket indexing may contain more than one selector:
+
+    value[i]
+    value[i, j]
+
+For built-in lists, strings, maps, and buffers, the selector is usually a single value such as an `Int`, `Range`, or `Regex`.
+For types with custom indexers, multiple selectors are packed into one list-like index object and passed to `getAt` / `putAt`.
+
 When you want to "flatten" it to single array, you can use splat syntax:
 
     [1, ...[2,3], 4]
@@ -763,6 +1006,7 @@ Set are unordered collection of unique elements, see [Set]. Sets are [Iterable]
     >>> void
 
 Please see [Set] for detailed description.
+For immutable set values, use `set.toImmutable()` and [ImmutableSet].
 
 # Maps
 
@@ -823,6 +1067,7 @@ Notes:
 - When you need computed (expression) keys or non-string keys, use `Map(...)` constructor with entries, e.g. `Map( ("a" + "b") => 1 )`, then merge with a literal if needed: `{ base: } + (computedKey => 42)`.
 
 Please see the [Map] reference for a deeper guide.
+For immutable map values, use `map.toImmutable()` and [ImmutableMap].
 
 # Flow control operators
 
@@ -849,6 +1094,37 @@ Or, more neat:
     >>> just 3
     >>> void
 
+## compile if
+
+`compile if` is a compile-time conditional. Unlike normal `if`, the compiler evaluates its condition while compiling
+the file and completely skips the untaken branch. This is useful when some class or package may or may not be
+available:
+
+    compile if (defined(Udp)) {
+        val socket = Udp()
+        println("udp is available")
+    } else {
+        println("udp is not available")
+    }
+
+`compile if` also supports single-statement branches:
+
+    compile if (defined(lyng.io.net) && !defined(Udp))
+        println("network module exists, but Udp is not visible here")
+    else
+        println("either Udp exists or the module is unavailable")
+
+Current condition syntax is intentionally limited to compile-time symbol checks:
+
+- `defined(Name)`
+- `defined(package.name)`
+- `!`, `&&`, `||`
+- parentheses
+
+Examples:
+    compile if (defined(Udp) && defined(Tcp))
+        println("both transports are available")
+
 ## When
 
 See also: [Comprehensive guide to `when`](when.md)
@@ -1037,8 +1313,8 @@ ends normally, without breaks. It allows override loop result value, for example
 to not calculate it in every iteration. For example, consider this naive prime number
 test function (remember function return it's last expression result):
 
-    fun naive_is_prime(candidate) {
-        val x = if( candidate !is Int) candidate.toInt() else candidate
+    fun naive_is_prime(candidate: Int) {
+        val x = candidate
         var divisor = 1
         while( ++divisor < x/2 || divisor == 2 ) {
             if( x % divisor == 0 ) break false
@@ -1113,12 +1389,48 @@ For loop are intended to traverse collections, and all other objects that suppor
 size and index access, like lists:
 
     var letters = 0
-    for( w in ["hello", "wolrd"]) {
-        letters += w.length
+    val words: List<String> = ["hello", "world"]
+    for( w in words) {
+        letters += (w as String).length
     }
     "total letters: "+letters
     >>> "total letters: 10"
 
+When you need a counting loop that goes backwards, use an explicit descending
+range:
+
+    var sum = 0
+    for( i in 5 downTo 1 ) {
+        sum += i
+    }
+    sum
+    >>> 15
+
+If the lower bound should be excluded, use `downUntil`:
+
+    val xs = []
+    for( i in 5 downUntil 1 ) {
+        xs.add(i)
+    }
+    xs
+    >>> [5,4,3,2]
+
+This is intentionally explicit: `5..1` is an empty ascending range, not an
+implicit reverse loop.
+
+Descending loops also support `step`:
+
+    val xs = []
+    for( i in 10 downTo 1 step 3 ) {
+        xs.add(i)
+    }
+    xs
+    >>> [10,7,4,1]
+
+For descending ranges, `step` stays positive. The direction comes from
+`downTo` / `downUntil`, so `10 downTo 1 step 3` is valid, while
+`10 downTo 1 step -3` is an error.
+
 For loop support breaks the same as while loops above:
 
     fun search(haystack, needle) {    
@@ -1227,7 +1539,7 @@ The same with `--`:
     sum
     >>> 5050
 
-There are self-assigning version for operators too:
+There is a self-assigning version for operators too:
 
     var count = 100
     var sum = 0
@@ -1248,6 +1560,12 @@ It could be open and closed:
     assert( 5 !in (1..<5) )
     >>> void
 
+Descending ranges are explicit too:
+
+    assertEquals([5,4,3,2,1], (5 downTo 1).toList())
+    assertEquals([5,4,3,2], (5 downUntil 1).toList())
+    >>> void
+
 Ranges could be inside other ranges:
 
     assert( (2..3) in (1..10) )
@@ -1265,6 +1583,14 @@ and you can use ranges in for-loops:
     >>> b
     >>> void
 
+Descending character ranges work the same way:
+
+    for( ch in 'e' downTo 'a' step 2 ) println(ch)
+    >>> e
+    >>> c
+    >>> a
+    >>> void
+
 See [Ranges](Range.md) for detailed documentation on it.
 
 # Time routines
@@ -1305,7 +1631,7 @@ than enum arrays, until `Lynon.encodeTyped` will be implemented.
     var result = null // here we will store the result
     >>> null
 
-# Integral data types
+# Built-in types
 
 | type   | description                     | literal samples     |
 |--------|---------------------------------|---------------------|
@@ -1315,6 +1641,7 @@ than enum arrays, until `Lynon.encodeTyped` will be implemented.
 | Char   | single unicode character        | `'S'`, `'\n'`       |
 | String | unicode string, no limits       | "hello" (see below) |
 | List   | mutable list                    | [1, "two", 3]       |
+| Object | top type for all values         |                     |
 | Void   | no value could exist, singleton | void                |
 | Null   | missing value, singleton        | null                |
 | Fn     | callable type                   |                     |
@@ -1327,28 +1654,94 @@ The type for the character objects is `Char`.
 
 ### String literal escapes
 
+Lyng string literals can use either double quotes or backticks:
+
+    val a = "hello"
+    val b = `hello`
+    assert(a == b)
+
 | escape | ASCII value           |
 |--------|-----------------------|
 | \n     | 0x10, newline         |
 | \r     | 0x13, carriage return |
 | \t     | 0x07, tabulation      |
 | \\     | \ slash character     |
-| \"     | " double quote        |
+| \uXXXX | unicode code point    |
+
+Delimiter-specific escapes:
+
+| form   | escape | value            |
+|--------|--------|------------------|
+| `"..."` | \"    | " double quote   |
+| `` `...` `` | \` | ` backtick       |
+
+Unicode escape form is exactly 4 hex digits, e.g. `"\u263A"` -> `☺`.
 
 Other `\c` combinations, where c is any char except mentioned above, are left intact, e.g.:
 
     val s = "\a"
-    assert(s[0] == '\')
+    assert(s[0] == '\\')
     assert(s[1] == 'a')
     >>> void
 
 same as:
 
     val s = "\\a"
-    assert(s[0] == '\')
+    assert(s[0] == '\\')
     assert(s[1] == 'a')
     >>> void
 
+### String interpolation
+
+Supported forms:
+
+- `$name`
+- `${expr}`
+
+Literal dollar forms:
+
+- `\$` -> `$`
+- `$$` -> `$`
+
+Example:
+
+    val name = "Lyng"
+    assertEquals("hello, Lyng!", "hello, $name!")
+    assertEquals("hello, Lyng!", `hello, $name!`)
+    assertEquals("sum=3", "sum=${1+2}")
+    assertEquals("sum=3", `sum=${1+2}`)
+    assertEquals("\$name", "\$name")
+    assertEquals("\$name", "$$name")
+    assertEquals("\$name", `\$name`)
+    assertEquals("\$name", `$$name`)
+    assertEquals("\\Lyng", "\\$name")
+    assertEquals("\\Lyng", `\\$name`)
+    >>> void
+
+Interpolation and `printf`-style formatting can be combined when needed:
+
+    val method = "transfer"
+    val argc = 2
+    val compact = "%s:%d"(method, argc)
+    assertEquals("call=transfer:2", "call=$compact")
+    assertEquals("[transfer:2] ok", "[${"%s:%d"(method, argc)}] ok")
+    >>> void
+
+Interpolation also works well with regex patterns. To keep a literal `$` in the
+regex, escape it in the resulting pattern:
+
+    val currency = "USD"
+    val amount = 15
+    val escapedDollar = "\\$"
+    val re = Regex("^${currency}${escapedDollar}${amount}$")
+    assert("USD$15" =~ re)
+    assert("USD15" !~ re)
+    >>> void
+
+If you need old literal behavior in a file, add a leading directive comment:
+
+    // feature: interpolation: off
+
 ### Char literal escapes
 
 Are the same as in string literals with little difference:
@@ -1360,6 +1753,9 @@ Are the same as in string literals with little difference:
 | \t     | 0x07, tabulation      |
 | \\     | \ slash character     |
 | \'     | ' apostrophe          |
+| \uXXXX | unicode code point    |
+
+For char literals, use `'\\'` to represent a single backslash character; `'\'` is invalid.
 
 ### Char instance members
 
@@ -1426,6 +1822,14 @@ Open-ended ranges could be used to get start and end too:
     assertEquals( "pult", "catapult"[ 4.. ])
     >>> void
 
+The same bracket syntax is also used by imported numeric modules such as `lyng.matrix`, where indexing can be multi-axis:
+
+    import lyng.matrix
+
+    val m: Matrix = matrix([[1, 2, 3], [4, 5, 6]])
+    assertEquals(6.0, m[1, 2])
+    >>> void
+
 ### String operations
 
 Concatenation is a `+`: `"hello " + name` works as expected. No confusion. There is also
@@ -1437,22 +1841,39 @@ Concatenation is a `+`: `"hello " + name` works as expected. No confusion. There
 
 Extraction:
 
-    "abcd42def"[ "\d+".re ].value
+    ("abcd42def"[ "\d+".re ] as RegexMatch).value
     >>> "42"
 
 Part match:
 
     assert( "abc foo def" =~ "f[oO]+".re )
-    assert( "foo" == $~.value )
+    assert( "foo" == ($~ as RegexMatch).value )
     >>> void
 
-Typical set of String functions includes:
+Replacing text:
+
+    assertEquals("bonono", "banana".replace('a', 'o'))
+    assertEquals("a-b-c", "a.b.c".replace(".", "-"))   // string patterns are literal
+    assertEquals("v#.#.#", "v1.2.3".replace("\d+".re, "#"))
+    assertEquals("v[1].[2].[3]", "v1.2.3".replace("(\d+)".re) { m -> "[" + m[1] + "]" })
+    >>> void
+
+Repeating the fragment:
+
+    assertEquals("hellohello", "hello"*2)
+    assertEquals("", "hello"*0)
+    >>> void
+
+A typical set of String functions includes:
 
 | fun/prop             | description / notes                                        |
 |----------------------|------------------------------------------------------------|
 | lower(), lowercase() | change case to unicode upper                               |
 | upper(), uppercase() | change case to unicode lower                               |
 | trim()               | trim space chars from both ends                            |
+| isEmpty()            | true if string is empty                                    |
+| isNotEmpty()         | true if string is not empty                                |
+| isBlank()            | true if empty or contains only whitespace                  |
 | startsWith(prefix)   | true if starts with a prefix                               |
 | endsWith(prefix)     | true if ends with a prefix                                 |
 | last()               | get last character of a string or throw                    |
@@ -1469,9 +1890,11 @@ Typical set of String functions includes:
 | s1 += s2             | self-modifying concatenation                               |
 | toReal()             | attempts to parse string as a Real value                   |
 | toInt()              | parse string to Int value                                  |
-| characters()         | create [List] of characters (1)                            |
+| characters           | create [List] of characters (1)                            |
 | encodeUtf8()         | returns [Buffer] with characters encoded to utf8           |
 | matches(re)          | matches the regular expression (2)                         |
+| replace(old, new)    | replace all literal or regex matches; regex needs [Regex]  |
+| replaceFirst(old,new)| replace the first literal or regex match                   |
 |                      |                                                            |
 
 (1)
@@ -1519,7 +1942,7 @@ See [math functions](math.md). Other general purpose functions are:
 | flow {}                               | create flow sequence, see [parallelism]                    |
 | delay, launch, yield                  | see [parallelism]                                          |
 | cached(builder)                       | [Lazy evaluation with `cached`](#lazy-evaluation-with-cached) |
-| let, also, apply, run                 | see above, flow controls                                   |
+| let, also, apply, run, with       | see above, flow controls                                   |
 
 (1)
 : `fn` is optional lambda returning string message to add to exception string. 
@@ -1535,6 +1958,7 @@ Lambda avoid unnecessary execution if assertion is not failed. for example:
 | π                                   | See [math](math.md)          |
 
 [List]: List.md
+[ImmutableList]: ImmutableList.md
 
 [Testing]: Testing.md
 
@@ -1546,13 +1970,15 @@ Lambda avoid unnecessary execution if assertion is not failed. for example:
 
 [Range]: Range.md
 
-[String]: development/String.md
+[String]: ../archived/development/String.md
 
 [string formatting]: https://github.com/sergeych/mp_stools?tab=readme-ov-file#sprintf-syntax-summary
 
 [Set]: Set.md
+[ImmutableSet]: ImmutableSet.md
 
 [Map]: Map.md
+[ImmutableMap]: ImmutableMap.md
 
 [Buffer]: Buffer.md
 
@@ -1670,7 +2096,7 @@ assertEquals(null, (buzz as? Foo)?.runA())
 
 Notes:
 - Resolution order uses C3 MRO (active): deterministic, monotonic order suitable for diamonds and complex hierarchies. Example: for `class D() : B(), C()` where both `B()` and `C()` derive from `A()`, the C3 order is `D → B → C → A`. The first visible match wins.
-- `private` is visible only inside the declaring class; `protected` is visible from the declaring class and any of its transitive subclasses. Qualialsofication (`this@Type`) or casts do not bypass visibility.
+- `private` is visible only inside the declaring class; `protected` is visible from the declaring class and its subclasses. Additionally, ancestors can access protected members of descendants if they override a member known to the ancestor. Qualification (`this@Type`) or casts do not bypass visibility.
 - Safe‑call `?.` works with `as?` for optional dispatch.
 
 ## Extension members
@@ -1685,7 +2111,7 @@ You can add new methods and properties to existing classes without modifying the
 
 ### Extension properties
 
-    val Int.isEven = this % 2 == 0
+    val Int.isEven get() = this % 2 == 0
     4.isEven
     >>> true
 
@@ -1695,6 +2121,30 @@ Example with custom accessors:
     "abc".firstChar
     >>> 'a'
 
+### Extension indexers
+
+Indexers can also be extended by overriding `getAt` and `putAt` on the receiver:
+
+```lyng
+object Storage
+
+var storageData = {}
+
+override fun Storage.getAt(key: String): Object? {
+    storageData[key]
+}
+
+override fun Storage.putAt(key: String, value: Object) {
+    storageData[key] = value
+}
+
+Storage["answer"] = 42
+val answer: Int? = Storage["answer"]
+assertEquals(42, answer)
+```
+
+This works for classes and named singleton `object` declarations. Bracket syntax is lowered to `getAt` / `putAt`, and multiple selectors are packed into one list-like index object the same way as other custom indexers.
+
 Extension members are **scope-isolated**: they are visible only in the scope where they are defined and its children. This prevents name collisions and improves security.
 
 To get details on OOP in Lyng, see [OOP notes](OOP.md).

docs/whats_new.md

@@ -1,9 +1,235 @@
 # What's New in Lyng
 
-This document highlights the latest additions and improvements to the Lyng language and its ecosystem.
+This document highlights the current Lyng release, **1.5.5**, and the broader additions from the 1.5 cycle.
+It is intentionally user-facing: new language features, new modules, new tools, and the practical things you can build with them.
+For a programmer-focused migration summary across 1.5.x, see `docs/whats_new_1_5.md`.
+
+## Release 1.5.5 Highlights
+
+- `1.5.5` extends the 1.5 line with practical database APIs, first-class calendar dates, and better coroutine building blocks.
+- The 1.5 line now brings together richer ranges and loops, interpolation, math modules, immutable and observable collections, richer `lyngio`, and much better CLI/IDE support.
+- `1.5.5` adds `Channel`, `LaunchPool`, and `joinAll()` so coroutine-heavy scripts can coordinate work more directly.
+- `1.5.5` adds `Date`, the portable `lyng.io.db` layer, SQLite/JDBC providers, and a compatibility `lyng.legacy_digest` module.
+- `1.5.5` also continues runtime/compiler hardening with better import dispatch, faster exact lambda calls, and correct `val +=`/`-=` behavior for mutating types versus real reassignment.
+- The docs, homepage samples, and release metadata now point at the current stable version.
+
+## User Highlights Across 1.5.x
+
+- Descending ranges and loops with `downTo` / `downUntil`
+- String interpolation with `$name` and `${expr}`
+- Backtick string literals for raw-ish string text
+- Decimal arithmetic, matrices/vectors, and complex numbers
+- Calendar `Date` support in `lyng.time`
+- `Channel`, `LaunchPool`, and `joinAll()` for coroutine workflows
+- Immutable collections and opt-in `ObservableList`
+- Rich `lyngio` modules for SQLite/JDBC databases, console, HTTP, WebSocket, TCP, and UDP
+- Legacy SHA-1 compatibility helpers in `lyng.legacy_digest`
+- CLI improvements including the built-in formatter `lyng fmt`
+- Better IDE support and stronger docs around the released feature set
 
 ## Language Features
 
+### Descending Ranges and Loops
+Lyng ranges are no longer just ascending. You can now write explicit descending ranges with inclusive or exclusive lower bounds.
+
+```lyng
+assertEquals([5,4,3,2,1], (5 downTo 1).toList())
+assertEquals([5,4,3,2], (5 downUntil 1).toList())
+
+for (i in 10 downTo 1 step 3) {
+    println(i)
+}
+```
+
+This also works for characters:
+
+```lyng
+assertEquals(['e','c','a'], ('e' downTo 'a' step 2).toList())
+```
+
+See [Range](Range.md).
+
+### String Interpolation
+Lyng 1.5.1 added built-in string interpolation:
+
+- `$name`
+- `${expr}`
+
+Literal dollar forms are explicit too:
+
+- `\$` -> `$`
+- `$$` -> `$`
+
+```lyng
+val name = "Lyng"
+assertEquals("hello, Lyng!", "hello, $name!")
+assertEquals("sum=3", "sum=${1+2}")
+assertEquals("\$name", "\$name")
+assertEquals("\$name", "$$name")
+```
+
+If you need legacy literal-dollar behavior in a file, add:
+
+```lyng
+// feature: interpolation: off
+```
+
+See [Tutorial](tutorial.md).
+
+### Matrix and Vector Module (`lyng.matrix`)
+Lyng now ships a dense linear algebra module with immutable double-precision `Matrix` and `Vector` types.
+
+It provides:
+
+- `matrix([[...]])` and `vector([...])`
+- matrix multiplication
+- matrix inversion
+- determinant, trace, rank
+- solving `A * x = b`
+- vector operations such as `dot`, `normalize`, `cross`, and `outer`
+
+```lyng
+import lyng.matrix
+
+val a: Matrix = matrix([[4, 7], [2, 6]])
+val inv: Matrix = a.inverse()
+assert(abs(inv.get(0, 0) - 0.6) < 1e-9)
+```
+
+Matrices also support Lyng-style slicing:
+
+```lyng
+import lyng.matrix
+
+val m: Matrix = matrix([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
+assertEquals(6.0, m[1, 2])
+val column: Matrix = m[0..2, 2]
+val tail: Matrix = m[1.., 1..]
+assertEquals([[3.0], [6.0], [9.0]], column.toList())
+assertEquals([[5.0, 6.0], [8.0, 9.0]], tail.toList())
+```
+
+See [Matrix](Matrix.md).
+
+### Multiple Selectors in Bracket Indexing
+Bracket indexing now accepts more than one selector:
+
+```lyng
+value[i]
+value[i, j]
+value[i, j, k]
+```
+
+For custom indexers, multiple selectors are packed into one list-like index object and dispatched through `getAt` / `putAt`.
+This is the rule used by `lyng.matrix` and by embedding APIs for Kotlin-backed indexers.
+
+### Decimal Arithmetic Module (`lyng.decimal`)
+Lyng now ships a first-class decimal module built as a regular extension library rather than a deep core special case.
+
+It provides:
+
+- `Decimal`
+- convenient `.d` conversions from `Int`, `Real`, and `String`
+- mixed arithmetic with `Int` and `Real`
+- local division precision and rounding control via `withDecimalContext(...)`
+
+```lyng
+import lyng.decimal
+
+assertEquals("3", (1 + 2.d).toStringExpanded())
+assertEquals("0.30000000000000004", (0.1 + 0.2).d.toStringExpanded())
+assertEquals("0.3", "0.3".d.toStringExpanded())
+
+assertEquals(
+    "0.3333333333",
+    withDecimalContext(10) { (1.d / 3.d).toStringExpanded() }
+)
+```
+
+The distinction between `Real -> Decimal` and exact decimal parsing is explicit by design:
+
+- `2.2.d` converts the current `Real` value
+- `"2.2".d` parses exact decimal text
+
+See [Decimal](Decimal.md).
+
+### Complex Numbers (`lyng.complex`)
+Lyng also ships a complex-number module for ordinary arithmetic in the complex plane.
+
+```lyng
+import lyng.complex
+
+assertEquals(Complex(1.0, 2.0), 1 + 2.i)
+assertEquals(Complex(2.0, 2.0), 2.i + 2)
+
+val z = 1 + π.i
+println(z.exp())
+```
+
+See [Complex](Complex.md).
+
+### Legacy Digest Module (`lyng.legacy_digest`)
+
+For situations where an external protocol or file format requires a SHA-1 value,
+Lyng now ships a `lyng.legacy_digest` module backed by a pure Kotlin/KMP
+implementation with no extra dependencies.
+
+> ⚠️ SHA-1 is cryptographically broken.  Use only for legacy-compatibility work.
+
+```lyng
+import lyng.legacy_digest
+
+val hex = LegacyDigest.sha1("abc")
+// → "a9993e364706816aba3e25717850c26c9cd0d89d"
+
+// Also accepts raw bytes:
+import lyng.buffer
+val buf = Buffer.decodeHex("616263")
+assertEquals(hex, LegacyDigest.sha1(buf))
+```
+
+The name `LegacyDigest` is intentional: it signals that these algorithms belong
+to a compatibility layer, not to a current security toolkit.
+
+See [LegacyDigest](LegacyDigest.md).
+
+### Binary Operator Interop Registry
+Lyng now provides a general mechanism for mixed binary operators through `lyng.operators`.
+
+This solves cases like:
+
+- `Int + MyType`
+- `Real < MyType`
+- `Int == MyType`
+
+without requiring changes to built-in classes.
+
+```lyng
+import lyng.operators
+
+class DecimalBox(val value: Int) {
+    fun plus(other: DecimalBox) = DecimalBox(value + other.value)
+    fun compareTo(other: DecimalBox) = value <=> other.value
+}
+
+OperatorInterop.register(
+    Int,
+    DecimalBox,
+    DecimalBox,
+    [BinaryOperator.Plus, BinaryOperator.Compare, BinaryOperator.Equals],
+    { x: Int -> DecimalBox(x) },
+    { x: DecimalBox -> x }
+)
+
+assertEquals(DecimalBox(3), 1 + DecimalBox(2))
+assert(1 < DecimalBox(2))
+assert(2 == DecimalBox(2))
+```
+
+`lyng.decimal` uses this same mechanism internally to interoperate with `Int` and `Real`.
+
+See [Operator Interop Registry](OperatorInterop.md).
+
 ### Class Properties with Accessors
 Classes now support properties with custom `get()` and `set()` accessors. Properties in Lyng do **not** have automatic backing fields; they are pure accessors.
 
@@ -24,6 +250,9 @@ class Person(private var _age: Int) {
 ### Private and Protected Setters
 You can now restrict the visibility of a property's or field's setter using `private set` or `protected set`. This allows members to be publicly readable but only writable from within the declaring class or its subclasses.
 
+### Refined Protected Visibility
+Ancestor classes can now access `protected` members of their descendants if it is an override of a member known to the ancestor. This enables base classes to call protected methods that are implemented or overridden in subclasses.
+
 ```lyng
 class Counter {
     var count = 0
@@ -98,13 +327,64 @@ Singleton objects are declared using the `object` keyword. They provide a conven
 
 ```lyng
 object Config {
-    val version = "1.2.3"
+    val version = "1.5.5"
     fun show() = println("Config version: " + version)
 }
 
 Config.show()
 ```
 
+Named singleton objects can also be used as extension receivers:
+
+```lyng
+object X {
+    fun base() = "base"
+}
+
+fun X.decorate(value): String {
+    this.base() + ":" + value.toString()
+}
+
+val X.tag get() = this.base() + ":tag"
+
+assertEquals("base:42", X.decorate(42))
+assertEquals("base:tag", X.tag)
+```
+
+### Nested Declarations and Lifted Enums
+You can now declare classes, objects, enums, and type aliases inside another class. These nested declarations live in the class namespace (no outer instance capture) and are accessed with a qualifier.
+
+```lyng
+class A {
+    class B(x?)
+    object Inner { val foo = "bar" }
+    enum E* { One, Two }
+}
+
+val ab = A.B()
+assertEquals(ab.x, null)
+assertEquals(A.Inner.foo, "bar")
+assertEquals(A.One, A.E.One)
+```
+
+The `*` on `enum E*` lifts entries into the enclosing class namespace (compile-time error on ambiguity).
+
+### Object Expressions
+You can now create anonymous objects that inherit from classes or interfaces using the `object : Base { ... }` syntax. These expressions capture their lexical scope and support multiple inheritance.
+
+```lyng
+val worker = object : Runnable {
+    override fun run() = println("Working...")
+}
+
+val x = object : Base(arg1), Interface1 {
+    val property = 42
+    override fun method() = this@object.property * 2
+}
+```
+
+Use `this@object` to refer to the innermost anonymous object instance when `this` is rebound.
+
 ### Unified Delegation Model
 A powerful new delegation system allows `val`, `var`, and `fun` members to delegate their logic to other objects using the `by` keyword.
 
@@ -123,8 +403,188 @@ var name by Observable("initial") { n, old, new ->
 
 The system features a unified interface (`getValue`, `setValue`, `invoke`) and a `bind` hook for initialization-time validation and configuration. See the [Delegation Guide](delegation.md) for more.
 
+### User-Defined Exception Classes
+You can now create custom exception types by inheriting from the built-in `Exception` class. Custom exceptions are real classes that can have their own fields and methods, and they work seamlessly with `throw` and `try-catch` blocks.
+
+```lyng
+class ValidationException(val field, m) : Exception(m)
+
+try {
+    throw ValidationException("email", "Invalid format")
+}
+catch(e: ValidationException) {
+    println("Error in " + e.field + ": " + e.message)
+}
+```
+
+### Assign-if-null Operator (`?=`)
+The new `?=` operator provides a concise way to assign a value only if the target is `null`. It is especially useful for setting default values or lazy initialization.
+
+```lyng
+var x = null
+x ?= 42          // x is now 42
+x ?= 100         // x remains 42 (not null)
+
+// Works with properties and index access
+config.port ?= 8080
+settings["theme"] ?= "dark"
+```
+
+The operator returns the final value of the receiver (the original value if it was not `null`, or the new value if the assignment occurred).
+
+### Transient Attribute (`@Transient`)
+The `@Transient` attribute can now be applied to class fields, constructor parameters, and static fields to exclude them from serialization.
+
+```lyng
+class MyData(@Transient val tempSecret, val publicData) {
+    @Transient var cachedValue = 0
+    var persistentValue = 42
+}
+```
+
+Key features:
+- **Serialization**: Transient members are omitted from both Lynon binary streams and JSON output.
+- **Structural Equality**: Transient fields are automatically ignored during `==` equality checks.
+- **Deserialization**: Transient constructor parameters with default values are correctly restored to those defaults upon restoration.
+
+### Value Clamping (`clamp`)
+A new `clamp()` function has been added to the standard library to limit a value within a specified range. It is available as both a global function and an extension method on all objects.
+
+```lyng
+// Global function
+clamp(15, 0..10)    // returns 10
+clamp(-5, 0..10)   // returns 0
+
+// Extension method
+val x = 15
+x.clamp(0..10)      // returns 10
+
+// Exclusive and open-ended ranges
+15.clamp(0..<10)    // returns 9
+15.clamp(..10)      // returns 10
+-5.clamp(0..)       // returns 0
+```
+
+`clamp()` correctly handles inclusive (`..`) and exclusive (`..<`) ranges. For discrete types like `Int` and `Char`, clamping to an exclusive upper bound returns the previous value.
+
+### Immutable Collections
+Lyng 1.5 adds immutable collection types for APIs that should not expose mutable state through aliases:
+
+- `ImmutableList`
+- `ImmutableSet`
+- `ImmutableMap`
+
+```lyng
+val a = ImmutableList(1,2,3)
+val b = a + 4
+
+assertEquals(ImmutableList(1,2,3), a)
+assertEquals(ImmutableList(1,2,3,4), b)
+```
+
+See [ImmutableList](ImmutableList.md), [ImmutableSet](ImmutableSet.md), and [ImmutableMap](ImmutableMap.md).
+
+### Observable Mutable Lists
+For reactive-style code, `lyng.observable` provides `ObservableList` with hooks and change streams.
+
+```lyng
+import lyng.observable
+
+val xs = [1,2].observable()
+xs.onChange { println("changed") }
+xs += 3
+```
+
+You can validate or reject mutations in `beforeChange`, listen in `onChange`, and consume structured change events from `changes()`.
+
+See [ObservableList](ObservableList.md).
+
+### Random API
+The standard library now includes a built-in random API plus deterministic seeded generators.
+
+```lyng
+val rng = Random.seeded(1234)
+assert(rng.next(1..10) in 1..10)
+assert(rng.next('a'..<'f') in 'a'..<'f')
+```
+
+Use:
+
+- `Random.nextInt()`
+- `Random.nextFloat()`
+- `Random.next(range)`
+- `Random.seeded(seed)`
+
 ## Tooling and Infrastructure
 
+### Rich Console Apps with `lyng.io.console`
+`lyngio` now includes a real console module for terminal applications:
+
+- TTY detection
+- screen clearing and cursor movement
+- alternate screen buffer
+- raw input mode
+- typed key and resize events
+
+```lyng
+import lyng.io.console
+
+Console.enterAltScreen()
+Console.clear()
+Console.moveTo(1, 1)
+Console.write("Hello from Lyng console app")
+Console.flush()
+Console.leaveAltScreen()
+```
+
+The repository includes a full interactive Tetris sample built on this API.
+
+See [lyng.io.console](lyng.io.console.md).
+
+### HTTP, WebSocket, TCP, and UDP in `lyngio`
+`lyngio` grew from filesystem/process support into a broader application-facing I/O library. In 1.5.x it includes:
+
+- `lyng.io.http` for HTTP/HTTPS client calls
+- `lyng.io.ws` for WebSocket clients
+- `lyng.io.net` for raw TCP/UDP transport
+
+HTTP example:
+
+```lyng
+import lyng.io.http
+
+val r = Http.get("https://example.com")
+println(r.status)
+println(r.text())
+```
+
+TCP example:
+
+```lyng
+import lyng.io.net
+
+val socket = Net.tcpConnect("127.0.0.1", 4040)
+socket.writeUtf8("ping")
+socket.flush()
+println(socket.readLine())
+socket.close()
+```
+
+WebSocket example:
+
+```lyng
+import lyng.io.ws
+
+val ws = Ws.connect("wss://example.com/socket")
+ws.sendText("hello")
+println(ws.receive())
+ws.close()
+```
+
+These modules are capability-gated and host-installed, keeping Lyng safe by default while making networked scripts practical when enabled.
+
+See [lyngio overview](lyngio.md), [lyng.io.db](lyng.io.db.md), [lyng.io.http](lyng.io.http.md), [lyng.io.ws](lyng.io.ws.md), and [lyng.io.net](lyng.io.net.md).
+
 ### CLI: Formatting Command
 A new `fmt` subcommand has been added to the Lyng CLI.
 
@@ -134,7 +594,28 @@ lyng fmt --in-place MyFile.lyng  # Format file in-place
 lyng fmt --check MyFile.lyng     # Check if file needs formatting
 ```
 
+### CLI: Better Terminal Workflows
+The CLI is no longer just a script launcher. In the 1.5 line it also gained:
+
+- built-in formatter support
+- integrated `lyng.io.console` support for terminal programs
+- downloadable packaged distributions for easier local use
+
+This makes CLI-first scripting and console applications much more practical than in earlier releases.
+
 ### IDEA Plugin: Autocompletion
 Experimental lightweight autocompletion is now available in the IntelliJ plugin. It features type-aware member suggestions and inheritance-aware completion.
 
 You can enable it in **Settings | Lyng Formatter | Enable Lyng autocompletion**.
+
+### Kotlin API: Exception Handling
+The `Obj.getLyngExceptionMessageWithStackTrace()` extension method has been added to simplify retrieving detailed error information from Lyng exception objects in Kotlin. Additionally, `getLyngExceptionMessage()` and `raiseAsExecutionError()` now accept an optional `Scope`, making it easier to use them when a scope is not immediately available.
+
+### Kotlin API: Bridge Reflection and Class Binding (Preferred Extensions)
+Lyng now provides a public Kotlin reflection bridge and a Lyng‑first class binding workflow. This is the **preferred** way to write Kotlin extensions and library integrations:
+
+- **Bridge resolver**: explicit handles for values, vars, and callables with predictable lookup rules.
+- **Class bridge binding**: declare extern surfaces in Lyng (`extern` members, or members inside `extern class/object`) and bind the implementations in Kotlin before the first instance is created.
+- **Extern declaration rule**: `extern class` / `extern object` are declaration-only; all members in their bodies are implicitly extern.
+
+See **Embedding Lyng** for full samples and usage details.

docs/whats_new_1_3.md

@@ -0,0 +1,133 @@
+# What's New in Lyng 1.3 (vs 1.2.* / master)
+
+This is a programmer-focused summary of what changed since the 1.2.* line on `master`. It highlights new language and type-system features, runtime/IDE improvements, and how to migrate code safely.
+
+## Highlights
+
+- Generics are now a first-class part of the type system, with bounds, variance, unions, and intersections.
+- Type aliases and type-expression checks (`T1 is T2`, `A in T`) enable richer static modeling.
+- Nested declarations inside classes, plus lifted enum entries via `enum E*`.
+- Stepped ranges (`step`) including iterable open-ended and real ranges.
+- Runtime and compiler speedups: more bytecode coverage, direct slot access, call-site caching.
+
+## Language and type system
+
+### Generics, bounds, and variance
+
+You can declare generic functions/classes with `<...>`, restrict them with bounds, and control variance.
+
+```lyng
+fun id<T>(x: T): T = x
+class Box<out T>(val value: T)
+
+fun sum<T: Int | Real>(x: T, y: T) = x + y
+class Named<T: Iterable & Comparable>(val data: T)
+```
+
+### Type aliases and type expressions
+
+Type aliases can name any type expression, including unions and intersections.
+
+```lyng
+type Num = Int | Real
+type Maybe<T> = T?
+```
+
+Type expressions can be checked directly:
+
+```lyng
+fun f<T>(xs: List<T>) {
+    assert( T is Int | String | Bool ) // type-subset check
+    assert( Int in T )                 // same relation as `Int is T`
+}
+```
+
+Value checks remain `x is T`. Type expression equality uses `==` and is structural.
+
+### Nullable shorthand for parameters
+
+Untyped parameters and constructor args can use `x?` as shorthand for `x: Object?`:
+
+```lyng
+class A(x?) { ... }
+fun f(x?) { x == null }
+```
+
+### List/map literal inference
+
+The compiler now infers element and key/value types from literals and spreads. Mixed element types produce unions.
+
+```lyng
+val a = [1, 2, 3]             // List<Int>
+val b = [1, "two", true]      // List<Int | String | Bool>
+val m = { "a": 1, "b": "x" } // Map<String, Int | String>
+```
+
+### Compile-time member access only
+
+Member access is resolved at compile time. On unknown types, only `Object` members are visible; other members require an explicit cast.
+
+```lyng
+fun f(x) {          // x: Object
+    x.toString()    // ok
+    x.size()        // compile-time error
+    (x as List).size()
+}
+```
+
+This removes runtime name-resolution fallbacks and makes errors deterministic.
+
+### Nested declarations and lifted enums
+
+Classes, objects, enums, and type aliases can be declared inside another class and accessed by qualifier. Enums can lift entries into the outer namespace with `*`.
+
+```lyng
+class A {
+    class B(x?)
+    object Inner { val foo = "bar" }
+    type Alias = B
+    enum E* { One, Two }
+}
+
+val b = A.B()
+assertEquals(A.One, A.E.One)
+```
+
+### Stepped ranges
+
+Ranges now support `step`, and open-ended/real ranges are iterable only with an explicit step.
+
+```lyng
+(1..5 step 2).toList()      // [1,3,5]
+(0.0..1.0 step 0.25).toList() // [0,0.25,0.5,0.75,1.0]
+(0.. step 1).take(3).toList() // [0,1,2]
+```
+
+## Tooling and performance
+
+- Bytecode compiler/VM coverage expanded (loops, expressions, calls), improving execution speed and consistency.
+- Direct frame-slot access and scoped slot addressing reduce lookup overhead, including in closures.
+- Call-site caching and numeric fast paths reduce hot-loop overhead.
+- IDE tooling updated for the new type system and nested declarations; MiniAst-based completion work continues.
+
+## Migration guide (from 1.2.*)
+
+1) Replace dynamic member access on unknown types
+- If you relied on runtime name resolution, add explicit casts or annotate types so the compiler can resolve members.
+
+2) Adopt new type-system constructs where helpful
+- Consider `type` aliases for complex unions/intersections.
+- Prefer generic signatures over `Object` when the API is parametric.
+
+3) Update range iteration where needed
+- Use `step` for open-ended or real ranges you want to iterate.
+
+4) Nullable shorthand is optional
+- If you used untyped nullable params, you can keep `x` (Object) or switch to `x?` (Object?) for clarity.
+
+## References
+
+- `docs/generics.md`
+- `docs/Range.md`
+- `docs/OOP.md`
+- `docs/BytecodeSpec.md`

docs/whats_new_1_5.md

@@ -0,0 +1,145 @@
+# What's New in Lyng 1.5 (vs 1.3.* / master)
+
+This document summarizes the significant changes and new features introduced in the 1.5 development cycle.
+
+## Principal changes
+
+### JIT compiler and compile-time types and symbols.
+
+This major improvement gives the following big advantages:
+
+- **Blazing Fast execution**: several times faster! (three to six times speedup in different scenarios).
+- **Better IDE support**: autocompletion, early error detection, types check.
+- **Error safety**: all symbols and types are checked at bound at compile-time. Many errors are detected earlier. Also, no risk that external or caller code would shadow some internally used symbols (especially in closures and inheritance).
+
+In particular, it means no slow and flaky runtime lookups. Once compiled, code guarantees that it will always call the symbol known at compile-time; runtime name lookup though does not guarantee it and can be source of hard to trace bugs. 
+
+### New stable API to create Kotlin extensions
+
+The API is fixed and will be kept with further Lyng core changes. It is now the recommended way to write Lyng extensions in Kotlin. It is much simpler and more elegant than the internal one. See [Kotlin Bridge Binding](../notes/kotlin_bridge_binding.md).
+
+Extern declaration clarification:
+- `extern class` / `extern object` are pure extern surfaces.
+- Members inside them are implicitly extern (`extern` on a member is optional/redundant).
+- Lyng method/property bodies for these declarations should be implemented as extensions instead.
+
+### Smart types system
+
+- **Deep inference**: The compiler analyzes types of symbols along the execution path and in many cases eliminates unnecessary casts or type specifications.
+- **Union and intersection types**: `A & B`, `A | B`.
+- **Generics**: Generic types are first-class citizens with support for [bounds and variance](generics.md). Type params are erased by default and are reified only when needed (e.g., `T::class`, `T is ...`, `as T`, or in extern-facing APIs), which enables checks like `A in T` when `T` is reified.
+- **Inner classes and enums**: Full support for nested declarations, including [Enums with lifting](OOP.md#lifted-enum-entries).
+
+## Other highlights
+
+- **The `return` Statement**: Added support for local and non-local returns using labels.
+- **Abstract Classes and Interfaces**: Full support for `abstract` members and the `interface` keyword.
+- **Singleton Objects**: Define singletons using the `object` keyword or use anonymous object expressions.
+- **Multiple Inheritance**: Enhanced multi-inheritance with predictable [C3 MRO resolution](OOP.md#multiple-inheritance-and-mro).
+- **Unified Delegation**: Powerful delegation model for `val`, `var`, and `fun` members. See [Delegation](delegation.md).
+- **Class Properties with Accessors**: Define `val` and `var` properties with custom `get()` and `set()`.
+- **Restricted Setter Visibility**: Use `private set` and `protected set` on fields and properties.
+- **Late-initialized `val`**: Support for `val` fields that are initialized in `init` blocks or class bodies.
+- **Transient Members**: Use `@Transient` to exclude members from serialization and equality checks.
+- **Named Arguments and Splats**: Improved call-site readability with `name: value` and map-based splats.
+- **Refined Visibility**: Improved `protected` access and `closed` modifier for better encapsulation.
+
+## Language Features
+
+### The `return` Statement
+You can now exit from the innermost enclosing callable (function or lambda) using `return`. Lyng also supports non-local returns to outer scopes using labels. See [Return Statement](return_statement.md).
+
+```lyng
+fun findFirst<T>(list: Iterable<T>, predicate: (T)->Bool): T? {
+    list.forEach {
+        if (predicate(it)) return@findFirst it
+    }
+    null
+}
+```
+
+### Abstract Classes and Interfaces
+Lyng now supports the `abstract` modifier for classes and their members. `interface` is introduced as a synonym for `abstract class`, allowing for rich multi-inheritance patterns.
+
+```lyng
+interface Shape {
+    abstract val area: Real
+    fun describe() = "Area: %g"(area)
+}
+
+class Circle(val radius: Real) : Shape {
+    override val area get = Math.PI * radius * radius
+}
+```
+
+### Class Properties with Accessors
+Properties can now have custom getters and setters. They do not have automatic backing fields, making them perfect for computed values or delegation.
+
+```lyng
+class Rectangle(var width: Real, var height: Real) {
+    val area: Real get() = width * height
+    
+    var squareSize: Real
+        get() = area
+        set(v) {
+            width = sqrt(v)
+            height = width
+        }
+}
+```
+
+### Singleton Objects
+Declare singletons or anonymous objects easily.
+
+```lyng
+object Database {
+    val connection = "connected"
+}
+
+val runner = object : Runnable {
+    override fun run() = println("Running!")
+}
+```
+
+### Named Arguments and Named Splats
+Improve call-site clarity by specifying argument names. You can also expand a Map into named arguments using the splat operator.
+
+```lyng
+fun configure(timeout: Int, retry: Int = 3) { ... }
+
+configure(timeout: 5000, retry: 5)
+val options = Map("timeout": 1000, "retry": 1)
+configure(...options)
+```
+
+### Modern Operators
+The `?=` operator allows for concise "assign if null" logic.
+
+```lyng
+var cache: Map? = null
+cache ?= Map("status": "ok") // Only assigns if cache is null
+```
+
+## Tooling and IDE
+
+- **IDEA Plugin**: Significant improvements to autocompletion, documentation tooltips, and natural language support (Grazie integration).
+- **CLI**: The `lyng fmt` command is now a first-class tool for formatting code with various options like `--check` and `--in-place`.
+- **Performance**: Ongoing optimizations in the bytecode VM and compiler for faster execution and smaller footprint.
+
+## Standard Library
+- **`with(self, block)`**: Scoped execution with a dedicated `self`.
+- **`clamp(value, min, max)`**: Easily restrict values to a range.
+
+## Migration Guide (from 1.3.*)
+
+1. **Check Visibility**: Refined `protected` and `private` rules may catch previously undetected invalid accesses.
+2. **Override Keyword**: Ensure all members that override ancestor declarations are marked with the `override` keyword (now mandatory).
+3. **Return in Shorthand**: Remember that `return` is forbidden in `=` shorthand functions; use block syntax if you need early exit.
+4. **Empty Map Literals**: Use `Map()` or `{:}` for empty maps, as `{}` is now strictly a block/lambda.
+
+## References
+- [Object Oriented Programming](OOP.md)
+- [Generics](generics.md)
+- [Return Statement](return_statement.md)
+- [Delegation](delegation.md)
+- [Tutorial](tutorial.md)

editors/lyng-textmate/syntaxes/lyng.tmLanguage.json

@@ -42,7 +42,7 @@
         { "name": "constant.numeric.decimal.lyng", "match": "(?<![A-Za-z_])(?:[0-9][0-9_]*)\\.(?:[0-9_]+)(?:[eE][+-]?[0-9_]+)?|(?<![A-Za-z_])(?:[0-9][0-9_]*)(?:[eE][+-]?[0-9_]+)?" }
       ]
     },
-    "annotations": { "patterns": [ { "name": "entity.name.label.at.lyng", "match": "@[\\p{L}_][\\p{L}\\p{N}_]*:" }, { "name": "storage.modifier.annotation.lyng", "match": "@[\\p{L}_][\\p{L}\\p{N}_]*" } ] },
+    "annotations": { "patterns": [ { "name": "entity.name.label.at.lyng", "match": "@[\\p{L}_][\\p{L}\\p{N}_]*" } ] },
     "mapLiterals": {
       "patterns": [
         {
@@ -74,11 +74,11 @@
         }
       ]
     },
-    "labels": { "patterns": [ { "name": "entity.name.label.lyng", "match": "[\\p{L}_][\\p{L}\\p{N}_]*:" } ] },
+    "labels": { "patterns": [ { "name": "entity.name.label.lyng", "match": "[\\p{L}_][\\p{L}\\p{N}_]*@" } ] },
     "directives": { "patterns": [ { "name": "meta.directive.lyng", "match": "^\\s*#[_A-Za-z][_A-Za-z0-9]*" } ] },
-    "declarations": { "patterns": [ { "name": "meta.function.declaration.lyng", "match": "\\b(fun|fn)\\s+(?:([\\p{L}_][\\p{L}\\p{N}_]*)\\.)?([\\p{L}_][\\p{L}\\p{N}_]*)", "captures": { "1": { "name": "keyword.declaration.lyng" }, "2": { "name": "entity.name.type.lyng" }, "3": { "name": "entity.name.function.lyng" } } }, { "name": "meta.type.declaration.lyng", "match": "\\b(?:class|enum|interface)\\s+([\\p{L}_][\\p{L}\\p{N}_]*)", "captures": { "1": { "name": "entity.name.type.lyng" } } }, { "name": "meta.variable.declaration.lyng", "match": "\\b(val|var)\\s+(?:([\\p{L}_][\\p{L}\\p{N}_]*)\\.)?([\\p{L}_][\\p{L}\\p{N}_]*)", "captures": { "1": { "name": "keyword.declaration.lyng" }, "2": { "name": "entity.name.type.lyng" }, "3": { "name": "variable.other.declaration.lyng" } } } ] },
-    "keywords": { "patterns": [ { "name": "keyword.control.lyng", "match": "\\b(?:if|else|when|while|do|for|try|catch|finally|throw|return|break|continue)\\b" }, { "name": "keyword.declaration.lyng", "match": "\\b(?:fun|fn|class|enum|interface|val|var|import|package|constructor|property|abstract|override|open|closed|extern|private|protected|static|get|set)\\b" }, { "name": "keyword.operator.word.lyng", "match": "\\bnot\\s+(?:in|is)\\b" }, { "name": "keyword.operator.word.lyng", "match": "\\b(?:and|or|not|in|is|as|as\\?)\\b" } ] },
-    "constants": { "patterns": [ { "name": "constant.language.lyng", "match": "(?:\\b(?:true|false|null|this)\\b|π)" } ] },
+    "declarations": { "patterns": [ { "name": "meta.function.declaration.lyng", "match": "\\b(fun|fn)\\s+(?:([\\p{L}_][\\p{L}\\p{N}_]*)\\.)?([\\p{L}_][\\p{L}\\p{N}_]*)", "captures": { "1": { "name": "keyword.declaration.lyng" }, "2": { "name": "entity.name.type.lyng" }, "3": { "name": "entity.name.function.lyng" } } }, { "name": "meta.type.declaration.lyng", "match": "\\b(?:class|enum|interface|object)(?:\\s+([\\p{L}_][\\p{L}\\p{N}_]*))?", "captures": { "1": { "name": "entity.name.type.lyng" } } }, { "name": "meta.variable.declaration.lyng", "match": "\\b(val|var)\\s+(?:([\\p{L}_][\\p{L}\\p{N}_]*)\\.)?([\\p{L}_][\\p{L}\\p{N}_]*)", "captures": { "1": { "name": "keyword.declaration.lyng" }, "2": { "name": "entity.name.type.lyng" }, "3": { "name": "variable.other.declaration.lyng" } } } ] },
+    "keywords": { "patterns": [ { "name": "keyword.control.lyng", "match": "\\b(?:if|else|when|while|do|for|try|catch|finally|throw|return|break|continue)\\b" }, { "name": "keyword.declaration.lyng", "match": "\\b(?:fun|fn|class|enum|interface|val|var|import|package|constructor|property|abstract|override|open|closed|extern|private|protected|static|get|set|object|init|by)\\b" }, { "name": "keyword.operator.word.lyng", "match": "\\bnot\\s+(?:in|is)\\b" }, { "name": "keyword.operator.word.lyng", "match": "\\b(?:and|or|not|in|is|as|as\\?)\\b" } ] },
+    "constants": { "patterns": [ { "name": "constant.language.lyng", "match": "(?:\\b(?:true|false|null|this(?:@[\\p{L}_][\\p{L}\\p{N}_]*)?)\\b|π)" } ] },
     "types": { "patterns": [ { "name": "storage.type.lyng", "match": "\\b(?:Int|Real|String|Bool|Char|Regex)\\b" }, { "name": "entity.name.type.lyng", "match": "\\b[A-Z][A-Za-z0-9_]*\\b(?!\\s*\\()" } ] },
     "operators": { "patterns": [ { "name": "keyword.operator.comparison.lyng", "match": "===|!==|==|!=|<=|>=|<|>" }, { "name": "keyword.operator.shuttle.lyng", "match": "<=>" }, { "name": "keyword.operator.arrow.lyng", "match": "=>|->|::" }, { "name": "keyword.operator.range.lyng", "match": "\\.\\.\\.|\\.\\.<|\\.\\." }, { "name": "keyword.operator.nullsafe.lyng", "match": "\\?\\.|\\?\\[|\\?\\(|\\?\\{|\\?:|\\?\\?" }, { "name": "keyword.operator.assignment.lyng", "match": "(?:\\+=|-=|\\*=|/=|%=|=)" }, { "name": "keyword.operator.logical.lyng", "match": "&&|\\|\\|" }, { "name": "keyword.operator.bitwise.lyng", "match": "<<|>>|&|\\||\\^|~" }, { "name": "keyword.operator.match.lyng", "match": "=~|!~" }, { "name": "keyword.operator.arithmetic.lyng", "match": "\\+\\+|--|[+\\-*/%]" }, { "name": "keyword.operator.other.lyng", "match": "[!?]" } ] },
     "punctuation": { "patterns": [ { "name": "punctuation.separator.comma.lyng", "match": "," }, { "name": "punctuation.terminator.statement.lyng", "match": ";" }, { "name": "punctuation.section.block.begin.lyng", "match": "[(]{1}|[{]{1}|\\[" }, { "name": "punctuation.section.block.end.lyng", "match": "[)]{1}|[}]{1}|\\]" }, { "name": "punctuation.accessor.dot.lyng", "match": "\\." }, { "name": "punctuation.separator.colon.lyng", "match": ":" } ] }

examples/content_index_db.lyng

@@ -0,0 +1,325 @@
+#!/usr/bin/env lyng
+
+import lyng.io.db
+import lyng.io.db.sqlite
+import lyng.io.fs
+
+val DB_FILE_NAME = "contents.db"
+val ANSI_ESC = "\u001b["
+val NEWLINE = "\n"
+val WINDOWS_SEPARATOR = "\\"
+val SQLITE_JOURNAL_SUFFIXES = ["-wal", "-shm", "-journal"]
+
+val USAGE_TEXT = "
+Lyng content index
+Scan a directory tree, diff it against a SQLite snapshot, and optionally refresh the snapshot.
+
+usage:
+lyng examples/content_index_db.lyng <root> [-u|--update]
+
+options:
+-u, --update   write the current scan back to $DB_FILE_NAME
+
+notes:
+- the database lives inside <root>/$DB_FILE_NAME
+- on first run the snapshot is created automatically
+- the script ignores its own SQLite sidecar files
+"
+
+val CREATE_FILE_INDEX_SQL = "
+create table if not exists file_index(
+    path text primary key not null,
+    size integer not null,
+    mtime integer not null
+)
+"
+
+val CREATE_CURRENT_SCAN_SQL = "
+create temp table current_scan(
+    path text primary key not null,
+    size integer not null,
+    mtime integer not null
+)
+"
+
+val SELECT_ADDED_SQL = "
+select
+c.path,
+c.size,
+c.mtime
+from current_scan c
+left join file_index f on f.path = c.path
+where f.path is null
+order by c.path
+"
+
+val SELECT_REMOVED_SQL = "
+select f.path, f.size, f.mtime
+from file_index f
+left join current_scan c on c.path = f.path
+where c.path is null
+order by f.path
+"
+
+val SELECT_CHANGED_SQL = "
+select c.path, f.size as old_size, c.size as new_size, f.mtime as old_mtime,
+       c.mtime as new_mtime
+from current_scan c
+join file_index f on f.path = c.path
+where c.size != f.size or c.mtime != f.mtime
+order by c.path
+"
+
+val DELETE_MISSING_SQL = "
+delete from file_index
+where not exists (
+    select 1
+    from current_scan c
+    where c.path = file_index.path
+)
+"
+
+val UPSERT_SCAN_SQL = "
+insert or replace into file_index(path, size, mtime)
+select path, size, mtime
+from current_scan
+"
+
+val INSERT_SCAN_ROW_SQL = "
+insert into current_scan(path, size, mtime)
+values(?, ?, ?)
+"
+
+val USE_COLOR = true
+
+class CliOptions(val rootText: String, val updateSnapshot: Bool) {}
+
+fun out(text: String? = null): Void {
+    if (text == null) {
+        print(NEWLINE)
+        return
+    }
+    print(text + NEWLINE)
+}
+
+fun paint(code: String, text: String): String {
+    if (!USE_COLOR) return text
+    ANSI_ESC + code + "m" + text + ANSI_ESC + "0m"
+}
+
+fun bold(text: String): String = paint("1", text)
+fun dim(text: String): String = paint("2", text)
+fun cyan(text: String): String = paint("36", text)
+fun green(text: String): String = paint("32", text)
+fun yellow(text: String): String = paint("33", text)
+fun red(text: String): String = paint("31", text)
+
+fun signed(value: Int): String = if (value > 0) "+" + value else value.toString()
+
+fun plural(count: Int, one: String, many: String): String {
+    if (count == 1) return one
+    many
+}
+
+fun childPath(parent: Path, name: String): Path {
+    val base = parent.toString()
+    if (base.endsWith("/") || base.endsWith(WINDOWS_SEPARATOR)) {
+        return Path(base + name)
+    }
+    Path(base + "/" + name)
+}
+
+fun relativePath(root: Path, file: Path): String {
+    val parts: List<String> = []
+    for (i in root.segments.size..<file.segments.size) {
+        parts.add(file.segments[i] as String)
+    }
+    parts.joinToString("/")
+}
+
+fun isDatabaseArtifact(relative: String): Bool {
+    relative == DB_FILE_NAME || SQLITE_JOURNAL_SUFFIXES.any { relative == DB_FILE_NAME + (it as String) }
+}
+
+fun printUsage(message: String? = null): Void {
+    if (message != null && message.trim().isNotEmpty()) {
+        out(red("error: ") + message)
+        out()
+    }
+
+    out(bold(USAGE_TEXT))
+}
+
+fun parseArgs(argv: List<String>): CliOptions? {
+    var rootText: String? = null
+    var updateSnapshot = false
+
+    for (arg in argv) {
+        when (arg) {
+            "-u", "--update" -> updateSnapshot = true
+            "-h", "--help" -> {
+                printUsage()
+                return null
+            }
+            else -> {
+                if (arg.startsWith("-")) {
+                    printUsage("unknown option: " + arg)
+                    return null
+                }
+                if (rootText != null) {
+                    printUsage("only one root path is allowed")
+                    return null
+                }
+                rootText = arg
+            }
+        }
+    }
+
+    if (rootText == null) {
+        printUsage("missing required <root> argument")
+        return null
+    }
+
+    CliOptions(rootText as String, updateSnapshot)
+}
+
+fun printBanner(root: Path, dbFile: Path, dbWasCreated: Bool, updateSnapshot: Bool): Void {
+    val mode =
+            if (dbWasCreated) "bootstrap snapshot"
+        else if (updateSnapshot) "scan + refresh snapshot"
+        else "scan only"
+
+    out(cyan("== Lyng content index =="))
+    out(dim("root: " + root))
+    out(dim("db:   " + dbFile))
+    out(dim("mode: " + mode))
+    out()
+}
+
+fun printSection(title: String, accent: (String)->String, rows: List<SqlRow>, render: (SqlRow)->String): Void {
+    out(accent(title + " (" + rows.size + ")"))
+    if (rows.isEmpty()) {
+        out(dim("  none"))
+        out()
+        return
+    }
+
+    for (row in rows) {
+        out(render(row))
+    }
+    out()
+}
+
+fun renderAdded(row: SqlRow): String {
+    val path = row["path"] as String
+    val size = row["size"] as Int
+    val mtime = row["mtime"] as Int
+    "  " + green("+") + " " + bold(path) + dim("  %12d B  mtime %d"(size, mtime))
+}
+
+fun renderRemoved(row: SqlRow): String {
+    val path = row["path"] as String
+    val size = row["size"] as Int
+    val mtime = row["mtime"] as Int
+    "  " + red("-") + " " + bold(path) + dim("  %12d B  mtime %d"(size, mtime))
+}
+
+fun renderChanged(row: SqlRow): String {
+    val path = row["path"] as String
+    val oldSize = row["old_size"] as Int
+    val newSize = row["new_size"] as Int
+    val oldMtime = row["old_mtime"] as Int
+    val newMtime = row["new_mtime"] as Int
+    val sizeDelta = newSize - oldSize
+    val mtimeDelta = newMtime - oldMtime
+
+    "  " + yellow("~") + " " + bold(path) +
+    dim(
+        "  size %d -> %d (%s B), mtime %d -> %d (%s ms)"(
+            oldSize,
+            newSize,
+            signed(sizeDelta),
+            oldMtime,
+            newMtime,
+            signed(mtimeDelta)
+    )
+    )
+}
+
+fun loadRows(tx: SqlTransaction, query: String): List<SqlRow> = tx.select(query).toList()
+
+fun main() {
+    val argv: List<String> = []
+    for (raw in ARGV as List) {
+        argv.add(raw as String)
+    }
+    val options = parseArgs(argv)
+    if (options == null) {
+        return
+    }
+
+    val root = Path(options.rootText)
+    if (!root.exists()) {
+        printUsage("root does not exist: " + root)
+        return
+    }
+    if (!root.isDirectory()) {
+        printUsage("root is not a directory: " + root)
+        return
+    }
+
+    val dbFile = childPath(root, DB_FILE_NAME)
+    val dbWasCreated = !dbFile.exists()
+    val shouldUpdateSnapshot = dbWasCreated || options.updateSnapshot
+
+    printBanner(root, dbFile, dbWasCreated, shouldUpdateSnapshot)
+
+    val db = openSqlite(dbFile.toString())
+
+    db.transaction { tx ->
+        tx.execute(CREATE_FILE_INDEX_SQL)
+
+        tx.execute("drop table if exists temp.current_scan")
+        tx.execute(CREATE_CURRENT_SCAN_SQL)
+
+        var scannedFiles = 0
+        for (rawEntry in root.glob("**")) {
+            val entry = rawEntry as Path
+            if (!entry.isFile()) continue
+
+            val relative = relativePath(root, entry)
+            if (isDatabaseArtifact(relative)) continue
+
+            val size = entry.size() ?: 0
+            val mtime = entry.modifiedAtMillis() ?: 0
+            tx.execute(INSERT_SCAN_ROW_SQL, relative, size, mtime)
+            scannedFiles++
+        }
+
+        val added = loadRows(tx, SELECT_ADDED_SQL)
+        val removed = loadRows(tx, SELECT_REMOVED_SQL)
+        val changed = loadRows(tx, SELECT_CHANGED_SQL)
+
+        val totalChanges = added.size + removed.size + changed.size
+
+        out(dim("scanned %d %s under %s"(scannedFiles, plural(scannedFiles, "file", "files"), root.toString())))
+        out(dim("detected %d %s"(totalChanges, plural(totalChanges, "change", "changes"))))
+        out()
+
+        printSection("Added", { green(it) }, added) { renderAdded(it) }
+        printSection("Removed", { red(it) }, removed) { renderRemoved(it) }
+        printSection("Changed", { yellow(it) }, changed) { renderChanged(it) }
+
+        if (shouldUpdateSnapshot) {
+            tx.execute(DELETE_MISSING_SQL)
+            tx.execute(UPSERT_SCAN_SQL)
+
+            val action = if (dbWasCreated) "created" else "updated"
+            out(cyan("snapshot " + action + " in " + dbFile.name))
+        } else {
+            out(dim("snapshot unchanged; re-run with -u or --update to persist the scan"))
+        }
+    }
+}
+
+main()

examples/extract_lynglang_version.lyng

@@ -0,0 +1,23 @@
+#!/env/bin lyng
+
+import lyng.io.http
+
+// Step 1: download the main lynglang.com page.
+val home = Http.get("https://lynglang.com").text()
+
+// Step 2: find the version-script reference in the page HTML.
+val jsRef = "src=\"([^\"]*lyng-version\\.js)\"".re.find(home)
+require(jsRef != null, "lyng-version.js reference not found on the homepage")
+
+// Step 3: extract the referenced script path from the first regex capture.
+val versionJsPath = jsRef[1]
+
+// Step 4: download the script that exposes `window.LYNG_VERSION`.
+val versionJs = Http.get("https://lynglang.com/" + versionJsPath).text()
+
+// Step 5: pull the actual version string from the JavaScript source.
+val versionMatch = "LYNG_VERSION\\s*=\\s*\"([^\"]+)\"".re.find(versionJs)
+require(versionMatch != null, "LYNG_VERSION assignment not found")
+
+// Step 6: print the discovered version for the user.
+println("Lynglang.com version: " + ((versionMatch as RegexMatch)[1]))

examples/fillspeed.lyng

@@ -0,0 +1,15 @@
+import lyng.time
+
+val n = 700_000
+
+fun tm<T>(block: ()->T): T {
+    val t = Instant()
+    block().also {
+        println("tm: ${Instant() - t}")
+    }
+}
+
+val x = tm { List.fill(n) { it * 10 + 1 } }
+val y = tm { List.fill(n, n + 10) { it * 10 + 1 } }
+tm { x.add(-1) }
+tm { y.add(-2) }

examples/free_fall.lyng

@@ -0,0 +1,76 @@
+fun calculateDepth(
+    T: Real,
+    m: Real,
+    d: Real,
+    rho: Real = 1.2,
+    c: Real = 340.0,
+    g: Real = 9.81,
+    Cd: Real = 0.5,
+    eps: Real = 1e-3,
+    maxIter: Int = 100
+): Real? {
+    // Площадь миделя
+    val r = d / 2.0
+    val A = π * r * r
+
+    // Коэффициент сопротивления
+    val k = 0.5 * Cd * rho * A
+
+    // Предельная скорость
+    val vTerm = sqrt(m * g / k)
+
+    // Функция времени падения с высоты h
+    fun tFall(h: Real): Real {
+        // Для численной стабильности при больших h используем логарифмическую форму
+        val arg = exp(g * h / (vTerm * vTerm))
+        // arcosh(x) = ln(x + sqrt(x^2 - 1))
+        val arcosh = ln(arg + sqrt(arg * arg - 1.0))
+        return vTerm / g * arcosh
+    }
+
+    // Полное расчётное время
+    fun Tcalc(h: Real): Real = tFall(h) + h / c
+
+    // Находим интервал, содержащий корень
+    // Нижняя граница: глубина не может быть отрицательной
+    var lo = 0.0
+    // Верхняя граница: сначала попробуем оценку по свободному падению (без звука)
+    var hi = 0.5 * g * T * T  // максимальная глубина, если бы не было сопротивления и звука
+    // Уточним hi, чтобы Tcalc(hi) было заведомо больше T
+    while (Tcalc(hi) < T && hi < 1e4) {
+        hi *= 2.0
+    }
+    // Проверка, что hi достаточно велико
+    if (Tcalc(hi) < T) return null // слишком большая глубина, не укладываемся в разумное
+
+    // Бисекция
+    var iter = 0
+    var h = (lo + hi) / 2.0
+    while (iter < maxIter && (hi - lo) > eps) {
+        val f = Tcalc(h) - T
+        if (abs(f) < eps) break
+        if (f > 0) {
+            hi = h
+        } else {
+            lo = h
+        }
+        h = (lo + hi) / 2.0
+        iter++
+    }
+
+    return h
+}
+
+    // Пример: T=12 секунд
+    val T = 26.0
+    val m = 1.0   // кг
+    val d = 0.1   // м
+
+    val depth = calculateDepth(T, m, d)
+    if (depth != null) {
+        println("Глубина: %.2f м"(depth))
+        // Для проверки выведем теоретическое время при найденной глубине
+        // (можно добавить функцию для самопроверки)
+    } else {
+        println("Расчёт не сошёлся")
+    }

examples/h2_basic.lyng

@@ -0,0 +1,43 @@
+import lyng.io.db
+import lyng.io.db.jdbc
+
+println("H2 JDBC demo: typed open, generic open, generated keys")
+
+val db = openH2("mem:lyng_h2_demo;DB_CLOSE_DELAY=-1")
+
+db.transaction { tx ->
+    tx.execute("create table if not exists person(id bigint auto_increment primary key, name varchar(120) not null, active boolean not null)")
+    tx.execute("delete from person")
+
+    val firstInsert = tx.execute(
+        "insert into person(name, active) values(?, ?)",
+        "Ada",
+        true
+    )
+    val firstId = firstInsert.getGeneratedKeys().toList()[0][0]
+    assertEquals(1, firstId)
+
+    tx.execute(
+        "insert into person(name, active) values(?, ?)",
+        "Linus",
+        false
+    )
+
+    val rows = tx.select("select id, name, active from person order by id").toList()
+    assertEquals(2, rows.size)
+    println("#" + rows[0]["id"] + " " + rows[0]["name"] + " active=" + rows[0]["active"])
+    println("#" + rows[1]["id"] + " " + rows[1]["name"] + " active=" + rows[1]["active"])
+}
+
+val genericDb = openDatabase(
+    "jdbc:h2:mem:lyng_h2_generic;DB_CLOSE_DELAY=-1",
+    Map()
+)
+
+val answer = genericDb.transaction { tx ->
+    tx.select("select 42 as answer").toList()[0]["answer"]
+}
+
+assertEquals(42, answer)
+println("Generic JDBC openDatabase(...) also works: answer=$answer")
+println("OK")

examples/pi-bench.lyng

@@ -0,0 +1,67 @@
+import lyng.time
+
+val WORK_SIZE = 500
+val THREADS = 1
+
+fn piSpigot(iThread: Int, n: Int) {
+    var piIter = 0
+    var pi = List.fill(n) { 0 }
+    val boxes = n * 10 / 3
+    var reminders = List.fill(boxes) { 2 }
+    var heldDigits = 0
+    for (i in 0..<n) {
+        var carriedOver = 0
+        var sum = 0
+        for (j in (boxes - 1) downTo 0) {
+            val denom = j * 2 + 1
+            reminders[j] *= 10
+            sum = reminders[j] + carriedOver
+            val quotient = sum / denom
+            reminders[j] = sum % denom
+            carriedOver = quotient * j
+        }
+        reminders[0] = sum % 10
+        var q = sum / 10
+        if (q == 9) {
+            ++heldDigits
+        } else if (q == 10) {
+            q = 0
+            for (k in 1..heldDigits) {
+                var replaced = pi[i - k]
+                if (replaced == 9) {
+                    replaced = 0
+                } else {
+                    ++replaced
+                }
+                pi[i - k] = replaced
+            }
+            heldDigits = 1
+        } else {
+            heldDigits = 1
+        }
+        pi[piIter] = q
+        ++piIter
+    }
+
+    var res = ""
+    for (i in (n - 8)..<n) {
+        res += pi[i]
+    }
+
+    println(iThread.toString() + ": " + res)
+    res
+}
+
+for( r in 0..100 ) {
+    val t0 = Instant()
+
+    println("piBench (lyng): THREADS = " + THREADS + ", WORK_SIZE = " + WORK_SIZE)
+    for (i in 0..<THREADS) {
+        piSpigot(i, WORK_SIZE)
+    }
+
+    val dt = Instant() - t0
+
+    println("all done, dt = ", dt)
+    delay(800)
+}

examples/pi-bench.py

@@ -0,0 +1,83 @@
+#  Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+import time
+from multiprocessing import Process
+
+def piSpigot(iThread, nx):
+    piIter = 0
+    pi = [None] * nx
+    boxes = nx * 10 // 3
+    reminders = [None]*boxes
+    i = 0
+    while i < boxes:
+        reminders[i] = 2
+        i += 1
+    heldDigits = 0
+    i = 0
+    while i < nx:
+        carriedOver = 0
+        sum = 0
+        j = boxes - 1
+        while j >= 0:
+            reminders[j] *= 10
+            sum = reminders[j] + carriedOver
+            quotient = sum // (j * 2 + 1)
+            reminders[j] = sum % (j * 2 + 1)
+            carriedOver = quotient * j
+            j -= 1
+        reminders[0] = sum % 10
+        q = sum // 10
+        if q == 9:
+            heldDigits += 1
+        elif q == 10:
+            q = 0
+            k = 1
+            while k <= heldDigits:
+                replaced = pi[i - k]
+                if replaced == 9:
+                    replaced = 0
+                else:
+                    replaced += 1
+                pi[i - k] = replaced
+                k += 1
+            heldDigits = 1
+        else:
+            heldDigits = 1
+        pi[piIter] = q
+        piIter += 1
+        i += 1
+    res = ""
+    for i in range(len(pi)-8, len(pi), 1):
+        res += str(pi[i])
+    print(str(iThread) + ": " + res)
+
+def createProcesses():
+    THREADS = 1
+    WORK_SIZE = 500
+    print("piBench (python3): THREADS = " + str(THREADS) + ", WORK_SIZE = " + str(WORK_SIZE))
+    pa = []
+    for i in range(THREADS):
+        p = Process(target=piSpigot, args=(i, WORK_SIZE))
+        p.start()
+        pa.append(p)
+    for p in pa:
+        p.join()
+
+if __name__ == "__main__":
+    t1 = time.time()
+    createProcesses()
+    dt = time.time() - t1
+    print("total time: %i ms" % (dt*1000))

examples/postgres_basic.lyng

@@ -0,0 +1,71 @@
+import lyng.io.db.jdbc
+
+/*
+    PostgreSQL JDBC demo.
+
+    Usage:
+    lyng examples/postgres_basic.lyng [jdbc-url] [user] [password]
+
+    Typical local URL:
+    jdbc:postgresql://127.0.0.1/postgres
+*/
+
+fun cliArgs(): List<String> {
+    val result: List<String> = []
+    for (raw in ARGV as List) {
+        result.add(raw as String)
+    }
+    return result
+}
+
+val argv = cliArgs()
+val URL = if (argv.size > 0) argv[0] else "jdbc:postgresql://127.0.0.1/postgres"
+val USER = if (argv.size > 1) argv[1] else ""
+val PASSWORD = if (argv.size > 2) argv[2] else ""
+
+println("PostgreSQL JDBC demo: typed open, generated keys, nested transaction")
+
+val db = openPostgres(URL, USER, PASSWORD)
+
+db.transaction { tx ->
+    tx.execute("create table if not exists lyng_pg_demo(id bigserial primary key, title text not null, done boolean not null)")
+    tx.execute("delete from lyng_pg_demo")
+
+    val firstInsert = tx.execute(
+        "insert into lyng_pg_demo(title, done) values(?, ?)",
+        "Verify PostgreSQL JDBC support",
+        false
+    )
+    val firstId = firstInsert.getGeneratedKeys().toList()[0][0]
+    println("First generated id=" + firstId)
+
+    tx.execute(
+        "insert into lyng_pg_demo(title, done) values(?, ?)",
+        "Review documentation",
+        true
+    )
+
+    try {
+        tx.transaction { inner ->
+            inner.execute(
+                "insert into lyng_pg_demo(title, done) values(?, ?)",
+                "This row is rolled back",
+                false
+            )
+            throw IllegalStateException("rollback nested")
+        }
+    } catch (_: IllegalStateException) {
+        println("Nested transaction rolled back as expected")
+    }
+
+    val rows = tx.select("select id, title, done from lyng_pg_demo order by id").toList()
+    for (row in rows) {
+        println("#" + row["id"] + " " + row["title"] + " done=" + row["done"])
+    }
+
+    val count = tx.select("select count(*) as count from lyng_pg_demo").toList()[0]["count"]
+    assertEquals(2, count)
+    println("Visible rows after nested rollback: " + count)
+}
+
+println("OK")

examples/sqlite_basic.lyng

@@ -0,0 +1,89 @@
+import lyng.io.db
+import lyng.io.db.sqlite
+import lyng.time
+
+println("SQLite demo: typed open, generic open, result sets, generated keys, nested rollback")
+
+// The typed helper is the simplest entry point when you know you want SQLite.
+val db = openSqlite(":memory:")
+
+db.transaction { tx ->
+    // Keep schema creation and data changes inside one transaction block.
+    tx.execute("create table task(id integer primary key autoincrement, title text not null, done integer not null, due_date date not null)")
+
+    // execute(...) is for side-effect statements. Generated keys are read from
+    // ExecutionResult rather than from a synthetic row-returning INSERT.
+    val firstInsert = tx.execute(
+        "insert into task(title, done, due_date) values(?, ?, ?)",
+        "Write a SQLite example",
+        false,
+        Date(2026, 4, 15)
+    )
+    val firstGeneratedKeys = firstInsert.getGeneratedKeys()
+    val firstId = firstGeneratedKeys.toList()[0][0]
+    assertEquals(1, firstId)
+
+    tx.execute(
+        "insert into task(title, done, due_date) values(?, ?, ?)",
+        "Review the DB API",
+        true,
+        Date(2026, 4, 16)
+    )
+
+    // Nested transactions are real savepoints. If the inner block fails,
+    // only the nested work is rolled back.
+    try {
+        tx.transaction { inner ->
+            inner.execute(
+                "insert into task(title, done, due_date) values(?, ?, ?)",
+                "This row is rolled back",
+                false,
+                Date(2026, 4, 17)
+            )
+            throw IllegalStateException("demonstrate nested rollback")
+        }
+    } catch (_: IllegalStateException) {
+        println("Nested transaction rolled back as expected")
+    }
+
+    // select(...) is for row-producing statements. ResultSet exposes metadata,
+    // cheap emptiness checks, iteration, and conversion to a plain list.
+    val tasks = tx.select("select id, title, done, due_date from task order by id")
+    assertEquals(false, tasks.isEmpty())
+    assertEquals(2, tasks.size())
+
+    println("Columns:")
+    for (column in tasks.columns) {
+        println("  " + column.name + " -> " + column.sqlType + " (native " + column.nativeType + ")")
+    }
+
+    val taskRows = tasks.toList()
+
+    println("Rows:")
+    for (row in taskRows) {
+        // Name lookups are case-insensitive and values are already converted.
+        println("  #" + row["ID"] + " " + row["title"] + " done=" + row["done"] + " due=" + row["due_date"])
+    }
+
+    // toList() materializes detached rows that stay usable after transaction close.
+    val snapshot = tx.select("select title, due_date from task order by id").toList()
+    assertEquals("Write a SQLite example", snapshot[0]["title"])
+    assertEquals(Date(2026, 4, 16), snapshot[1]["due_date"])
+
+    val count = tx.select("select count(*) as count from task").toList()[0]["count"]
+    assertEquals(2, count)
+    println("Visible rows after nested rollback: $count")
+}
+
+// The generic entry point stays useful for config-driven code.
+val genericDb = openDatabase(
+    "sqlite::memory:",{ foreignKeys: true, busyTimeoutMillis: 1000 }
+)
+
+val answer = genericDb.transaction { tx ->
+    tx.select("select 42 as answer").toList()[0]["answer"]
+}
+
+assertEquals(42, answer)
+println("Generic openDatabase(...) also works: answer=$answer")
+println("OK")

examples/tcp-server.lyng

@@ -0,0 +1,68 @@
+import lyng.buffer
+import lyng.io.net
+
+val host = "127.0.0.1"
+val port = 8092
+val N = 5
+val server = Net.tcpListen(port, host)
+println("start tcp server at $host:$port")
+
+fun serveClient(client: TcpSocket) = launch {
+    try {
+        while (true) {
+            val data = client.read()
+            if (data == null) break
+            var line = (data as Buffer).decodeUtf8()
+            line = "[" + client.remoteAddress() + "]> " + line
+            println(line)
+        }
+    } catch (e) {
+        println("ERROR [reader]: " + e)
+    }
+}
+
+fun serveRequests(server: TcpServer) = launch {
+    val readers = []
+    try {
+        for (i in 0..<5) {
+            val client = server.accept()
+            println("accept new connection: " + client.remoteAddress())
+            readers.add(serveClient(client as TcpSocket))
+        }
+    } catch (e) {
+        println("ERROR [listener]: " + e)
+    } finally {
+        server.close()
+    }
+    for (i in 0..<readers.size) {
+        val reader = readers[i]
+        (reader as Deferred).await()
+    }
+}
+
+val srv = serveRequests(server as TcpServer)
+
+var clients = []
+for (i in 0..<N) {
+    //delay(500)
+    clients.add(launch {
+        try{
+            val socket = Net.tcpConnect(host, port)
+            socket.writeUtf8("ping1ping2ping3ping4ping5")
+            socket.flush()
+            socket.close()
+        } catch (e) {
+            println("ERROR [client]: " + e)
+        }
+    })
+}
+
+for (i in 0..<clients.size) {
+    val c = clients[i]
+    (c as Deferred).await()
+    println("client done")
+}
+
+srv.await()
+delay(10000)
+println("FIN")

examples/tcpserver.lyng

@@ -0,0 +1,60 @@
+import lyng.io.net
+
+val host = "127.0.0.1"
+val clientCount = 1000
+val clientWindow = 128
+val server = Net.tcpListen(0, host, clientWindow, true)
+val port = server.localAddress().port
+
+fun payloadFor(index: Int) = "$index:${Random.nextInt()}:${Random.nextInt()}"
+
+val serverJob = launch {
+    try {
+        while (true) {
+            val client = server.accept()
+            launch {
+                try {
+                    client.readLine()?.let { source ->
+                        client.writeUtf8("pong: $source\n")
+                        client.flush()
+                    }
+                } finally {
+                    client.close()
+                }
+            }
+        }
+    } catch (e) {
+        if (server.isOpen()) {
+            throw e
+        }
+    } finally {
+        if (server.isOpen()) {
+            server.close()
+        }
+    }
+}
+
+var completed = 0
+for (batchStart in 0..<clientCount step clientWindow) {
+    val batchEnd = if (batchStart + clientWindow < clientCount) batchStart + clientWindow else clientCount
+    val replies = (batchStart..<batchEnd).map { index ->
+        val payload = payloadFor(index)
+        launch {
+            val socket = Net.tcpConnect(host, port) as TcpSocket
+            try {
+                socket.writeUtf8(payload + "\n")
+                socket.flush()
+                val reply = socket.readLine()
+                assertEquals("pong: $payload", reply)
+            } finally {
+                socket.close()
+            }
+        }
+    }.joinAll()
+    completed += replies.size
+}
+
+assertEquals(clientCount, completed)
+server.close()
+serverJob.await()
+println("OK: $clientCount concurrent tcp clients")

examples/tetris_console.lyng

@@ -0,0 +1,822 @@
+#!/usr/bin/env lyng
+
+/*
+ * Lyng Console Tetris (interactive sample)
+ *
+ * Controls:
+ *  - Left/Right arrows or A/D: move
+ *  - Up arrow or W: rotate
+ *  - Down arrow or S: soft drop
+ *  - Space: hard drop
+ *  - P or Escape: pause
+ *  - Q: quit
+
+    Tsted to score:
+sergeych@sergeych-XPS-17-9720:~$ ~/dev/lyng/examples/tetris_console.lyng
+Bye.
+Score: 435480
+Lines: 271
+Level: 28
+Ssergeych@sergeych-XPS-17-9720:~$
+ */
+
+import lyng.io.console
+import lyng.io.fs
+
+val MIN_COLS = 56
+val MIN_ROWS = 24
+val PANEL_WIDTH = 24
+val BOARD_MARGIN_ROWS = 5
+val BOARD_MIN_W = 10
+val BOARD_MAX_W = 16
+val BOARD_MIN_H = 16
+val BOARD_MAX_H = 28
+val LEVEL_LINES_STEP = 10
+val DROP_FRAMES_BASE = 15
+val DROP_FRAMES_MIN = 3
+val FRAME_DELAY_MS = 35
+val RESIZE_WAIT_MS = 250
+val MAX_PENDING_INPUTS = 64
+val ROTATION_KICKS = [0, -1, 1, -2, 2]
+val ANSI_ESC = "\u001b["
+val ANSI_RESET = ANSI_ESC + "0m"
+val ERROR_LOG_PATH = "/tmp/lyng_tetris_errors.log"
+val UNICODE_BLOCK = "██"
+val UNICODE_TOP_LEFT = "┌"
+val UNICODE_TOP_RIGHT = "┐"
+val UNICODE_BOTTOM_LEFT = "└"
+val UNICODE_BOTTOM_RIGHT = "┘"
+val UNICODE_HORIZONTAL = "──"
+val UNICODE_VERTICAL = "│"
+val UNICODE_DOT = "· "
+val PIECES: List<Piece> = []
+
+type Cell = List<Int>
+type Rotation = List<Cell>
+type Rotations = List<Rotation>
+type Row = List<Int>
+type Board = List<Row>
+
+class Piece(val name: String, val rotations: Rotations) {}
+class RotateResult(val ok: Bool, val rot: Int, val px: Int) {}
+class GameState(
+    pieceId0: Int,
+    nextId0: Int,
+    next2Id0: Int,
+    px0: Int,
+    py0: Int,
+) {
+    var pieceId = pieceId0
+    var nextId = nextId0
+    var next2Id = next2Id0
+    var rot = 0
+    var px = px0
+    var py = py0
+    var score = 0
+    var totalLines = 0
+    var level = 1
+    var running = true
+    var gameOver = false
+    var paused = false
+}
+class LoopFrame(val resized: Bool, val originRow: Int, val originCol: Int) {}
+class InputBuffer {
+    private val mutex: Mutex = Mutex()
+    private val items: List<String> = []
+
+    fun push(value: String): Void {
+        mutex.withLock {
+            if (items.size >= MAX_PENDING_INPUTS) {
+                items.removeAt(0)
+            }
+            items.add(value)
+        }
+    }
+
+    fun drain(): List<String> {
+        val out: List<String> = []
+        mutex.withLock {
+            while (items.size > 0) {
+                out.add(items[0])
+                items.removeAt(0)
+            }
+        }
+        out
+    }
+}
+
+fun clearAndHome() {
+    Console.clear()
+    Console.home()
+}
+
+fun logError(message: String, err: Object?): Void {
+    try {
+        var details = ""
+        if (err != null) {
+            try {
+                details = ": " + err
+            } catch (_: Object) {
+                details = ": <error-format-failed>"
+            }
+        }
+        Path(ERROR_LOG_PATH).appendUtf8(message + details + "\n")
+    } catch (_: Object) {
+        // Never let logging errors affect gameplay.
+    }
+}
+
+fun repeatText(s: String, n: Int): String {
+    var out: String = ""
+    for (i in 0..<n) {
+        out += s
+    }
+    out
+}
+
+fun max<T>(a: T, b: T): T = if (a > b) a else b
+
+fun emptyRow(width: Int): Row {
+    val r: Row = []
+    for (x in 0..<width) {
+        r.add(0)
+    }
+    r
+}
+
+fun createBoard(width: Int, height: Int): Board {
+    val b: Board = []
+    for (y in 0..<height) {
+        b.add(emptyRow(width))
+    }
+    b
+}
+
+fun colorize(text: String, sgr: String, useColor: Bool): String {
+    if (!useColor) return text
+    ANSI_ESC + sgr + "m" + text + ANSI_RESET
+}
+
+fun blockText(pieceId: Int, useColor: Bool): String {
+    if (pieceId <= 0) return "  "
+    val sgr = when (pieceId) {
+        1 -> "36" // I
+        2 -> "33" // O
+        3 -> "35" // T
+        4 -> "32" // S
+        5 -> "31" // Z
+        6 -> "34" // J
+        7 -> "93" // L
+        else -> "37"
+    }
+    colorize(UNICODE_BLOCK, sgr, useColor)
+}
+
+fun emptyCellText(useColor: Bool): String {
+    colorize(UNICODE_DOT, "90", useColor)
+}
+
+fun canPlace(board: Board, boardW: Int, boardH: Int, pieceId: Int, rot: Int, px: Int, py: Int): Bool {
+    try {
+        if (pieceId < 1 || pieceId > 7) return false
+        val piece: Piece = PIECES[pieceId - 1]
+        if (rot < 0 || rot >= piece.rotations.size) return false
+        val cells = piece.rotations[rot]
+
+        for (cell in cells) {
+            val x = px + cell[0]
+            val y = py + cell[1]
+
+            if (x < 0 || x >= boardW) return false
+            if (y >= boardH) return false
+
+            if (y >= 0) {
+                if (y >= board.size) return false
+                val row = board[y]
+                if (row == null) return false
+                if (row[x] != 0) return false
+            }
+        }
+        true
+    } catch (_: Object) {
+        false
+    }
+}
+
+fun lockPiece(board: Board, pieceId: Int, rot: Int, px: Int, py: Int): Void {
+    val piece: Piece = PIECES[pieceId - 1]
+    val cells = piece.rotations[rot]
+
+    for (cell in cells) {
+        val x = px + cell[0]
+        val y = py + cell[1]
+
+        if (y >= 0 && y < board.size) {
+            val row = board[y]
+            if (row != null) {
+                row[x] = pieceId
+            }
+        }
+    }
+}
+
+fun clearCompletedLines(board: Board, boardW: Int, boardH: Int): Int {
+    val b = board
+    var y = boardH - 1
+    var cleared = 0
+
+    while (y >= 0) {
+        if (y >= b.size) {
+            y--
+            continue
+        }
+        val row = b[y]
+        if (row == null) {
+            b.removeAt(y)
+            b.insertAt(0, emptyRow(boardW))
+            cleared++
+            continue
+        }
+        var full = true
+        for (x in 0..<boardW) {
+            if (row[x] == 0) {
+                full = false
+                break
+            }
+        }
+
+        if (full) {
+            b.removeAt(y)
+            b.insertAt(0, emptyRow(boardW))
+            cleared++
+        } else {
+            y--
+        }
+    }
+
+    cleared
+}
+
+fun activeCellId(pieceId: Int, rot: Int, px: Int, py: Int, x: Int, y: Int): Int {
+    val piece: Piece = PIECES[pieceId - 1]
+    val cells = piece.rotations[rot]
+
+    for (cell in cells) {
+        val ax = px + cell[0]
+        val ay = py + cell[1]
+        if (ax == x && ay == y) return pieceId
+    }
+
+    0
+}
+
+fun tryRotateCw(board: Board, boardW: Int, boardH: Int, pieceId: Int, rot: Int, px: Int, py: Int): RotateResult {
+    val piece: Piece = PIECES[pieceId - 1]
+    val rotations = piece.rotations.size
+    val nr = (rot + 1) % rotations
+    for (kx in ROTATION_KICKS) {
+        val nx = px + kx
+        if (canPlace(board, boardW, boardH, pieceId, nr, nx, py) == true) {
+            return RotateResult(true, nr, nx)
+        }
+    }
+    RotateResult(false, rot, px)
+}
+
+fun nextPreviewLines(pieceId: Int, useColor: Bool): List<String> {
+    val out: List<String> = []
+    if (pieceId <= 0) {
+        for (i in 0..<4) {
+            out.add("        ")
+        }
+        return out
+    }
+
+    val piece: Piece = PIECES[pieceId - 1]
+    val cells = piece.rotations[0]
+
+    for (y in 0..<4) {
+        var line = ""
+        for (x in 0..<4) {
+            var filled = false
+            for (cell in cells) {
+                if (cell[0] == x && cell[1] == y) {
+                    filled = true
+                    break
+                }
+            }
+            line += if (filled) blockText(pieceId, useColor) else "  "
+        }
+        out.add(line)
+    }
+    out
+}
+
+fun render(
+    state: GameState,
+    board: Board,
+    boardW: Int,
+    boardH: Int,
+    prevFrameLines: List<String>,
+    originRow: Int,
+    originCol: Int,
+    useColor: Bool,
+): List<String> {
+    val bottomBorder = UNICODE_BOTTOM_LEFT + repeatText(UNICODE_HORIZONTAL, boardW) + UNICODE_BOTTOM_RIGHT
+
+    val panel: List<String> = []
+    val nextPiece: Piece = PIECES[state.nextId - 1]
+    val next2Piece: Piece = PIECES[state.next2Id - 1]
+    val nextName = nextPiece.name
+    val next2Name = next2Piece.name
+    val preview = nextPreviewLines(state.nextId, useColor)
+
+    panel.add("Lyng Tetris")
+    panel.add("")
+    panel.add("Score: " + state.score)
+    panel.add("Lines: " + state.totalLines)
+    panel.add("Level: " + state.level)
+    panel.add("")
+    panel.add("Next: " + nextName)
+    panel.add("")
+    for (pl in preview) panel.add(pl)
+    panel.add("After: " + next2Name)
+    panel.add("")
+    panel.add("Keys:")
+    panel.add("A/D or arrows")
+    panel.add("W/Up: rotate")
+    panel.add("S/Down: drop")
+    panel.add("Space: hard drop")
+    panel.add("P/Esc: pause")
+    panel.add("Q: quit")
+
+    val frameLines: List<String> = []
+
+    for (y in 0..<boardH) {
+        var line = UNICODE_VERTICAL
+
+        for (x in 0..<boardW) {
+            val a = activeCellId(state.pieceId, state.rot, state.px, state.py, x, y)
+            val row = if (y < board.size) board[y] else null
+            val b = if (row == null) 0 else row[x]
+            val id = if (a > 0) a else b
+            line += if (id > 0) blockText(id, useColor) else emptyCellText(useColor)
+        }
+
+        line += UNICODE_VERTICAL
+
+        if (y < panel.size) {
+            line += "   " + panel[y]
+        }
+
+        frameLines.add(line)
+    }
+
+    frameLines.add(bottomBorder)
+
+    for (i in 0..<frameLines.size) {
+        val line = frameLines[i]
+        val old = if (i < prevFrameLines.size) prevFrameLines[i] else null
+        if (old != line) {
+            Console.moveTo(originRow + i, originCol)
+            Console.clearLine()
+            Console.write(line)
+        }
+    }
+    frameLines
+}
+
+fun fitLine(line: String, width: Int): String {
+    val maxLen = if (width > 0) width else 0
+    if (maxLen <= 0) return ""
+    if (line.size >= maxLen) return line[..<maxLen]
+    line + repeatText(" ", maxLen - line.size)
+}
+
+fun renderPauseOverlay(
+    originRow: Int,
+    originCol: Int,
+    boardW: Int,
+    boardH: Int,
+): Void {
+    val contentWidth = boardW * 2 + 2 + 3 + PANEL_WIDTH
+    val contentHeight = boardH + 1
+
+    val lines: List<String> = []
+    lines.add("PAUSED")
+    lines.add("")
+    lines.add("Any key: continue game")
+    lines.add("Esc: exit game")
+    lines.add("")
+    lines.add("Move: A/D or arrows")
+    lines.add("Rotate: W or Up")
+    lines.add("Drop: S/Down, Space hard drop")
+
+    var innerWidth = 0
+    for (line in lines) {
+        if (line.size > innerWidth) innerWidth = line.size
+    }
+    innerWidth += 4
+    val maxInner = max(12, contentWidth - 2)
+    if (innerWidth > maxInner) innerWidth = maxInner
+    if (innerWidth % 2 != 0) innerWidth--
+
+    val boxWidth = innerWidth + 2
+    val boxHeight = lines.size + 2
+
+    val left = originCol + max(0, (contentWidth - boxWidth) / 2)
+    val top = originRow + max(0, (contentHeight - boxHeight) / 2)
+
+    val topBorder = UNICODE_TOP_LEFT + repeatText(UNICODE_HORIZONTAL, innerWidth / 2) + UNICODE_TOP_RIGHT
+    val bottomBorder = UNICODE_BOTTOM_LEFT + repeatText(UNICODE_HORIZONTAL, innerWidth / 2) + UNICODE_BOTTOM_RIGHT
+
+    Console.moveTo(top, left)
+    Console.write(topBorder)
+
+    for (i in 0..<lines.size) {
+        Console.moveTo(top + 1 + i, left)
+        Console.write(UNICODE_VERTICAL + fitLine("  " + lines[i], innerWidth) + UNICODE_VERTICAL)
+    }
+
+    Console.moveTo(top + boxHeight - 1, left)
+    Console.write(bottomBorder)
+}
+
+fun waitForMinimumSize(minCols: Int, minRows: Int): Object {
+    while (true) {
+        val g = Console.geometry()
+        val cols = g?.columns ?: 0
+        val rows = g?.rows ?: 0
+
+        if (cols >= minCols && rows >= minRows) return g
+
+        clearAndHome()
+        val lines: List<String> = []
+        lines.add("Lyng Tetris needs at least %sx%s terminal size."(minCols, minRows))
+        lines.add("Current: %sx%s"(cols, rows))
+        lines.add("Resize the console window to continue.")
+
+        val visibleLines = if (rows < lines.size) rows else lines.size
+        if (visibleLines > 0) {
+            val startRow = max(1, ((rows - visibleLines) / 2) + 1)
+            for (i in 0..<visibleLines) {
+                val line = lines[i]
+                val startCol = max(1, ((cols - line.size) / 2) + 1)
+                Console.moveTo(startRow + i, startCol)
+                Console.clearLine()
+                Console.write(line)
+            }
+            Console.flush()
+        }
+        delay(RESIZE_WAIT_MS)
+    }
+}
+
+fun scoreForLines(cleared: Int, level: Int): Int {
+    when (cleared) {
+        1 -> 100 * level
+        2 -> 300 * level
+        3 -> 500 * level
+        4 -> 800 * level
+        else -> 0
+    }
+}
+
+// Classic 7 tetrominoes, minimal rotations per piece.
+fun cell(x: Int, y: Int): Cell { [x, y] }
+fun rot(a: Cell, b: Cell, c: Cell, d: Cell): Rotation {
+    val r: Rotation = []
+    r.add(a)
+    r.add(b)
+    r.add(c)
+    r.add(d)
+    r
+}
+
+val iRots: Rotations = []
+iRots.add(rot(cell(0,1), cell(1,1), cell(2,1), cell(3,1)))
+iRots.add(rot(cell(2,0), cell(2,1), cell(2,2), cell(2,3)))
+PIECES.add(Piece("I", iRots))
+
+val oRots: Rotations = []
+oRots.add(rot(cell(1,0), cell(2,0), cell(1,1), cell(2,1)))
+PIECES.add(Piece("O", oRots))
+
+val tRots: Rotations = []
+tRots.add(rot(cell(1,0), cell(0,1), cell(1,1), cell(2,1)))
+tRots.add(rot(cell(1,0), cell(1,1), cell(2,1), cell(1,2)))
+tRots.add(rot(cell(0,1), cell(1,1), cell(2,1), cell(1,2)))
+tRots.add(rot(cell(1,0), cell(0,1), cell(1,1), cell(1,2)))
+PIECES.add(Piece("T", tRots))
+
+val sRots: Rotations = []
+sRots.add(rot(cell(1,0), cell(2,0), cell(0,1), cell(1,1)))
+sRots.add(rot(cell(1,0), cell(1,1), cell(2,1), cell(2,2)))
+PIECES.add(Piece("S", sRots))
+
+val zRots: Rotations = []
+zRots.add(rot(cell(0,0), cell(1,0), cell(1,1), cell(2,1)))
+zRots.add(rot(cell(2,0), cell(1,1), cell(2,1), cell(1,2)))
+PIECES.add(Piece("Z", zRots))
+
+val jRots: Rotations = []
+jRots.add(rot(cell(0,0), cell(0,1), cell(1,1), cell(2,1)))
+jRots.add(rot(cell(1,0), cell(2,0), cell(1,1), cell(1,2)))
+jRots.add(rot(cell(0,1), cell(1,1), cell(2,1), cell(2,2)))
+jRots.add(rot(cell(1,0), cell(1,1), cell(0,2), cell(1,2)))
+PIECES.add(Piece("J", jRots))
+
+val lRots: Rotations = []
+lRots.add(rot(cell(2,0), cell(0,1), cell(1,1), cell(2,1)))
+lRots.add(rot(cell(1,0), cell(1,1), cell(1,2), cell(2,2)))
+lRots.add(rot(cell(0,1), cell(1,1), cell(2,1), cell(0,2)))
+lRots.add(rot(cell(0,0), cell(1,0), cell(1,1), cell(1,2)))
+PIECES.add(Piece("L", lRots))
+
+if (!Console.isSupported()) {
+    println("Console API is not supported in this runtime.")
+    void
+} else if (!Console.isTty()) {
+    println("This sample needs an interactive terminal (TTY).")
+    void
+} else {
+    waitForMinimumSize(MIN_COLS, MIN_ROWS)
+
+    val g0 = Console.geometry()
+    val cols = g0?.columns ?: MIN_COLS
+    val rows = g0?.rows ?: MIN_ROWS
+
+    val boardW = clamp((cols - PANEL_WIDTH) / 2, BOARD_MIN_W..BOARD_MAX_W)
+    val boardH = clamp(rows - BOARD_MARGIN_ROWS, BOARD_MIN_H..BOARD_MAX_H)
+
+    val board: Board = createBoard(boardW, boardH)
+
+    fun nextPieceId() {
+        Random.next(1..7)
+    }
+
+    val state: GameState = GameState(
+        nextPieceId(),
+        nextPieceId(),
+        nextPieceId(),
+        (boardW / 2) - 2,
+        -1,
+    )
+    var prevFrameLines: List<String> = []
+
+    var forceRedraw = false
+    val inputBuffer: InputBuffer = InputBuffer()
+
+    val rawModeEnabled = Console.setRawMode(true)
+    if (!rawModeEnabled) {
+        println("Raw keyboard mode is not available in this terminal/runtime.")
+        println("Use jlyng in an interactive terminal with raw input support.")
+        void
+    } else {
+        val useColor = Console.ansiLevel() != ConsoleAnsiLevel.NONE
+        Console.enterAltScreen()
+        Console.setCursorVisible(false)
+        clearAndHome()
+
+        fun resetActivePiece(s: GameState): Void {
+            s.pieceId = nextPieceId()
+            s.rot = 0
+            s.px = (boardW / 2) - 2
+            s.py = -1
+        }
+
+        fun applyKeyInput(s: GameState, key: String): Void {
+            try {
+                if (key == "__CTRL_C__") {
+                    s.running = false
+                }
+                else if (s.paused) {
+                    if (key == "Escape") {
+                        s.running = false
+                    } else {
+                        s.paused = false
+                        forceRedraw = true
+                    }
+                }
+                else if (key == "p" || key == "P" || key == "Escape") {
+                    s.paused = true
+                    forceRedraw = true
+                }
+                else if (key == "q" || key == "Q") {
+                    s.running = false
+                }
+                else if (key == "ArrowLeft" || key == "a" || key == "A") {
+                    if (canPlace(board, boardW, boardH, s.pieceId, s.rot, s.px - 1, s.py) == true) s.px--
+                }
+                else if (key == "ArrowRight" || key == "d" || key == "D") {
+                    if (canPlace(board, boardW, boardH, s.pieceId, s.rot, s.px + 1, s.py) == true) s.px++
+                }
+                else if (key == "ArrowUp" || key == "w" || key == "W") {
+                    val rr: RotateResult = tryRotateCw(board, boardW, boardH, s.pieceId, s.rot, s.px, s.py)
+                    if (rr.ok) {
+                        s.rot = rr.rot
+                        s.px = rr.px
+                    }
+                }
+                else if (key == "ArrowDown" || key == "s" || key == "S") {
+                    if (canPlace(board, boardW, boardH, s.pieceId, s.rot, s.px, s.py + 1) == true) {
+                        s.py++
+                        s.score++
+                    }
+                }
+                else if (key == " ") {
+                    while (canPlace(board, boardW, boardH, s.pieceId, s.rot, s.px, s.py + 1) == true) {
+                        s.py++
+                        s.score += 2
+                    }
+                }
+            } catch (inputErr: Object) {
+                logError("applyKeyInput recovered after error", inputErr)
+                resetActivePiece(s)
+            }
+        }
+
+        var inputRunning = true
+        launch {
+            while (inputRunning) {
+                try {
+                    for (ev in Console.events()) {
+                        if (!inputRunning) break
+
+                        // Isolate per-event failures so one bad event does not unwind the stream.
+                        try {
+                            if (ev is ConsoleKeyEvent) {
+                                val ke = ev as ConsoleKeyEvent
+                                if (ke.type == ConsoleEventType.KEY_DOWN) {
+                                    var key = ""
+                                    var ctrl = false
+                                    try {
+                                        key = ke.key
+                                    } catch (keyErr: Object) {
+                                        logError("Input key read error", keyErr)
+                                    }
+                                    try {
+                                        ctrl = ke.ctrl == true
+                                    } catch (_: Object) {
+                                        ctrl = false
+                                    }
+                                    if (key == "") {
+                                        logError("Dropped key event with empty/null key", null)
+                                        continue
+                                    }
+                                    val mapped = if (ctrl && (key == "c" || key == "C")) "__CTRL_C__" else key
+                                    inputBuffer.push(mapped)
+                                }
+                            }
+                        } catch (eventErr: Object) {
+                            // Keep the input stream alive; report for diagnostics.
+                            logError("Input event error", eventErr)
+                        }
+                    }
+                } catch (err: Object) {
+                    // Recover stream-level failures by recreating event stream in next loop turn.
+                    if (!inputRunning) break
+                    logError("Input stream recovered after error", err)
+                    Console.setRawMode(true)
+                    delay(50)
+                }
+            }
+        }
+
+        fun pollLoopFrame(): LoopFrame? {
+            val g = Console.geometry()
+            val c = g?.columns ?: 0
+            val r = g?.rows ?: 0
+            if (c < MIN_COLS || r < MIN_ROWS) {
+                waitForMinimumSize(MIN_COLS, MIN_ROWS)
+                clearAndHome()
+                prevFrameLines = []
+                return null
+            }
+
+            val contentCols = boardW * 2 + 2 + 3 + PANEL_WIDTH
+            val contentRows = boardH + 1
+            val requiredCols = max(MIN_COLS, contentCols)
+            val requiredRows = max(MIN_ROWS, contentRows)
+            if (c < requiredCols || r < requiredRows) {
+                waitForMinimumSize(requiredCols, requiredRows)
+                clearAndHome()
+                prevFrameLines = []
+                return null
+            }
+
+            val originCol = max(1, ((c - contentCols) / 2) + 1)
+            val originRow = max(1, ((r - contentRows) / 2) + 1)
+            LoopFrame(false, originRow, originCol)
+        }
+
+        fun advanceGravity(s: GameState, frame: Int): Int {
+            s.level = 1 + (s.totalLines / LEVEL_LINES_STEP)
+            val dropEvery = max(DROP_FRAMES_MIN, DROP_FRAMES_BASE - s.level)
+
+            var nextFrame = frame + 1
+            if (nextFrame < dropEvery) return nextFrame
+            nextFrame = 0
+
+            if (canPlace(board, boardW, boardH, s.pieceId, s.rot, s.px, s.py + 1) == true) {
+                s.py++
+                return nextFrame
+            }
+
+            lockPiece(board, s.pieceId, s.rot, s.px, s.py)
+
+            val cleared = clearCompletedLines(board, boardW, boardH)
+            if (cleared > 0) {
+                s.totalLines += cleared
+                s.score += scoreForLines(cleared, s.level)
+            }
+
+            s.pieceId = s.nextId
+            s.nextId = s.next2Id
+            s.next2Id = nextPieceId()
+            s.rot = 0
+            s.px = (boardW / 2) - 2
+            s.py = -1
+
+            if (!canPlace(board, boardW, boardH, s.pieceId, s.rot, s.px, s.py)) {
+                s.gameOver = true
+            }
+            nextFrame
+        }
+
+        try {
+            if (!canPlace(board, boardW, boardH, state.pieceId, state.rot, state.px, state.py)) {
+                state.gameOver = true
+            }
+
+            var frame = 0
+            var shouldStop = false
+            var prevOriginRow = -1
+            var prevOriginCol = -1
+            var prevPaused = false
+            while (!shouldStop) {
+                val frameData = pollLoopFrame()
+                if (frameData == null) {
+                    frame = 0
+                    prevPaused = false
+                    continue
+                }
+
+                val toApply = inputBuffer.drain()
+                if (toApply.size > 0) {
+                    for (k in toApply) {
+                        applyKeyInput(state, k)
+                        if (!state.running || state.gameOver) break
+                    }
+                }
+                if (!state.running || state.gameOver) {
+                    shouldStop = true
+                } else {
+                    val localForceRedraw = forceRedraw
+                    forceRedraw = false
+                    val movedOrigin = frameData.originRow != prevOriginRow || frameData.originCol != prevOriginCol
+                    if (frameData.resized || movedOrigin || localForceRedraw) {
+                        clearAndHome()
+                        prevFrameLines = []
+                    }
+                    prevOriginRow = frameData.originRow
+                    prevOriginCol = frameData.originCol
+                    if (!state.paused) {
+                        frame = advanceGravity(state, frame)
+                    }
+                    prevFrameLines = render(
+                        state,
+                        board,
+                        boardW,
+                        boardH,
+                        prevFrameLines,
+                        frameData.originRow,
+                        frameData.originCol,
+                        useColor
+                    )
+                    if (state.paused && (!prevPaused || frameData.resized || movedOrigin)) {
+                        renderPauseOverlay(frameData.originRow, frameData.originCol, boardW, boardH)
+                    }
+                    prevPaused = state.paused
+                }
+                Console.flush()
+                delay(FRAME_DELAY_MS)
+            }
+        } finally {
+            inputRunning = false
+            Console.setRawMode(false)
+            Console.setCursorVisible(true)
+            Console.leaveAltScreen()
+            Console.flush()
+        }
+
+        if (state.gameOver) {
+            println("Game over.")
+        } else {
+            println("Bye.")
+        }
+        println("Score: %s"(state.score))
+        println("Lines: %s"(state.totalLines))
+        println("Level: %s"(state.level))
+    }
+}

gradle.properties

@@ -1,5 +1,5 @@
 #
-# Copyright 2025 Sergey S. Chernov real.sergeych@gmail.com
+# Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -16,7 +16,7 @@
 #
 
 #Gradle
-org.gradle.jvmargs=-Xmx2048M -Dfile.encoding=UTF-8 -Dkotlin.daemon.jvm.options\="-Xmx2048M"
+org.gradle.jvmargs=-Xmx4096M -Dfile.encoding=UTF-8 -Dkotlin.daemon.jvm.options\="-Xmx2048M"
 org.gradle.caching=true
 org.gradle.configuration-cache=true
 #Kotlin
@@ -27,9 +27,6 @@ kotlin.mpp.enableCInteropCommonization=true
 android.useAndroidX=true
 android.nonTransitiveRClass=true
 
-# other
-kotlin.native.cacheKind.linuxX64=none
-
 # Workaround: Ensure Gradle uses a JDK with `jlink` available for AGP's JDK image transform.
 # On this environment, the system JDK 21 installation lacks `jlink`, causing
 # :lynglib:androidJdkImage to fail. Point Gradle to a JDK that includes `jlink`.
@@ -37,4 +34,6 @@ kotlin.native.cacheKind.linuxX64=none
 #org.gradle.java.home=/home/sergeych/.jdks/corretto-21.0.9
 android.experimental.lint.migrateToK2=false
 android.lint.useK2Uast=false
-kotlin.mpp.applyDefaultHierarchyTemplate=true
+kotlin.mpp.applyDefaultHierarchyTemplate=false
+
+org.gradle.parallel=true

gradle/libs.versions.toml

@@ -1,27 +1,59 @@
 [versions]
 agp = "8.5.2"
 clikt = "5.0.3"
-kotlin = "2.3.0"
+mordant = "3.0.2"
+kotlin = "2.3.20"
 android-minSdk = "24"
 android-compileSdk = "34"
 kotlinx-coroutines = "1.10.2"
+kotlinx-datetime = "0.6.1"
 mp_bintools = "0.3.2"
+ionspin-bignum = "0.3.10"
+multik = "0.3.0"
 firebaseCrashlyticsBuildtools = "3.0.3"
 okioVersion = "3.10.2"
 compiler = "3.2.0-alpha11"
+ktor = "3.3.1"
+slf4j = "2.0.17"
+sqlite-jdbc = "3.50.3.0"
+h2 = "2.4.240"
+postgresql = "42.7.8"
+testcontainers = "1.20.6"
+hikaricp = "6.2.1"
 
 [libraries]
 clikt = { module = "com.github.ajalt.clikt:clikt", version.ref = "clikt" }
+clikt-core = { module = "com.github.ajalt.clikt:clikt-core", version.ref = "clikt" }
 clikt-markdown = { module = "com.github.ajalt.clikt:clikt-markdown", version.ref = "clikt" }
+mordant-core = { module = "com.github.ajalt.mordant:mordant-core", version.ref = "mordant" }
+mordant-jvm-jna = { module = "com.github.ajalt.mordant:mordant-jvm-jna", version.ref = "mordant" }
 kotlin-test = { module = "org.jetbrains.kotlin:kotlin-test", version.ref = "kotlin" }
 kotlinx-coroutines-test = { module = "org.jetbrains.kotlinx:kotlinx-coroutines-test", version.ref = "kotlinx-coroutines" }
 kotlinx-coroutines-core = { module = "org.jetbrains.kotlinx:kotlinx-coroutines-core", version.ref = "kotlinx-coroutines" }
+kotlinx-datetime = { module = "org.jetbrains.kotlinx:kotlinx-datetime", version.ref = "kotlinx-datetime" }
 mp_bintools = { module = "net.sergeych:mp_bintools", version.ref = "mp_bintools" }
+ionspin-bignum = { module = "com.ionspin.kotlin:bignum", version.ref = "ionspin-bignum" }
+multik-default = { module = "org.jetbrains.kotlinx:multik-default", version.ref = "multik" }
 firebase-crashlytics-buildtools = { group = "com.google.firebase", name = "firebase-crashlytics-buildtools", version.ref = "firebaseCrashlyticsBuildtools" }
 okio = { module = "com.squareup.okio:okio", version.ref = "okioVersion" }
 okio-fakefilesystem = { module = "com.squareup.okio:okio-fakefilesystem", version.ref = "okioVersion" }
 okio-nodefilesystem = { module = "com.squareup.okio:okio-nodefilesystem", version.ref = "okioVersion" }
 compiler = { group = "androidx.databinding", name = "compiler", version.ref = "compiler" }
+ktor-client-core = { module = "io.ktor:ktor-client-core", version.ref = "ktor" }
+ktor-client-cio = { module = "io.ktor:ktor-client-cio", version.ref = "ktor" }
+ktor-client-curl = { module = "io.ktor:ktor-client-curl", version.ref = "ktor" }
+ktor-client-darwin = { module = "io.ktor:ktor-client-darwin", version.ref = "ktor" }
+ktor-client-js = { module = "io.ktor:ktor-client-js", version.ref = "ktor" }
+ktor-client-winhttp = { module = "io.ktor:ktor-client-winhttp", version.ref = "ktor" }
+ktor-client-websockets = { module = "io.ktor:ktor-client-websockets", version.ref = "ktor" }
+ktor-network = { module = "io.ktor:ktor-network", version.ref = "ktor" }
+slf4j-nop = { module = "org.slf4j:slf4j-nop", version.ref = "slf4j" }
+sqlite-jdbc = { module = "org.xerial:sqlite-jdbc", version.ref = "sqlite-jdbc" }
+h2 = { module = "com.h2database:h2", version.ref = "h2" }
+postgresql = { module = "org.postgresql:postgresql", version.ref = "postgresql" }
+testcontainers = { module = "org.testcontainers:testcontainers", version.ref = "testcontainers" }
+testcontainers-postgresql = { module = "org.testcontainers:postgresql", version.ref = "testcontainers" }
+hikaricp = { module = "com.zaxxer:HikariCP", version.ref = "hikaricp" }
 
 [plugins]
 androidLibrary = { id = "com.android.library", version.ref = "agp" }