v1.5.5 release

Fix escaping immutable closure captures
Add VPS deployment target and Ansible provisioning for lynglang.com
2026-04-23 15:18:47 +03:00 · 2026-04-22 18:50:49 +03:00 · 2026-04-22 12:31:01 +03:00 · 2026-04-22 08:56:02 +03:00 · 2026-04-21 23:25:07 +03:00 · 2026-04-21 21:37:04 +03:00
788 changed files with 146531 additions and 6068 deletions
@@ -0,0 +1 @@
 *.zip filter=lfs diff=lfs merge=lfs -text
@@ -1,54 +0,0 @@
 # This workflow uses actions that are not certified by GitHub.
 # They are provided by a third-party and are governed by
 # separate terms of service, privacy policy, and support
 # documentation.
 # This workflow will build a Java project with Gradle and cache/restore any dependencies to improve the workflow execution time
 # For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-java-with-gradle
 name: Java CI with Gradle
 on:
  push:
    branches: [ "main" ]
  pull_request:
    branches: [ "main" ]
  workflow_call:
 permissions:
  contents: read
 jobs:
  build:
    strategy:
      matrix:
        include:
          - target: iosSimulatorArm64Test
            os: macos-latest
          - target: jvmTest
            os: ubuntu-latest
          - target: linuxX64Test
            os: ubuntu-latest
          - target: testDebugUnitTest
            os: ubuntu-latest
          - target: testReleaseUnitTest
            os: ubuntu-latest
    runs-on: ${{ matrix.os }}
    steps:
    - uses: actions/checkout@v3
    - name: Validate Gradle Wrapper
      uses: gradle/actions/wrapper-validation@v3
    - uses: actions/cache@v3
      with:
        path: |
          ~/.konan
        key: ${{ runner.os }}-${{ hashFiles('**/.lock') }}
    - name: Set up JDK 17
      uses: actions/setup-java@v3
      with:
        java-version: '17'
        distribution: 'temurin'
    - name: Build with Gradle
      uses: gradle/gradle-build-action@ce999babab2de1c4b649dc15f0ee67e6246c994f
      with:
        arguments: ${{ matrix.target }}
@@ -1,25 +0,0 @@
 name: Publish
 on:
  release:
    types: [released, prereleased]
 jobs:
  publish:
    name: Release build and publish
    runs-on: macOS-latest
    steps:
      - name: Check out code
        uses: actions/checkout@v4
      - name: Set up JDK 21
        uses: actions/setup-java@v4
        with:
          distribution: 'zulu'
          java-version: 21
      - name: Publish to MavenCentral
        run: ./gradlew publishToMavenCentral --no-configuration-cache
        env:
          ORG_GRADLE_PROJECT_mavenCentralUsername: ${{ secrets.MAVEN_CENTRAL_USERNAME }}
          ORG_GRADLE_PROJECT_mavenCentralPassword: ${{ secrets.MAVEN_CENTRAL_PASSWORD }}
          ORG_GRADLE_PROJECT_signingInMemoryKeyId: ${{ secrets.SIGNING_KEY_ID }}
          ORG_GRADLE_PROJECT_signingInMemoryKeyPassword: ${{ secrets.SIGNING_PASSWORD }}
          ORG_GRADLE_PROJECT_signingInMemoryKey: ${{ secrets.GPG_KEY_CONTENTS }}
@@ -14,3 +14,20 @@ xcuserdata
 .gigaide
 /kotlin-js-store/yarn.lock
 /test.lyng
 /sample_texts/1.txt.gz
 /kotlin-js-store/wasm/yarn.lock
 /distributables
 .output*.txt
 debug.log
 /build.log
 /test.md
 /build_output.txt
 /build_output_full.txt
 /check_output.txt
 /compile_jvm_output.txt
 /compile_metadata_output.txt
 test_output*.txt
 /site/src/version-template/lyng-version.js
 /bugcontents.db
 /bugs/
 contents.db
@@ -0,0 +1,11 @@
 # Lyng Project Guidelines
 This project uses the Lyng scripting language for multiplatform scripting.
 ## Coding in Lyng
 When writing, refactoring, or analyzing Lyng code:
 - **Reference**: Always use `LYNG_AI_SPEC.md` in the project root as the primary source of truth for syntax and idioms.
 - **File Extensions**: Use `.lyng` for all script files.
 - **Implicit Coroutines**: Remember that all Lyng functions are implicitly coroutines; do not look for `async/await`.
 - **Everything is an Expression**: Leverage the fact that blocks, if-statements, and loops return values.
 - **Maps vs Blocks**: Be careful: `{}` is a block/lambda, use `Map()` for an empty map.
@@ -0,0 +1,28 @@
 <component name="ProjectRunConfigurationManager">
  <configuration default="false" name="Tests in 'lyng.lynglib.jvmTest'" type="GradleRunConfiguration" factoryName="Gradle">
    <ExternalSystemSettings>
      <option name="executionName" />
      <option name="externalProjectPath" value="$PROJECT_DIR$" />
      <option name="externalSystemIdString" value="GRADLE" />
      <option name="scriptParameters" value="" />
      <option name="taskDescriptions">
        <list />
      </option>
      <option name="taskNames">
        <list>
          <option value=":lynglib:cleanJvmTest" />
          <option value=":lynglib:jvmTest" />
        </list>
      </option>
      <option name="vmOptions" />
    </ExternalSystemSettings>
    <ExternalSystemDebugServerProcess>false</ExternalSystemDebugServerProcess>
    <ExternalSystemReattachDebugProcess>true</ExternalSystemReattachDebugProcess>
    <ExternalSystemDebugDisabled>false</ExternalSystemDebugDisabled>
    <DebugAllEnabled>false</DebugAllEnabled>
    <RunAsTest>true</RunAsTest>
    <GradleProfilingDisabled>false</GradleProfilingDisabled>
    <GradleCoverageDisabled>false</GradleCoverageDisabled>
    <method v="2" />
  </configuration>
 </component>
@@ -0,0 +1,44 @@
 <!--
  ~ Copyright 2025 Sergey S. Chernov real.sergeych@gmail.com
  ~
  ~ Licensed under the Apache License, Version 2.0 (the "License");
  ~ you may not use this file except in compliance with the License.
  ~ You may obtain a copy of the License at
  ~
  ~     http://www.apache.org/licenses/LICENSE-2.0
  ~
  ~ Unless required by applicable law or agreed to in writing, software
  ~ distributed under the License is distributed on an "AS IS" BASIS,
  ~ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  ~ See the License for the specific language governing permissions and
  ~ limitations under the License.
  ~
  -->
 <component name="ProjectRunConfigurationManager">
  <configuration default="false" name="lyng:site [jsBrowserDevelopmentRun]" type="GradleRunConfiguration" factoryName="Gradle">
    <ExternalSystemSettings>
      <option name="executionName" />
      <option name="externalProjectPath" value="$PROJECT_DIR$/site" />
      <option name="externalSystemIdString" value="GRADLE" />
      <option name="scriptParameters" value="--continuous" />
      <option name="taskDescriptions">
        <list />
      </option>
      <option name="taskNames">
        <list>
          <option value="jsBrowserDevelopmentRun" />
        </list>
      </option>
      <option name="vmOptions" />
    </ExternalSystemSettings>
    <ExternalSystemDebugServerProcess>true</ExternalSystemDebugServerProcess>
    <ExternalSystemReattachDebugProcess>true</ExternalSystemReattachDebugProcess>
    <ExternalSystemDebugDisabled>false</ExternalSystemDebugDisabled>
    <DebugAllEnabled>false</DebugAllEnabled>
    <RunAsTest>false</RunAsTest>
    <GradleProfilingDisabled>true</GradleProfilingDisabled>
    <GradleCoverageDisabled>true</GradleCoverageDisabled>
    <method v="2" />
  </configuration>
 </component>
@@ -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`.
@@ -0,0 +1,152 @@
 # Changelog
 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
 - No unreleased entries yet.
 ## 1.5.5 (2026-04-23)
 ### 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.
@@ -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.
@@ -0,0 +1,3 @@
 The Lyng programming language.
 Copyright (c) 2024-2025 Sergey Chernov real.sergeych@gmail.com
@@ -1,57 +1,120 @@
-# Lyng: modern scripting for kotlin multiplatform
+# Lyng: ideal scripting for kotlin multiplatform
-A KMP library and a standalone interpreter
+__Please visit the project homepage: [https://lynglang.com](https://lynglang.com) and a [telegram channel](https://t.me/lynglang).__
- simple, compact, intuitive and elegant modern code style:
+__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.
 - simple, compact, intuitive and elegant modern code:
 ```lyng
 class Point(x, y) {
   fun dist() { sqrt(x*x + y*y) } 
 }
-Point(3,4).dist() //< 5
+
 // Auto-named arguments shorthand (x: is x: x):
 val x = 3
 val y = 4
 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
+- 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:
+- 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.
-    delay(1.5) // coroutine is delayed for 1.5s, thread is not blocked!
+- Any Unicode letters can be used as identifiers: `assert( sin(π/2) == 1 )`.
 ```
 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)/
 - functional style and OOP together, multiple inheritance, implementing interfaces for existing classes, writing extensions.
 - 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 library
+### Add dependency to your project
-TBD
+```kotlin
 val lyngVersion = "1.5.5"
 repositories {
    // ...
    maven("https://gitea.sergeych.net/api/packages/SergeychWorks/maven")
 }
 ```
 And add dependency to the proper place in your project, it could look like:
 ```kotlin
 comminMain by getting {
    dependencies {
        // ...
        implementation("net.sergeych:lynglib:$lyngVersion")
    }
 }
 ```
 Now you can import lyng and use it:
 ### Execute script:
 ```kotlin
-assertEquals("hello, world", eval(""" 
+import net.sergeych.lyng.*
-    "hello, " + "world" 
+
-    """).toString())
+// we need a coroutine to start, as Lyng
 // is a coroutine based language, async topdown
 runBlocking {
    val session = EvalSession()
    assert(5 == session.eval(""" 3*3 - 4 """).toInt())
    session.eval(""" println("Hello, Lyng!") """)
 }
 ```
 ### Exchanging information
-Script is executed over some `Context`. Create instance of the context,
+The preferred host runtime is `EvalSession`. It owns the script scope and any coroutines
-add your specific vars and functions to it, an call over it:
+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.*
 runBlocking {
    val session = EvalSession()
    val scope = session.getScope().apply {
        // simple function
-val context = Context().apply {
+        addFn("sumOf") {
    addFn("addArgs") {
            var sum = 0.0
            for (a in args) sum += a.toDouble()
            ObjReal(sum)
@@ -62,14 +125,47 @@ val context = Context().apply {
        // suspend fun doSomeWork(text: String): Int
        addFn("doSomeWork") {
            // this _is_ a suspend lambda, we can call suspend function,
-        // and it won't consume the thread:
+            // and it won't consume the thread.
-        doSomeWork(args[0].toString()).toObj()
+            // note that in kotlin handler, `args` is a list of `Obj` arguments
            // and return value from this lambda should be Obj too:
            doSomeWork(args[0]).toObj()
        }
    }
-// adding constant:
+
-context.eval("addArgs(1,2,3)") // <- 6
+    // execute through the session:
    session.eval("sumOf(1,2,3)") // <- 6
 }
 ```
-Note that the context stores all changes in it so you can make calls on a single context 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)
 The IDEA plugin provides a fast, lightweight BASIC completion for Lyng code (IntelliJ IDEA 2024.3+).
 What it does:
 - Global suggestions: in-scope parameters, same-file declarations (functions/classes/vals), imported modules, and stdlib symbols.
 - Member completion after dot: offers only members of the inferred receiver type. It works for chained calls like `Path(".." ).lines().` (suggests `Iterator` methods), and for literals like `"abc".` (String methods) or `[1,2,3].` (List/Iterable methods).
 - Inheritance-aware: shows direct class members first, then inherited. For example, `List` also exposes common `Collection`/`Iterable` methods.
 - Static/namespace members: `Name.` lists only static members when `Name` is a known class or container (e.g., `Math`).
 - Performance: suggestions are capped; prefix filtering is early; parsing is cached; computation is cancellation-friendly.
 What it does NOT do (yet):
 - No heavy resolve or project-wide indexing. It’s best-effort, driven by a tiny MiniAst + built-in docs registry.
 - No control/data-flow type inference.
 Enable/disable:
 - Settings | Lyng Formatter → "Enable Lyng autocompletion (experimental)" (default: ON).
 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? 
@@ -83,22 +179,78 @@ Designed to add scripting to kotlin multiplatform application in easy and effici
 # Language 
- dynamic 
+- Javascript, WasmJS, native, JVM, android - batteries included.
- async
+- dynamic types in most elegant and concise way
- multithreaded (coroutines could be dispatched using threads on appropriate platforms, automatically)
+- async, 100% coroutines, supports multiple cores where platform supports thread
 - good for functional an object-oriented style
-## By-stage
+# Language Roadmap
-Here are plans to develop it:
+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.
-### First stage
+Ready features:
-Interpreted, precompiled into threaded code, actually. Dynamic types.
+- [x] Language platform and independent command-line launcher
 - [x] Integral types and user classes, variables and constants, functions
 - [x] lambdas and closures, coroutines for all callables
 - [x] while-else, do-while-else, for-else loops with break-continue returning values and labels support
 - [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?)
 - [x] modules
 - [x] string formatting and tools
 - [x] launch, deferred, CompletableDeferred, Mutex, etc.
 - [x] multiline strings
 - [x] typesafe bit-effective serialization
 - [x] compression/decompression (integrated in serialization)
 - [x] dynamic fields
 - [x] function annotations
 - [x] better stack reporting
 - [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
-### Second stage
+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.  
-Will add:
+## plan: towards v2.0 Next Generation
- optimizations
+- [x] site with integrated interpreter to give a try
- p-code serialization
+- [x] kotlin part public API good docs, integration focused
- static typing
+- [x] type specifications
 - [x] Textmate Bundle
 - [x] IDEA plugin
 - [x] source docs and maybe lyng.md to a standard
 - [x] aggressive optimizations
 ## After 1.5 "Ideal scripting"
 * __we are here now ;)__
 - propose your feature! 
 ## Authors
@-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.
 <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.
 [parallelism]: docs/parallelism.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.__
@@ -0,0 +1,28 @@
 [//]: # (excludeFromIndex)
 # String
 # This document is for developer notes only
 --------------------------
 ## Interpolation proposal
 """no $iterpolation"""
    val inpterpolation1 = "foo"
    "${interpolation1}"
    >>> "foo"
 "$interpolation2"
 "no $$ interpolatino"
 ## Regexp vs div / ?
 ```EBNF
 regex_literal = "/", { regchar }, "/", [ flag ]
 foag = "i" | "n"....
 regchar = x`x`
 ```
@@ -1,5 +1,7 @@
 # Modules inclusion
 [//]: # (excludeFromIndex)
 Module is, at the low level, a statement that modifies a given context by adding
 here local and exported symbols, performing some tasks and even returning some value
 we don't need for now.
@@ -0,0 +1,15 @@
 [//]: # (excludeFromIndex)
 Provide:
    fun outer(a1)
        // a1 is caller.a1:arg
        val a1_local = a1 + 1
        // we return lambda:
        { it ->
            // a1_local
            a1_lcoal + it
        }
    }
@@ -0,0 +1,34 @@
 ## Fix: prevent cycles in scope parent chain during pooled frame reuse
 ### What changed
 - Scope.resetForReuse now fully detaches a reused scope from its previous chain/state before re-parenting:
  - sets `parent = null` and regenerates `frameId`
  - clears locals/slots/bindings caches
  - only after that, validates the new parent with `ensureNoCycle` and assigns it
 - ScopePool.borrow on all targets (JVM, Android, JS, Native, Wasm) now has a defensive fallback:
  - if `resetForReuse` throws `IllegalStateException` indicating a parent-chain cycle, the pool allocates a fresh `Scope` instead of failing.
 ### Why
 In some nested call patterns (notably instance method calls where an instance is produced by another function and immediately used), the same pooled `Scope` object can be rebound into a chain that already (transitively) contains it. Reassigning `parent` in that case forms a structural cycle, which `ensureNoCycle` correctly detects and throws. This could surface as:
 ```
 IllegalStateException: cycle detected in scope parent chain assignment
  at net.sergeych.lyng.Scope.ensureNoCycle(...)
  at net.sergeych.lyng.Scope.resetForReuse(...)
  at net.sergeych.lyng.ScopePool.borrow(...)
  ... during instance method invocation
 ```
 The fix removes the failure mode by:
 1) Detaching the reused frame from its prior chain/state before validating and assigning the new parent.
 2) Falling back to a new frame allocation if a cycle is still detected (extremely rare and cheap vs. a crash).
 ### Expected effects
 - Eliminates sporadic `cycle detected in scope parent chain` crashes during instance method invocation.
 - No change to public API or normal semantics.
 - Pooling remains enabled by default; the fallback only triggers on the detected cycle edge case.
 - Negligible performance impact: fresh allocation is used only when a cycle would have crashed the VM previously.
 ### Notes
 - The fix is platform-wide (all ScopePool actuals are covered).
 - We recommend adding/keeping a regression test that exercises: a class with a method, a function returning an instance, and an exported function calling the instance method. The test should pass without exceptions.
@@ -0,0 +1,25 @@
 # Migration of Instant and Clock
 ## History
 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 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). 
 Later JetBrains added serializers for their new `Instant` and `Clock` types, but strangely not in the stdlib, but in newer versions of `kotlinx.serialization`. This means that plain upgrade of dependencies to 2.2 is not enough to make them work.
 ## Solution
 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, 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 such "brilliant" ideas from IDEA ideologspersons.
 Sorry for inconvenience and send a ray of hate to JetBrains ;)
@@ -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 } }
@@ -0,0 +1,95 @@
 # Map literals — refined proposal
 [//]: # (excludeFromIndex)
 Implement JavaScript-like literals for maps. The syntax and semantics align with named arguments in function calls, but map literals are expressions that construct `Map` values.
 Keys can be either:
 - string literals: `{ "some key": value }`, or
 - identifiers: `{ name: expr }`, where the key becomes the string `"name"`.
 Identifier shorthand inside map literals is supported:
 - `{ name: }` desugars to `{ "name": name }`.
 Property access sugar is not provided for maps: use bracket access only, e.g. `m["a"]`, not `m.a`.
 Examples:
 ```lyng
 val x = 2
 val m = { "a": 1, x: x*10, y: }
 assertEquals(1, m["a"])      // string-literal key
 assertEquals(20, m["x"])     // identifier key
 assertEquals(2, m["y"])      // identifier shorthand
 ```
 Spreads (splats) in map literals are allowed and merged left-to-right with “rightmost wins” semantics:
 ```lyng
 val base = { a: 1, b: 2 }
 val m = { a: 0, ...base, b: 3, c: 4 }
 assertEquals(1, m["a"])  // base overwrites a:0
 assertEquals(3, m["b"])  // literal overwrites spread
 assertEquals(4, m["c"])  // new key
 ```
 Trailing commas are allowed (optional):
 ```lyng
 val m = {
  "a": 1,
  b: 2,
  ...other,
 }
 ```
 Duplicate keys among literal entries (including identifier shorthand) are a compile-time error:
 ```lyng
 { foo: 1, "foo": 2 }   // error: duplicate key "foo"
 { foo:, foo: 2 }        // error: duplicate key "foo"
 ```
 Spreads are evaluated at runtime. Overlaps from spreads are resolved by last write wins. If a spread is not a map, or yields a map with non-string keys, it’s a runtime error.
 Merging with `+`/`+=` and entries:
 ```lyng
 ("1" => 10) + ("2" => 20)     // Map("1"=>10, "2"=>20)
 { "1": 10 } + ("2" => 20)     // same
 { "1": 10 } + { "2": 20 }    // same
 var m = { "a": 1 }
 m += ("b" => 2)                 // m = { "a":1, "b":2 }
 ```
 Rightmost wins on duplicates consistently across spreads and merges. All map merging operations require string keys; encountering a non-string key during merge is a runtime error.
 Empty map literal `{}` is not supported to avoid ambiguity with blocks/lambdas. Use `Map()` for an empty map.
 Lambda disambiguation
 - A `{ ... }` with typed lambda parameters must have a top-level `->` in its header. The compiler disambiguates by looking for a top-level `->`. If none is found, it attempts to parse a map literal; if that fails, it is parsed as a lambda or block.
 Grammar (EBNF)
 ```
 ws                    = zero or more whitespace (incl. newline/comments)
 map_literal           = '{' ws map_entries ws '}'
 map_entries           = map_entry ( ws ',' ws map_entry )* ( ws ',' )?
 map_entry             = map_key ws ':' ws map_value_opt
                      | '...' ws expression
 map_key               = string_literal | ID
 map_value_opt         = expression | ε   // ε allowed only if map_key is ID
 ```
 Notes:
 - Identifier shorthand (`id:`) is allowed only for identifiers, not string-literal keys.
 - Spreads accept any expression; at runtime it must yield a `Map` with string keys.
 - Duplicate keys are detected at compile time among literal keys; spreads are merged at runtime with last-wins.
 Rationale
 - The `{ name: value }` style is familiar and ergonomic.
 - Disambiguation with lambdas leverages the required `->` in typed lambda headers.
 - Avoiding `m.a` sidesteps method/field shadowing and keeps semantics clear.
@@ -0,0 +1,70 @@
 # Named arguments in calls
 [//]: # (excludeFromIndex)
 Extend function/method calls to allow setting arguments by name using colon syntax at call sites. This is especially convenient with many parameters and default values.
 Examples:
 ```lyng
    fun test(a="foo", b="bar", c="bazz") { [a, b, c] }
    assertEquals(test(b: "b"), ["foo", "b", "bazz"])
    assertEquals(test("a", c: "c"), ["a", "bar", "c"])
 ```
 Rules:
 - Named arguments are optional. If named arguments are present, their order is not important.
 - Named arguments must follow positional arguments; positional arguments cannot follow named ones (the only exception is the syntactic trailing block outside parentheses, see below).
 - A named argument cannot reassign a parameter already set positionally.
 - If the last parameter is already assigned by a named argument (or named splat), the trailing-lambda rule must NOT apply: a following `{ ... }` after the call is an error.
 Rationale for using `:` instead of `=` in calls: in Lyng, assignment `=` is an expression; using `:` avoids ambiguity and keeps declarations (`name: Type`) distinct from call sites, where casting uses `as` / `as?`.
 Migration note: earlier drafts/examples used `name = value`. The final syntax is `name: value` at call sites.
 ## Extended call argument splats: named splats
 With named arguments, splats (`...`) are extended to support maps as named splats. When a splat evaluates to a Map, its entries provide name→value assignments:
 ```lyng
    fun test(a="a", b="b", c="c", d="d") { [a, b, c, d] }
    assertEquals(test("A?", ...Map("d" => "D!", "b" => "B!")), ["A?", "B!", "c", "D!"])
 ```
 Constraints for named splats:
 - Only string keys are allowed in map splats; otherwise, a clean error is thrown.
 - Named splats cannot reassign parameters already set (positionally or by earlier named arguments/splats).
 - Named splats follow the same ordering as named arguments: they must appear after all positional arguments and positional splats.
 ## Trailing-lambda interaction
 Lyng supports a syntactic trailing block after a call: `f(args) { ... }`. With named args/splats, if the last parameter is already assigned by name, the trailing block must not apply and the call is an error:
 ```lyng
    fun f(x, onDone) { onDone(x) }
    f(x: 1) { 42 }    // ERROR: last parameter already assigned by name
    f(1) { 42 }       // OK
 ```
 ## Errors (non-exhaustive)
 - Positional argument after any named argument inside parentheses: error.
 - Positional splat after any named argument: error.
 - Duplicate named assignment (directly or via map splats): error.
 - Unknown parameter name in a named argument/splat: error.
 - Map splat with non-string keys: error.
 - Attempt to target the ellipsis parameter by name: error.
 ## Notes
 - Declarations continue to use `:` for types, while call sites use `:` for named arguments and `as` / `as?` for type casts/checks.
@@ -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.
@@ -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
@@ -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
@@ -0,0 +1,192 @@
 #!/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.
 #
 #
 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 [[ $? != 0 ]]; then
    echo
    echo -- rsync failed. deploy was not finished. deployed version has not been affected
    echo
    exit 100
  fi
 }
 # Update docs/idea_plugin.md to point to the latest built IDEA plugin zip
 # from ./distributables before building the site. The change is temporary and
 # the original file is restored right after the build.
 DOC_IDEA_PLUGIN="docs/idea_plugin.md"
 DOC_IDEA_PLUGIN_BACKUP="${DOC_IDEA_PLUGIN}.deploy_backup"
 function updateIdeaPluginDownloadLink() {
  if [[ ! -f "$DOC_IDEA_PLUGIN" ]]; then
    echo "WARN: $DOC_IDEA_PLUGIN not found; skipping plugin link update"
    return 0
  fi
  # Find the most recently modified plugin zip
  local latest
  latest=$(ls -t distributables/lyng-idea-*.zip 2>/dev/null | head -n 1)
  if [[ -z "$latest" ]]; then
    echo "WARN: no distributables/lyng-idea-*.zip found; leaving $DOC_IDEA_PLUGIN unchanged"
    return 0
  fi
  local base
  base=$(basename "$latest")
  local version
  version="${base#lyng-idea-}"
  version="${version%.zip}"
  local url
  url="https://lynglang.com/distributables/${base}"
  local newline
  newline="### [Download plugin v${version}](${url})"
  # Backup and rewrite the specific markdown line if present
  cp "$DOC_IDEA_PLUGIN" "$DOC_IDEA_PLUGIN_BACKUP" || {
    echo "ERROR: can't backup $DOC_IDEA_PLUGIN"; return 1; }
  # Replace the line that starts with the download header; if not found, append it
  awk -v repl="$newline" 'BEGIN{done=0} \
    /^### \[Download plugin v/ { print repl; done=1; next } \
    { print } \
    END { if (done==0) exit 42 }' "$DOC_IDEA_PLUGIN_BACKUP" > "$DOC_IDEA_PLUGIN"
  local rc=$?
  if [[ $rc -eq 42 ]]; then
    echo "WARN: download link not found in $DOC_IDEA_PLUGIN; appending generated link"
    echo >> "$DOC_IDEA_PLUGIN"
    echo "$newline" >> "$DOC_IDEA_PLUGIN"
  elif [[ $rc -ne 0 ]]; then
    echo "ERROR: failed to update $DOC_IDEA_PLUGIN; restoring original"
    mv -f "$DOC_IDEA_PLUGIN_BACKUP" "$DOC_IDEA_PLUGIN" 2>/dev/null
    return 1
  fi
 }
 # target settings  (-t com | -t vps)
 case "$target" in
  com)
    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
    ;;
  *)
    echo "*** ERROR: unknown target '$target' (use -t com | -t vps)"
    echo "*** stop"
    exit 101
 esac
 die() { echo "ERROR: $*" 1>&2 ; exit 1; }
 function refreshTextmateZip() {
  echo "Refreshing distributables/lyng-textmate.zip from editors/..."
  mkdir -p distributables
  # We use -r for recursive and -q for quiet (optional)
  # -j can be used if we want to junk paths, but the request says "contents of editors/"
  # usually we want to preserve the structure inside editors/
  (cd editors && zip -rq ../distributables/lyng-textmate.zip .)
 }
 if [[ "$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"
  ./gradlew site:jsBrowserDistribution
  BUILD_RC=$?
  # Always restore original doc if backup exists
  if [[ -f "$DOC_IDEA_PLUGIN_BACKUP" ]]; then
    mv -f "$DOC_IDEA_PLUGIN_BACKUP" "$DOC_IDEA_PLUGIN"
  fi
  if [[ $BUILD_RC -ne 0 ]]; then
    echo
    echo -- build failed. deploy aborted.
    echo
    exit 100
  fi
 fi
 # Prepare working dir
 ssh -p ${SSH_PORT} ${SSH_HOST} "
  cd ${ROOT}
  rm -rd build 2>/dev/null
  if [ -d release ]; then
    echo copying current release
    cp -r release build
  else
    echo creating first release
    mkdir release
    mkdir build
  fi
 ";
 if [[ "$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
 echo
 echo finalizing the deploy...
 ssh -p ${SSH_PORT} ${SSH_HOST} "
  cd ${ROOT}
  rm -rd release~
  mv release release~
  mv build release
  cd release
  # in this project we needn't restart back when we deploy the front
  #  ~/bin/restart_service
 ";
 if [[ $? != 0 ]]; then
  echo
  echo -- finalization failed. the rease might be corrupt. rolling back is not yet implemented.
  echo
  exit 100
 fi
 echo
 echo Deploy successful
 echo
@@ -0,0 +1,256 @@
 #!/bin/bash
 #
 # Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
 #     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
 #
 set -e
 # Configuration
 DOCS_DIR="docs"
 OUTPUT_DIR="distributables"
 TEMP_DIR="build/temp_docs"
 MERGED_MD="$TEMP_DIR/merged.md"
 OUTPUT_HTML="$OUTPUT_DIR/lyng_documentation.html"
 mkdir -p "$OUTPUT_DIR"
 mkdir -p "$TEMP_DIR"
 # Files that should come first in specific order
 PRIORITY_FILES=(
    "tutorial.md"
    "OOP.md"
    "advanced_topics.md"
    "declaring_arguments.md"
    "scopes_and_closures.md"
    "exceptions_handling.md"
    "when.md"
    "parallelism.md"
    "Testing.md"
 )
 # Files that should come next (reference)
 REFERENCE_FILES=(
    "Collection.md"
    "Iterable.md"
    "Iterator.md"
    "List.md"
    "Set.md"
    "Map.md"
    "Array.md"
    "Buffer.md"
    "RingBuffer.md"
    "Range.md"
    "Real.md"
    "Regex.md"
    "math.md"
    "time.md"
 )
 # Files that are about integration/tools
 INTEGRATION_FILES=(
    "serialization.md"
    "json_and_kotlin_serialization.md"
    "embedding.md"
    "lyng_cli.md"
    "lyng.io.fs.md"
    "formatter.md"
    "EfficientIterables.md"
 )
 # Tracking processed files to avoid duplicates
 PROCESSED_PATHS=()
 is_excluded() {
    local full_path="$1"
    if grep -q "excludeFromIndex" "$full_path"; then
        return 0 # true in bash
    fi
    return 1 # false
 }
 process_file() {
    local rel_path="$1"
    local full_path="$DOCS_DIR/$rel_path"
    if [[ ! -f "$full_path" ]]; then
        return
    fi
    if is_excluded "$full_path"; then
        echo "Skipping excluded: $rel_path"
        return
    fi
    # Check for duplicates
    for p in "${PROCESSED_PATHS[@]}"; do
        if [[ "$p" == "$rel_path" ]]; then
            return
        fi
    done
    PROCESSED_PATHS+=("$rel_path")
    echo "Processing: $rel_path"
    # 1. Add an anchor for the file based on its path
    local anchor_name=$(echo "$rel_path" | sed 's/\//_/g')
    echo "<div id=\"$anchor_name\"></div>" >> "$MERGED_MD"
    echo "" >> "$MERGED_MD"
    # 2. Append content with fixed links
    # - [text](file.md) -> [text](#file.md)
    # - [text](dir/file.md) -> [text](#dir_file.md)
    # - [text](file.md#anchor) -> [text](#anchor)
    # - Fix image links: [alt](../images/...) -> [alt](images/...) if needed, but none found yet.
    cat "$full_path" | \
    perl -pe 's/\[([^\]]+)\]\(([^)]+)\.md\)/"[$1](#" . ($2 =~ s|\/|_|gr) . ".md)"/ge' | \
    perl -pe 's/\[([^\]]+)\]\(([^)]+)\.md#([^)]+)\)/[$1](#$3)/g' >> "$MERGED_MD"
    echo -e "\n\n---\n\n" >> "$MERGED_MD"
 }
 # Start with an empty merged file
 echo "% Lyng Language Documentation" > "$MERGED_MD"
 echo "" >> "$MERGED_MD"
 # 1. Process priority files
 for f in "${PRIORITY_FILES[@]}"; do
    process_file "$f"
 done
 # 2. Process reference files
 for f in "${REFERENCE_FILES[@]}"; do
    process_file "$f"
 done
 # 3. Process integration files
 for f in "${INTEGRATION_FILES[@]}"; do
    process_file "$f"
 done
 # 4. Process remaining files in docs root
 for f in "$DOCS_DIR"/*.md; do
    rel_f=${f#"$DOCS_DIR/"}
    process_file "$rel_f"
 done
 # 5. Process remaining files in subdirs (like samples)
 find "$DOCS_DIR" -name "*.md" | sort | while read -r f; do
    rel_f=${f#"$DOCS_DIR/"}
    process_file "$rel_f"
 done
 echo "Running pandoc to generate $OUTPUT_HTML..."
 # Use a basic but clean CSS
 pandoc "$MERGED_MD" -o "$OUTPUT_HTML" \
    --toc --toc-depth=2 \
    --standalone \
    --embed-resources \
    --metadata title="Lyng Language Documentation" \
    --css <(echo "
        body { 
            font-family: -apple-system, BlinkMacSystemFont, \"Segoe UI\", Helvetica, Arial, sans-serif;
            line-height: 1.6; 
            max-width: 1000px; 
            margin: 0 auto; 
            padding: 2em; 
            color: #24292e;
            background-color: #fff;
        }
        code { 
            background-color: rgba(27,31,35,0.05); 
            padding: 0.2em 0.4em; 
            border-radius: 3px; 
            font-family: SFMono-Regular, Consolas, \"Liberation Mono\", Menlo, monospace;
            font-size: 85%;
        }
        pre { 
            background-color: #f6f8fa; 
            padding: 16px; 
            overflow: auto; 
            border-radius: 3px;
            line-height: 1.45;
        }
        pre code {
            background-color: transparent;
            padding: 0;
            font-size: 100%;
        }
        h1 { 
            border-bottom: 1px solid #eaecef; 
            padding-bottom: 0.3em; 
            margin-top: 24px;
            margin-bottom: 16px;
            font-weight: 600;
        }
        h2 { 
            border-bottom: 1px solid #eaecef; 
            padding-bottom: 0.3em; 
            margin-top: 24px;
            margin-bottom: 16px;
            font-weight: 600;
        }
        hr { 
            height: 0.25em;
            padding: 0;
            margin: 24px 0;
            background-color: #e1e4e8;
            border: 0;
        }
        blockquote {
            padding: 0 1em;
            color: #6a737d;
            border-left: 0.25em solid #dfe2e5;
            margin: 0 0 16px 0;
        }
        nav#TOC { 
            background: #f9f9f9; 
            padding: 1em; 
            border: 1px solid #eee; 
            margin-bottom: 2.5em; 
            border-radius: 6px;
        }
        nav#TOC ul { 
            list-style: none; 
            padding-left: 1.5em; 
        }
        nav#TOC > ul {
            padding-left: 0;
        }
        table {
            border-spacing: 0;
            border-collapse: collapse;
            margin-top: 0;
            margin-bottom: 16px;
        }
        table th, table td {
            padding: 6px 13px;
            border: 1px solid #dfe2e5;
        }
        table tr {
            background-color: #fff;
            border-top: 1px solid #c6cbd1;
        }
        table tr:nth-child(2n) {
            background-color: #f6f8fa;
        }
    ")
 echo "-------------------------------------------------------"
 echo "Done! Documentation generated successfully."
 echo "Location: $OUTPUT_HTML"
 echo "-------------------------------------------------------"
@@ -0,0 +1,33 @@
 #!/bin/bash
 #
 # Copyright 2026 Sergey S. Chernov real.sergeych@gmail.com
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
 #
 #     http://www.apache.org/licenses/LICENSE-2.0
 #
 # Unless required by applicable law or agreed to in writing, software
 # distributed under the License is distributed on an "AS IS" BASIS,
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
 #
 set -e
 archive=./lyng/build/distributions/lyng-jvm.zip
 install_root="$HOME/bin/jlyng-jvm"
 launcher="$install_root/lyng-jvm/bin/lyng"
 ./gradlew :lyng:jvmDistZip
 mkdir -p ./distributables
 cp "$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"
@@ -1,5 +1,22 @@
 #!/bin/bash
 #
 # Copyright 2025 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
 file=./lyng/build/bin/linuxX64/releaseExecutable/lyng.kexe
@@ -8,3 +25,6 @@ file=./lyng/build/bin/linuxX64/releaseExecutable/lyng.kexe
 strip $file
 upx   $file
 cp $file ~/bin/lyng
 cp $file ./distributables/lyng
 zip ./distributables/lyng-linuxX64 ./distributables/lyng
 rm ./distributables/lyng
@@ -0,0 +1,3 @@
 #!/usr/bin/env lyng
 println("Hello from lyng!")
@@ -1,5 +1,61 @@
 /*
 * 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.
 *
 */
 plugins {
    alias(libs.plugins.androidLibrary) apply false
    alias(libs.plugins.kotlinMultiplatform) apply  false
    alias(libs.plugins.vanniktech.mavenPublish) apply false
 }
 // Convenience alias to run the IntelliJ IDE with the Lyng plugin from the project root.
 // Usage: ./gradlew runIde
 // It simply delegates to :lyng-idea:runIde provided by the Gradle IntelliJ Plugin.
 tasks.register<org.gradle.api.DefaultTask>("runIde") {
    group = "intellij"
    description = "Run IntelliJ IDEA with the Lyng plugin (:lyng-idea)"
    dependsOn(":lyng-idea:runIde")
 }
 tasks.register<Exec>("generateDocs") {
    group = "documentation"
    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()
        )
    }
 }
@@ -0,0 +1,45 @@
 # Array
 It's an interface if the [Collection] that provides indexing access, like `array[3] = 0`.
 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:
 ## Binary search
 When applied to sorted arrays, binary search allow to quicly find an index of the element in the array, or where to insert it to keep order:
    val coll = [1,2,3,4,5]
    assertEquals( 2, coll.binarySearch(3) )
    assertEquals( 0, coll.binarySearch(1) )
    assertEquals( 4, coll.binarySearch(5) )
    val src = (1..50).toList().shuffled()
    val result = []
    for( x in src ) {
        val i = result.binarySearch(x)
        assert( i < 0 )
        result.insertAt(-i-1, x)
    }
    assertEquals( src.sorted(), result )
    >>> void
 So `binarySearch(x)` returns:
 - index of `x`, a non-negative number
 - negative: `x` not found, but if inserted at position `-returnedValue-1` will leave array sorted.
 To pre-sort and array use `Iterable.sorted*` or in-place `List.sort*` families, see [List] and [Iterable] docs.
 [Collection]: Collection.md
 [Iterable]: Iterable.md
 [List]: List.md
 [ImmutableList]: ImmutableList.md
@@ -0,0 +1,152 @@
 # Binary `Buffer`
 Buffers are effective unsigned byte arrays of fixed size. Buffers content is mutable,
 unlike its size. Buffers are comparable and implement [Array], thus [Collection] and [Iterable]. Buffer iterators return
 its contents as unsigned bytes converted to `Int`
 Buffers needs to be imported with `import lyng.buffer`:
    import lyng.buffer
    assertEquals(5, Buffer("Hello").size)
    >>> void
 Buffer is _immutable_, there is a `MutableBuffer` with same interface but mutable.
 ## Constructing
 There are a lo of ways to construct a buffer:
    import lyng.buffer
    // from string using utf8 encoding:
    assertEquals( 5, Buffer("hello").size )
    // from bytes, e.g. integers in range 0..255
    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 )
    // Empty buffer of fixed size:
    assertEquals(100, Buffer(100).size)
    assertEquals(0, Buffer(100)[0])
    // Note that you can use list iteral to create buffer with 1 byte:
    assertEquals(1, Buffer([100]).size)
    assertEquals(100, Buffer([100])[0])
    >>> void
 ## Accessing and modifying
 Buffer implement [Array] and therefore can be accessed, and `MutableBuffers` also modified:
    import lyng.buffer
    val b1 = Buffer( 1, 2, 3)
    assertEquals( 2, b1[1] )
    val b2 = b1.toMutable()
    assertEquals( 2, b1[1] )
    b2[1]++
    b2[0] = 100
    assertEquals( Buffer(100, 3, 3), b2)
    // b2 is a mutable copy so b1 has not been changed:
    assertEquals( Buffer(1, 2, 3), b1)
    >>> void
 Buffer provides concatenation with another Buffer:
    import lyng.buffer
    val b = Buffer(101, 102)
    assertEquals( Buffer(101, 102, 1, 2), b + [1,2])
    >>> void
 Please note that indexed bytes are _readonly projection_, e.g. you can't modify these with
 ## Comparing
 Buffers are comparable with other buffers (and notice there are _mutable_ buffers, bu default buffers ar _immutable_):
    import lyng.buffer
    val b1 = Buffer(1, 2, 3)
    val b2 = Buffer(1, 2, 3)
    val b3 = MutableBuffer(b2)
    b3[0] = 101
    assert( b3 > b1 )
    assert( b2 == b1 )
    // longer string with same prefix is considered bigger:
    assert( b2 + "!".characters() > b1 )
    // note that characters() provide Iterable of characters that 
    // can be concatenated to Buffer
    >>> void
 ## Slicing
 As with [List], it is possible to use ranges as indexes to slice a Buffer:
    import lyng.buffer
    val a = Buffer( 100, 101, 102, 103, 104, 105 )
    assertEquals( a[ 0..1 ], Buffer(100, 101) )
    assertEquals( a[ 0 ..< 2 ], Buffer(100, 101) )
    assertEquals( a[  ..< 2 ], Buffer(100, 101) )
    assertEquals( a[  4.. ], Buffer(104, 105) )
    assertEquals( a[  2..3 ], Buffer(102, 103) )
    >>> void
 ## Encoding
 You can encode `String` to buffer using buffer constructor, as was shown. Also, buffer supports out of the box base64 (
 which is used in `toString`) and hex encoding:
    import lyng.buffer
    // to UTF8 and back:
    val b = Buffer("hello")
    assertEquals( "hello", b.decodeUtf8() )
    // to base64 and back:
    assertEquals( b, Buffer.decodeBase64(b.base64) )
    assertEquals( b, Buffer.decodeHex(b.hex) )
    >>> void
 ## Members
 | name                       | meaning                                        | type          |
 |----------------------------|------------------------------------------------|---------------|
 | `size`                     | size                                           | Int           |
 | `decodeUtf8`               | decode to String using UTF8 rules              | Any           |
 | `+`                        | buffer concatenation                           | Any           |
 | `toMutable()`              | create a mutable copy                          | MutableBuffer |
 | `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)        |               |
 (1)
 : optimized implementation that override `Iterable` one
 (2)
 : base64url alphabet is used without trailing '=', which allows string to be used in URI without escaping. Note that
 decoding supports both traditional and URL alphabets automatically, and ignores filling `=` characters. Base64URL is
 well known and mentioned in the internet, for example, [here](https://base64.guru/standards/base64url).
 (3)
 : `BitInput` is a bit buffer that is used, for example, in [Lynon.decode](serialization.md)
 Also, it inherits methods from [Iterable] and [Array].
 [Range]: Range.md
 [Iterable]: Iterable.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.
@@ -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
@@ -0,0 +1,31 @@
 # Collection
 Is a [Iterable] with known `size`, a finite [Iterable]:
    class Collection : 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                                          |
 |------------------------|------------------------------------------------------|
 (1)
 : `comparator(a,b)` should return -1 if `a < b`, +1 if `a > b` or zero.
 See [List], [Set], [Iterable] and [Efficient Iterables in Kotlin Interop](EfficientIterables.md)
 [Iterable]: Iterable.md
 [List]: List.md
 [Set]: Set.md
 [Map]: Map.md
 [ImmutableList]: ImmutableList.md
 [ImmutableSet]: ImmutableSet.md
 [ImmutableMap]: ImmutableMap.md
 [ObservableList]: ObservableList.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.
@@ -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).
@@ -0,0 +1,92 @@
 # Efficient Iterables in Kotlin Interop
 Lyng provides high-performance iteration mechanisms that allow Kotlin-side code to interact with Lyng iterables efficiently and vice versa.
 ## 1. Enumerating Lyng Objects from Kotlin
 To iterate over a Lyng object (like a `List`, `Set`, or `Range`) from Kotlin code, use the virtual `enumerate` method:
 ```kotlin
 val lyngList: Obj = ... 
 lyngList.enumerate(scope) { item ->
    println("Processing $item")
    true // return true to continue, false to break
 }
 ```
 ### Why it's efficient:
 - **Zero allocation**: Unlike traditional iterators, it doesn't create a `LyngIterator` object or any intermediate wrappers.
 - **Direct access**: Subclasses like `ObjList` override `enumerate` to iterate directly over their internal Kotlin collections.
 - **Reduced overhead**: It avoids multiple `invokeInstanceMethod` calls for `hasNext()` and `next()` on every step, which would normally involve dynamic dispatch and scope overhead.
 ## 2. Reactive Enumeration with Flow
 If you prefer a reactive approach or need to integrate with Kotlin Coroutines flows, use `toFlow()`:
 ```kotlin
 lyngList.toFlow(scope).collect { item ->
    // ...
 }
 ```
 *Note: `toFlow()` internally uses the Lyng iterator protocol (`iterator()`, `hasNext()`, `next()`), so it's slightly less efficient than `enumerate()` for performance-critical loops, but more idiomatic for flow-based processing.*
 ## 3. Creating Efficient Iterables for Lyng in Kotlin
 When implementing a custom object in Kotlin that should be iterable in Lyng (e.g., usable in `for (x in myObj) { ... }`), follow these steps to ensure maximum performance.
 ### A. Inherit from `Obj` and use `ObjIterable`
 Ensure your object's class has `ObjIterable` as a parent so the Lyng compiler recognizes it as an iterable.
 ```kotlin
 class MyCollection(val items: List<Obj>) : Obj() {
    override val objClass = MyCollection.type
    companion object {
        val type = ObjClass("MyCollection", ObjIterable).apply {
            // Provide a Lyng-side iterator for compatibility with 
            // manual iterator usage in Lyng scripts.
            // Using ObjKotlinObjIterator if items are already Obj instances:
            addFn("iterator") { 
                ObjKotlinObjIterator(thisAs<MyCollection>().items.iterator()) 
            }
        }
    }
 }
 ```
 ### B. Override `enumerate` for Maximum Performance
 The Lyng compiler's `for` loops use the `enumerate` method. By overriding it in your Kotlin class, you provide a "fast path" for iteration.
 ```kotlin
 class MyCollection(val items: List<Obj>) : Obj() {
    // ...
    override suspend fun enumerate(scope: Scope, callback: suspend (Obj) -> Boolean) {
        for (item in items) {
            // If callback returns false, it means 'break' was called in Lyng
            if (!callback(item)) break
        }
    }
 }
 ```
 ### C. Use `ObjInt.of()` for Numeric Data
 If your iterable contains integers, always use `ObjInt.of(Long)` instead of the `ObjInt(Long)` constructor. Lyng maintains a cache for small integers (-128 to 127), which significantly reduces object allocations and GC pressure during tight loops.
 ```kotlin
 // Efficiently creating an integer object
 val obj = ObjInt.of(42L) 
 // Or using extension methods which also use the cache:
 val obj2 = 42.toObj()
 val obj3 = 42L.toObj()
 ```
 #### Note on `toObj()` extensions:
 While `<reified T> T.toObj()` is convenient, using specific extensions like `Int.toObj()` or `Long.toObj()` is slightly more efficient as they use the `ObjInt` cache.
 ## 4. Summary of Best Practices
 - **To Consume**: Use `enumerate(scope) { item -> ... true }`.
 - **To Implement**: Override `enumerate` in your `Obj` subclass.
 - **To Register**: Use `ObjIterable` (or `ObjCollection`) as a parent class in your `ObjClass` definition.
 - **To Optimize**: Use `ObjInt.of()` (or `.toObj()`) for all integer object allocations.
@@ -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
@@ -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
@@ -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
@@ -1,37 +1,246 @@
 # Iterable interface
-The inteface which requires iterator to be implemented:
+The interface for anything that can be iterated, e.g. finite or infinite ordered set of data that can be accessed
 sequentially. Almost any data container in `Lyng` implements it: `List`, `Set`, `Buffer`, `RingBuffer`, `BitBuffer`,
 `Range` and many others are `Iterable`, also `Collection` and `Array` interfaces inherit it.
-    fun iterator(): Iterator
+`Map` and `String` have `Iterable` members to access its contents too.
 Please see also [Collection] interface: many iterables are also collections, and it adds important features.
 ## Definition:
 Iterable is a class that provides function that creates _the iterator_:
    class Iterable {
        abstract fun iterator()
    }
 Note that each call of `iterator()` must provide an independent iterator.
 Iterator itself is a simple interface that should provide only to method:
-    interface Iterable {
+    class Iterator {
-        fun hasNext(): Bool
+        abstract fun hasNext(): Bool
        fun next(): Obj
    }
 Just remember at this stage typed declarations are not yet supported.
-Having `Iterable` in base classes allows to use it in for loop. Also, each `Iterable` has some utility functions available:
+Having `Iterable` in base classes allows to use it in for loop. Also, each `Iterable` has some utility functions
 available, for example
-## Abstract methods
+    val r = 1..10  // Range is Iterable!  
    assertEquals( [9,10], r.takeLast(2).toList() )
    assertEquals( [1,2,3], r.take(3).toList() )
    assertEquals( [9,10], r.drop(8).toList() )
    assertEquals( [1,2], r.dropLast(8).toList() )
    >>> void
 ## joinToString
 This methods convert any iterable to a string joining string representation of each element, optionally transforming it
 and joining using specified separator.
    Iterable.joinToString(separator=' ', transformer=null)
 - if `Iterable` `isEmpty`, the empty string `""` is returned.
 - `separator` is inserted between items when there are more than one.
 - `transformer` of specified is applied to each element, otherwise its `toString()` method is used.
 Here is the sample:
    assertEquals( (1..3).joinToString(), "1 2 3")
    assertEquals( (1..3).joinToString(":"), "1:2:3")
    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:
    assertEquals( 6, [1,2,3].sum() )
    assertEquals( 12, [1,2,3].sumOf { it*2 } )
    // sum of empty collections is null:
    assertEquals( null, [].sum() )
    assertEquals( null, [].sumOf { 2*it } )
    >>> void
 ## map, filter and their variations
 Used to transform or filter the whole iterable stream:
    val source = [1,2,3,4]
    // map: transform every element to something else
    assertEquals(["n1", "n2", "n3", "n4"], source.map { "n"+it } )
    // filter: keep only elements matching the predicate
    assertEquals([2, 4], source.filter { it % 2 == 0 } )
    // count: count elements matching the predicate
    assertEquals(2, source.count { it % 2 == 0 } )
    // mapNotNull: transform every element, skipping null results:
    assertEquals(["n1", "n2", "n4"], source.mapNotNull { if( it == 3 ) null else "n"+it } )
    // filterNotNull: skip all null elements:
    assertEquals([1, 2, 4], [1, 2, null, 4].filterNotNull())
    >>> void
 You can also use flow variations that return a cold `Flow` instead of a `List`, which is useful for large or infinite sequences:
    val source = [1, 2, 3, 4]
    // filterFlow: returns a Flow of filtered elements
    assert( source.filterFlow { it % 2 == 0 } is Flow )
    // filterFlowNotNull: returns a Flow of non-null elements
    assert( [1, null, 2].filterFlowNotNull() is Flow )
    >>> void
 ## minOf and maxOf
 Find the minimum or maximum value of a function applied to each element:
    val source = ["abc", "de", "fghi"]
    assertEquals(2, source.minOf { (it as String).length })
    assertEquals(4, source.maxOf { (it as String).length })
    >>> void
 ## flatten and flatMap
 Work with nested collections:
    val nested = [[1, 2], [3, 4]]
    // flatten: combine nested collections into one list
    assertEquals([1, 2, 3, 4], nested.flatten())
    // flatMap: map each element to a collection and flatten the result
    assertEquals([1, 10, 2, 20], [1, 2].flatMap { [it, it*10] })
    >>> void
 ## findFirst and findFirstOrNull
 Search for the first element that satisfies the given predicate:
    val source = [1, 2, 3, 4]
    assertEquals( 2, source.findFirst { it % 2 == 0 } )
    assertEquals( 2, source.findFirstOrNull { it % 2 == 0 } )
    // findFirst throws if not found:
    assertThrows( NoSuchElementException ) { source.findFirst { it > 10 } }
    // findFirstOrNull returns null if not found:
    assertEquals( null, source.findFirstOrNull { it > 10 } )
    >>> void
 ## Instance methods:
 | 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 |
 | indexOf(i)             | return index if the first encounter of i or a negative value if not found       |
 | associateBy(kf)        | create a map where keys are returned by kf that will be called for each element |
 | filter(p)              | create a list of elements matching predicate `p`                |
 | count(p)               | count elements matching predicate `p`                            |
 | filterFlow(p)          | create a [Flow] of elements matching predicate `p`               |
 | filterNotNull()        | create a list of non-null elements                               |
 | filterFlowNotNull()    | create a [Flow] of non-null elements                             |
 | minOf(f)               | return minimum value of `f` applied to elements                  |
 | maxOf(f)               | return maximum value of `f` applied to elements                  |
 | flatten()              | flatten nested collections into a single [List]                  |
 | flatMap(f)             | map each element with `f` and flatten results into a [List]      |
 | findFirst(p)           | return first element matching predicate `p` or throw (1)         |
 | findFirstOrNull(p)     | return first element matching predicate `p` or `null`                            |
 | first                  | first element (1)                                                               |
 | last                   | last element (1)                                                                |
 | take(n)                | return [Iterable] of up to n first elements                                     |
 | takeLast(n)            | return [Iterable] of up to n last elements                                      |
 | drop(n)                | return new [Iterable] without first n elements                                  |
 | dropLast(n)            | return new [Iterable] without last n elements                                   |
 | sum()                  | return sum of the collection applying `+` to its elements (3)                   |
 | sumOf(f)               | sum of the modified collection items (3)                                        |
 | sorted()               | return [List] with collection items sorted naturally                            |
 | 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                                              |
 (1)
 :: throws `NoSuchElementException` if there is no such element
 (2)
 :: `joinToString(separator=" ", transformer=null)`: separator is inserted between items if there are more than one, transformer is
 optional function applied to each item that must return result string for an item, otherwise `item.toString()` is used.
 (3)
 :: sum of empty collection is `null`
 ## Abstract methods:
    fun iterator(): Iterator
-## Instance methods
+For high-performance Kotlin-side interop and custom iterable implementation details, see [Efficient Iterables in Kotlin Interop](EfficientIterables.md).
 ### toList()
 Creates a list by iterating to the end. So, the Iterator should be finite to be used with it.
 ## Included in interfaces:
- Collection, Array, [List]
+- [Collection], Array, [List]
 ## Implemented in classes:
- [List], [Range]
+- [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
@@ -9,6 +9,9 @@ To implement the iterator you need to implement only two abstract methods:
 ### hasNext(): Bool
    // lets test
    // offset
 Should return `true` if call to `next()` will return valid next element.
 ### next(): Obj
@@ -20,6 +23,8 @@ must throw `ObjIterationFinishedError`.
 Iterators are returned when implementing [Iterable] interface.
 For high-performance Kotlin-side interop and custom iterable implementation details, see [Efficient Iterables in Kotlin Interop](EfficientIterables.md).
 ## Implemented for classes:
 - [List], [Range]
@@ -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`
@@ -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.
@@ -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`:
@@ -20,14 +22,21 @@ indexing is zero-based, as in C/C++/Java/Kotlin, etc.
    list[1]
    >>> 20
-Using negative indexes has a special meaning: _offset from the end of the list_:
+There is a shortcut for the last:
    val list = [10, 20, 30]
-    list[-1]
+    [list.last, list.lastIndex]
-    >>> 30
+    >>> [30,2]
 __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:
@@ -36,9 +45,20 @@ 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 concatenate [Iterable] objects instead of appending them. To append [Iterable] instance itself, use `list.add`:
+To append to lists, use `+=` with elements, lists and any [Iterable] instances, but beware it will
 concatenate [Iterable] objects instead of appending them. To append [Iterable] instance itself, use `list.add`:
    var list = [1, 2]
    val other = [3, 4]
@@ -57,7 +77,28 @@ To append to lists, use `+=` with elements, lists and any [Iterable] instances,
    >>> void
 ## Removing elements
 List is mutable, so it is possible to remove its contents. To remove a single element
 by index use:
    assertEquals( [1,2,3].removeAt(1), [1,3] )
    assertEquals( [1,2,3].removeAt(0), [2,3] )
    assertEquals( [1,2,3].removeLast(), [1,2] )
    >>> void
 There is a way to remove a range (see [Range] for more on ranges):
    assertEquals( [1, 4], [1,2,3,4].removeRange(1..2))
    assertEquals( [1, 4], [1,2,3,4].removeRange(1..<3))
    >>> void
 Open end ranges remove head and tail elements:
    assertEquals( [3, 4, 5], [1,2,3,4,5].removeRange(..1))
    assertEquals( [3, 4, 5], [1,2,3,4,5].removeRange(..<2))
    assertEquals( [1, 2], [1,2,3,4,5].removeRange( (2..) ))
    >>> void
 ## Comparisons
@@ -70,21 +111,162 @@ To append to lists, use `+=` with elements, lists and any [Iterable] instances,
    assert( [1, 2, 3] !== [1, 2, 3])
    >>> void
 ## Destructuring
 Lists can be used as L-values for destructuring assignments. This allows you to unpack list elements into multiple variables.
 ### Basic Destructuring
 ```lyng
 val [a, b, c] = [1, 2, 3]
 ```
 ### With Splats (Variadic)
 A single ellipsis `...` can be used to capture remaining elements into a list. It can be placed at the beginning, middle, or end of the pattern.
 ```lyng
 val [head, rest...] = [1, 2, 3] // head=1, rest=[2, 3]
 val [first, middle..., last] = [1, 2, 3, 4, 5] // first=1, middle=[2, 3, 4], last=5
 ```
 ### Nested Patterns
 Destructuring patterns can be nested to unpack multi-dimensional lists.
 ```lyng
 val [a, [b, c...], d] = [1, [2, 3, 4], 5]
 ```
 ### Reassignment
 Destructuring can also be used to reassign existing variables:
 ```lyng
 [x, y] = [y, x] // Swap values
 ```
 ## In-place sort
 List could be sorted in place, just like [Collection] provide sorted copies, in a very like way:
    val l1 = [6,3,1,9]
    l1.sort()
    assertEquals( [1,3,6,9], l1)
    l1.sortBy { -it }
    assertEquals( [1,3,6,9].reversed(), l1)
    l1.sort() // 1 3 6 9
    l1.sortBy { it % 4 }
    // 1,3,6,9 gives, mod 4:
    // 1 3 2 1
    // we hope we got it also stable:
    assertEquals( [1,9,6,3], l1)
    >>> void
 ## Members
 | name                          | meaning                                      | type        |
-|-----------------------------------|-------------------------------------|----------|
+|-------------------------------|----------------------------------------------|-------------|
 | `size`                        | current size                                 | Int         |
 | `add(elements...)`            | add one or more elements to the end          | Any         |
-| `addAt(index,elements...)`        | insert elements at position         | Int, Any |
+| `insertAt(index,elements...)` | insert elements at position                  | Int, Any    |
 | `removeAt(index)`             | remove element at position                   | Int         |
-| `removeRangeInclusive(start,end)` | remove range, inclusive (1)         | Int, Int |
+| `remove(from,toNonInclusive)` | remove range from (incl) to (nonincl)        | Int, Int    |
-|                                   |                                     |          |
+| `remove(Range)`               | remove range                                 | Range       |
 | `removeLast()`                | remove last element                          |             |
 | `removeLast(n)`               | remove n last elements                       | Int         |
 | `contains(element)`           | check the element is in the list (1)         |             |
 | `[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        |
 | `shuffle()`                   | in-place shuffle contents                    |             |
 | `toString()`                 | string representation like `[a,b,c]`         |             |
 (1)
-: end-inclisiveness allows to use negative indexes to, for exampe, remove several last elements, like `list.removeRangeInclusive(-2, -1)` will remove two last elements.
+: optimized implementation that override `Array` one
 (2)
 : `+=` append either a single element, or all elements if the List or other Iterable
 instance is appended. If you want to append an Iterable object itself, use `add` instead.
 (3)
 : predicate is called on each element, and the returned values are used to sort in natural
 order, e.g. is same as `list.sortWith { a,b -> predicate(a) <=> predicate(b) }`
 (4)
 : comparator callable takes tho arguments and must return: negative value when first is less,
 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  |
 |------------------|--------------------------------|-------|
 | `last`           | last element (throws if empty) |       |
 | `lastOrNull`     | last element or null           |       |
 | `lastIndex`      |                                | Int   |
 | `indices`        | range of indexes               | Range |
 | `contains(item)` | test that item is in the list  |       |
 (1)
 : end-inclisiveness allows to use negative indexes to, for exampe, remove several last elements, like
 `list.removeRangeInclusive(-2, -1)` will remove two last elements.
-# Notes
+[Range]: Range.md
-Could be rewritten using array as a class but List as the interface
+[Iterable]: Iterable.md
 [ImmutableList]: ImmutableList.md
 [ObservableList]: ObservableList.md
@@ -0,0 +1,181 @@
 # Map
 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].
 Important thing is that maps can't contain `null`: it is used to return from missing elements.
 Constructed map instance is of class `Map` and implements `Collection` (and therefore `Iterable`).
    val oldForm = Map( "foo" => 1, "bar" => "buzz" )
    assert(oldForm is Map)
    assert(oldForm.size == 2)
    assert(oldForm is Iterable)
    >>> void
 Notice usage of the `=>` operator that creates `MapEntry`, which also implements [Collection] of
 two items, first, at index zero, is a key, second, at index 1, is the value. You can use lists too.
 Map keys could be any objects (hashable, e.g. with reasonable hashCode, most of standard types are). You can access elements with indexing operator:
    val map = Map( ["foo", 1], ["bar", "buzz"], [42, "answer"] )
    assert( map["bar"] == "buzz")
    assert( map[42] == "answer" )
    assertEquals( null, map["nonexisting"])
    assert( map.getOrNull(101) == null )
    assert( map.getOrPut(911) { "nine-eleven" } == "nine-eleven" )
    // now 91 entry is set:
    assert( map[911] == "nine-eleven" )
    map["foo"] = -1
    assert( map["foo"] == -1)
    >>> void
 ## Map literals { ... }
 Lyng supports JavaScript-like map literals. Keys can be string literals or identifiers, and there is a handy identifier shorthand:
 - String key: `{ "a": 1 }`
 - Identifier key: `{ foo: 2 }` is the same as `{ "foo": 2 }`
 - Identifier shorthand: `{ foo: }` is the same as `{ "foo": foo }`
 Access uses brackets: `m["a"]`.
    val x = 10
    val y = 10
    val m = { "a": 1, x: x * 2, y: }
    assertEquals(1, m["a"])      // string-literal key
    assertEquals(20, m["x"])     // identifier key
    assertEquals(10, m["y"])     // identifier shorthand expands to y: y
    >>> void
 Trailing commas are allowed for nicer diffs and multiline formatting:
    val m = {
        "a": 1,
        b: 2,
    }
    assertEquals(1, m["a"]) 
    assertEquals(2, m["b"]) 
    >>> void
 Empty `{}` is reserved for blocks/lambdas; use `Map()` for an empty map.
 To remove item from the collection. use `remove`. It returns last removed item or null. Be careful if you 
 hold nulls in the map - this is not a recommended practice when using `remove` returned value. `clear()` 
 removes all.
    val map = Map( "foo" => 1, "bar" => "buzz", [42, "answer"] )
    assertEquals( 1, map.remove("foo") )
    assert( map.getOrNull("foo") == null)
    assert( map.size == 2 )
    map.clear()
    assert( map.size == 0 )
    >>> void
 Map implements [contains] method that checks _the presence of the key_ in the map:
    val map = Map( ["foo", 1], ["bar", "buzz"], [42, "answer"] )
    assert( "foo" in map )
    assert( "answer" !in map )
    >>> void
 To iterate maps it is convenient to use `keys` method that returns [Set] of keys (keys are unique:
    val map = Map( ["foo", 1], ["bar", "buzz"], [42, "answer"] )
    for( k in map.keys ) println(map[k])
    >>> 1
    >>> buzz
    >>> answer
    >>> void
 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 ) {
        val e: MapEntry = entry as MapEntry
        println("map[%s] = %s"(e.key, e.value))
    }
    void
    >>> map[foo] = 1
    >>> map[bar] = buzz
    >>> map[42] = answer
    >>> void
 There is a shortcut to use `MapEntry` to create maps: operator `=>` which creates `MapEntry`:
    val entry = "answer" => 42
    assert( entry is MapEntry )
    >>> void
 And you can use it to construct maps:
    val map = Map( "foo" => 1, "bar" => 22)
    assertEquals(1, map["foo"])
    assertEquals(22, map["bar"])
    >>> void
 Or use `.toMap` on anything that implements [Iterable] and which elements implements [Array] with 2 elements size, for example, `MapEntry`:
    val map = ["foo" => 1, "bar" => 22].toMap()
    assert( map is Map )
    assertEquals(1, map["foo"])
    assertEquals(22, map["bar"])
    >>> void
 It is possible also to get values as [List] (values are not unique):
    val map = Map( ["foo", 1], ["bar", "buzz"], [42, "answer"] )
    assertEquals(map.values, [1, "buzz", "answer"] )
    >>> void
 Map could be tested to be equal: when all it key-value pairs are equal, the map
 is equal.
    val m1 = Map(["foo", 1])
    val m2 = Map(["foo", 1])
    val m3 = Map(["foo", 2])
    assert( m1 == m2 )
    // but the references are different:
    assert( m1 !== m2 )
    // different maps:
    assert( m1 != m3 )
    >>> void
 ## Spreads and merging
 Inside map literals you can spread another map with `...` and items will be merged left-to-right; rightmost wins:
    val base = { a: 1, b: 2 }
    val m = { a: 0, ...base, b: 3, c: 4 }
    assertEquals(1, m["a"])  // base overwrites a:0
    assertEquals(3, m["b"])  // literal overwrites spread
    assertEquals(4, m["c"])  // new key
    >>> void
 Maps and entries can also be merged with `+` and `+=`:
    val m1 = ("x" => 1) + ("y" => 2)
    assertEquals(1, m1["x"])
    assertEquals(2, m1["y"])
    val m2 = { "a": 10 } + ("b" => 20)
    assertEquals(10, m2["a"])
    assertEquals(20, m2["b"])
    var m3 = { a: 1 }
    m3 += ("b" => 2)
    assertEquals(1, m3["a"])
    assertEquals(2, m3["b"])
    >>> void
 Notes:
 - Map literals always use string keys (identifier keys are converted to strings).
 - Spreads inside map literals and `+`/`+=` merges allow any objects as keys.
 - 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
@@ -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.
@@ -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
@@ -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).
@@ -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
+Finite ranges are [Iterable] and can be used in loops and list conversions.
-use finite ranges in loops and convert it to lists:
+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:
@@ -87,7 +153,7 @@ You can use Char as both ends of the closed range:
 Exclusive end char ranges are supported too:
    ('a'..<'c').toList 
-    >>> ['a', 'b']
+    >>> [a,b]
 # Instance members
@@ -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          |
+| step            | explicit iteration step      | Any?          |
-| end             |                              | Bool          |
+| 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
@@ -16,9 +16,12 @@ you can use it's class to ensure type:
 ## Member functions
 | name            | meaning                                                     | 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 |
 |                 |                                                             |      |
 |                 |                                                             |      |
 |                 |                                                             |      |
@@ -0,0 +1,104 @@
 # Regular expressions
 In lyng, you create regular expressions using class `Regex` or `String.re` methods:
    assert( "\d*".re is Regex )
    assert( Regex("\d*") is Regex )
    >>> void
 We plan to add slash syntax at some point.
 To check that some string matches as whole to some regex:
    assert( "123".matches("\d{3}".re) )
    assert( !"123".matches("\d{4}".re) )
    assert( !"1234".matches("\d".re) )
    >>> void
 To check that _part of the string_ matches some regular expession, use _match operator_ `=~` just like in Ruby, and its
 counterpart, _not match_ operator `!~`:
    assert( "abc123def" =~ "\d\d\d".re )
    assert( "abc" !~ "\d\d\d".re )
    >>> void
 When you need to find groups, and more detailed match information, use `Regex.find`:
    val result: RegexMatch? = Regex("abc(\d)(\d)(\d)").find( "bad456 good abc123")
    assert( result  != null )
    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
 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 )
    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`.
 Note that `=~` and `!~` operators against strings and regular expressions are commutative, e.g. regular expression and a
 string can be either left or right operator, but not both:
    assert( "abc" =~ "\wc".re )
    assert( "abc" !~ "\w1c".re )
    assert( "a\wc".re =~ "abcd" )
    assert( "a[a-z]c".re !~ "a2cd" )
    >>> void
 Also, string indexing is Regex-aware, and works like `Regex.find` (_not findall!_):
    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
 # Regex class reference
 | name         | description                         | notes |
 |--------------|-------------------------------------|-------|
 | matches(str) | true if the whole `str` matches     |       |
 | find(str)    | find first match in `str` or null   | (1)   |
 | findAll(str) | find all matches in `str` as [List] | (1)   |
 (1)
 :: See `RegexMatch` class description below
 # RegexMatch
 | name  | description                               | notes |
 |-------|-------------------------------------------|-------|
 | range | the [Range] of the match in source string |       |
 | value | the value that matches                    |       |
 | [n]   | [List] of group matches                   | (1)   |
 (1)
 :: the [0] element is always value, [1] is group 1 match of any, etc.
 [List]: List.md
 [Range]: Range.md
@@ -0,0 +1,52 @@
 # RingBuffer
 This is a fixed size buffer that allow to store N last elements with _O(1)_ effectiveness (no data shifting).
 Here is the sample:
    val r = RingBuffer(3)
    assert( r is RingBuffer )
    assertEquals(0, r.size)
    assertEquals(3, r.capacity)
    r += 10
    assertEquals(1, r.size)
    assertEquals(10, r.first)
    r += 20
    assertEquals(2, r.size)
    assertEquals( [10, 20], r.toList() )
    r += 30
    assertEquals(3, r.size)
    assertEquals( [10, 20, 30], r.toList() )
    // now first value is lost:
    r += 40
    assertEquals(3, r.size)
    assertEquals( [20, 30, 40], r.toList() )
    assertEquals(3, r.capacity)
    >>> void
 Ring buffer implements [Iterable], so any of its methods are available for `RingBuffer`, e.g. `first`, `last`, `toList`,
 `take`, `drop`, `takelast`, `dropLast`, etc.
 ## Constructor
    RinbBuffer(capacity: Int)
 ## Instance methods
 | method      | description            | remarks |
 |-------------|------------------------|---------|
 | capacity    | max size of the buffer |         |
 | size        | current size           | (1)     |
 | operator += | add new item           | (1)     |
 | add(item)   | add new item           | (1)     |
 | iterator()  | return iterator        | (1)     |
 (1)
 : Ringbuffer is not threadsafe, protect it with a mutex to avoid RC where necessary.
 [Iterable]: Iterable.md
@@ -0,0 +1,96 @@
 # 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 )
    assert( "foo" in set)
    assert( "bar" !in set)
    >>> void
 ## Set is collection and therefore [Iterable]:
    assert( Set(1,2) is Set)
    assert( Set(1,2) is Iterable)
    assert( Set(1,2) is Collection)
    >>> void
 So it supports all methods from [Iterable]; set is not, though, an [Array] and has
 no indexing. Use [set.toList] as needed.
 ## Set operations
    // Union
    assertEquals( Set(1,2,3,4), Set(3, 1) + Set(2, 4))
    // intersection
    assertEquals( Set(1,4), Set(3, 1, 4).intersect(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 `-`:
    assertEquals( Set( 1, 2), Set(1, 2, 4, 3) - Set(3, 4))
    >>> void
 ## Adding elements
    var s = Set()
    s += 1
    assertEquals( Set(1), s)
    s += [3, 3, 4]
    assertEquals( Set(3, 4, 1), s)
    >>> void
 ## Removing elements
 List is mutable, so it is possible to remove its contents. To remove a single element
 by index use:
    var s = Set(1,2,3)
    s.remove(2)
    assertEquals( s, Set(1,3) )
    s = Set(1,2,3)
    s.remove(2,1)
    assertEquals( s, Set(3) )
    >>> void
 Note that `remove` returns true if at least one element was actually removed and false
 if the set has not been changed.
 ## Comparisons and inclusion
 Sets are only equal when contains exactly same elements, order, as was said, is not significant:
    assert( Set(1, 2) == Set(2, 1) )
    assert( Set(1, 2, 2) == Set(2, 1) )
    assert( Set(1, 3) != Set(2, 1) )
    assert( 1 in Set(5,1))
    assert( 10 !in Set(5,1))
    >>> void
 ## Members
 | name                | meaning                              | type  |
 |---------------------|--------------------------------------|-------|
 | `size`              | current size                         | Int   |
 | `+=`                | add one or more elements             | Any   |
 | `+`, `union`         | union sets                           | Any   |
 | `-`, `subtract`     | subtract sets                        | Any   |
 | `*`, `intersect`    | subtract sets                        | Any   |
 | `remove(items...)`  | remove one or more items             | Range |
 | `contains(element)` | check the element is in the list (1) |       |
 (1)
 : optimized implementation that override `Iterable` one
 Also, it inherits methods from [Iterable].
 [Range]: Range.md
 [ImmutableSet]: ImmutableSet.md
@@ -0,0 +1,121 @@
 # Testing and Assertions
 Lyng provides several built-in functions for testing and verifying code behavior. These are available in all scripts.
 ## Basic Assertions
 ### `assert`
 Assert that a condition is true.
    assert(condition, message=null)
 - `condition`: A boolean expression.
 - `message` (optional): A string message to include in the exception if the assertion fails.
 If the condition is false, it throws an `AssertionFailedException`.
 ```lyng
 assert(1 + 1 == 2)
 assert(true, "This should be true")
 ```
 ### `assertEquals` and `assertEqual`
 Assert that two values are equal. `assertEqual` is an alias for `assertEquals`.
    assertEquals(expected, actual)
    assertEqual(expected, actual)
 If `expected != actual`, it throws an `AssertionFailedException` with a message showing both values.
 ```lyng
 assertEquals(4, 2 * 2)
 assertEqual("hello", "hel" + "lo")
 ```
 ### `assertNotEquals`
 Assert that two values are not equal.
    assertNotEquals(unexpected, actual)
 If `unexpected == actual`, it throws an `AssertionFailedException`.
 ```lyng
 assertNotEquals(5, 2 * 2)
 ```
 ## Exception Testing
 ### `assertThrows`
 Assert that a block of code throws an exception.
    assertThrows(code)
    assertThrows(expectedExceptionClass, code)
 - `expectedExceptionClass` (optional): The class of the exception that is expected to be thrown.
 - `code`: A lambda block or statement to execute.
 If the code does not throw an exception, an `AssertionFailedException` is raised. 
 If an `expectedExceptionClass` is provided, the thrown exception must be of that class (or its subclass), otherwise an error is raised.
 `assertThrows` returns the caught exception object if successful.
 ```lyng
 // Just assert that something is thrown
 assertThrows { 1 / 0 }
 // Assert that a specific exception class is thrown
 assertThrows(NoSuchElementException) { 
    [1, 2, 3].findFirst { it > 10 } 
 }
 // You can use the returned exception
 val ex = assertThrows { throw Exception("custom error") }
 assertEquals("custom error", ex.message)
 ```
 ## Other Validation Functions
 While not strictly for testing, these functions help in defensive programming:
 ### `require`
    require(condition, message="requirement not met")
 Throws an `IllegalArgumentException` if the condition is false. Use this for validating function arguments.
 If we want to evaluate the message lazily:
    require(condition) { "requirement not met: %s"(someData) }
 In this case, formatting will only occur if the condition is not met.
 ### `check`
    check(condition, message="check failed")
 Throws an `IllegalStateException` if the condition is false. Use this for validating internal state.
 With lazy message evaluation:
    check(condition) { "check failed: %s"(someData) }
 In this case, formatting will only occur if the condition is not met.
 ### TODO
 It is easy to mark some code and make it throw a special exception at cone with:
    TODO()
 or
    TODO("some message")
 It raises an `NotImplementedException` with the given message. You can catch it
 as any other exception when necessary.
 Many IDE and editors have built-in support for marking code with TODOs.
@@ -1,5 +1,7 @@
 # Advanced topics
 __See also:__ [parallelism].
 ## Closures/scopes isolation
 Each block has own scope, in which it can safely use closures and override
@@ -87,3 +89,80 @@ Lambda functions remember their scopes, so it will work the same as previous:
    println(c)
    >> 1
    >> void
 # Elements of functional programming
 With ellipsis and splats you can create partial functions, manipulate
 arguments list in almost arbitrary ways. For example:
    // Swap first and last arguments for call
    fun swap_args(first, others..., last, f) { 
        f(last, ...others, first) 
    }
    fun glue(args...) {
        var result = ""
        for( a in args ) result += a
    }
    // loop variables are read-only inside the loop body
    assertEquals( 
        "4231", 
        swap_args( 1, 2, 3, 4, glue)
    )
    >>> void
 # Annotations
 Annotation in Lyng resembles these proposed for Javascript. Annotation is just regular functions that, if used as annotation, are called when defining a function, var, val or class. 
 ## Function annotation
 When used without params, annotation calls a function with two arguments: actual function name and callable function body. Function annotation __must return callable for the function__, either what it received as a second argument (most often), or something else. Annotation name convention is upper scaled: 
    var annotated = false
    // this is annotation function:
    fun Special(name, body) {
        assertEquals("foo", name)
        annotated = true
        { body(it) + 100 }
    }
    @Special
    fun foo(value) { value + 1 }
    assert(annotated)
    assertEquals(111, foo( 10 ))
    >>> void
 Function annotation can have more args specified at call time. There arguments must follow two mandatory ones (name and body). Use default values in order to allow parameterless annotation to be used simultaneously.
    val registered = Map()
    // it is recommended to provide defaults for extra parameters:
    fun Registered(name, body, overrideName = null) {
        registered[ overrideName ?: name ] = body
        body
    }
    // witout parameters is Ok as we provided default value
    @Registered
    fun foo() { "called foo" }
    @Registered("bar") 
    fun foo2() { "called foo2" }
    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
 ## Scopes and Closures: resolution and safety
 Closures and dynamic scope graphs require care to avoid accidental recursion and to keep name resolution predictable. See the dedicated page for detailed rules, helper APIs, and best practices:
 - Scopes and Closures: resolution and safety → [scopes_and_closures.md](scopes_and_closures.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.
@@ -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.
@@ -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`).
@@ -1,24 +0,0 @@
 # Classes
 ## Declaring
    class Foo1
    class Foo2() // same, empty constructor
    class Foo3() { // full
    }
    class Foo4 { // Only body
    }
 ```
 class_declaration = ["abstract",] "class" [, constructor] [, body]
 constructor = "(", [field [, field]] ")
 field = [visibility ,] [access ,] name [, typedecl]
 body = [visibility] ("var", vardecl) | ("val", vardecl) | ("fun", fundecl)
 visibility = "private" | "protected" | "internal"
 ```
 ### Abstract classes
 Contain one pr more abstract methods which must be implemented; though they
 can have constructors, the instances of the abstract classes could not be
 created independently
@@ -1,11 +1,13 @@
 # Declaring arguments in Lyng
 [//]: # (topMenu)
 It is a common thing that occurs in many places in Lyng, function declarations, 
 lambdas and class declarations.
 ## Regular
-## default values
+## Default values
 Default parameters should not be mixed with mandatory ones:
@@ -32,13 +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
@@ -47,7 +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
@@ -56,7 +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")
@@ -73,6 +80,13 @@ destructuring arrays when calling functions and lambdas:
    getFirstAndLast( ...(1..10) ) // see "splats" section below
    >>> [1,10]
 Note that array destructuring can also be used in assignments:
    val [first, middle..., last] = [1, 2, 3, 4, 5]
    [x, y] = [y, x] // Swap
 See [tutorial] and [List] documentation for more details on destructuring assignments.
 # Splats
 Ellipsis allows to convert argument lists to lists. The inversa algorithm that converts [List],
@@ -83,7 +97,7 @@ or whatever implementing [Iterable], is called _splats_. Here is how we use it:
    }
    val array = [1,2,3]
    testSplat("start", ...array, "end")
-    >>> ["start", 1, 2, 3, "end"]
+    >>> [start,1,2,3,end]
    >>> void
 There could be any number of splats at any positions. You can splat any other [Iterable] type:
@@ -93,8 +107,76 @@ There could be any number of splats at any positions. You can splat any other [I
    }
    val range = 1..3
    testSplat("start", ...range, "end")
-    >>> ["start", 1, 2, 3, "end"]
+    >>> [start,1,2,3,end]
    >>> void
 ## Named arguments in calls
 Lyng supports named arguments at call sites using colon syntax `name: value`. 
 ### Shorthand for Named Arguments
 If you want to pass a variable as a named argument and the variable has the same name as the parameter, you can omit the value and use the shorthand `name:`. This is highly readable and matches the shorthand for map literals.
 ```lyng
    fun test(a, b, c) { [a, b, c] }
    val a = 1
    val b = 2
    val c = 3
    // Explicit:
    assertEquals([1, 2, 3], test(a: a, b: b, c: c))
    // Shorthand (preferred):
    assertEquals([1, 2, 3], test(a:, b:, c:))
 ```
 This shorthand is elegant, reduces boilerplate, and is consistent with Lyng's map literal syntax. It works for both function calls and class constructors.
 Rules for named arguments:
 - Named arguments must follow positional arguments. After the first named argument, no positional arguments may appear inside the parentheses.
 - The only exception is the syntactic trailing block after the call: `f(args) { ... }`. This block is outside the parentheses and is handled specially (see below).
 - A named argument cannot reassign a parameter already set positionally.
 - If the last parameter has already been assigned by a named argument (or named splat), a trailing block is not allowed and results in an error.
 Why `:` and not `=` at call sites? In Lyng, `=` is an expression (assignment), so we use `:` to avoid ambiguity. This is a key difference from **Kotlin**, which uses `=` for named arguments. Declarations in Lyng continue to use `:` for types, while call sites use `as` / `as?` for type operations.
 ## Named splats (map splats)
 Splat (`...`) of a Map provides named arguments to the call. Only string keys are allowed. You can use the same auto-substitution shorthand inside map literals used for splats:
 ```lyng
    fun test(a="a", b="b", c="c", d="d") { [a, b, c, d] }
    val b = "B!"
    val d = "D!"
    // Auto-substitution in map literal:
    val patch = { d:, b: }
    val r = test("A?", ...patch)
    assertEquals(["A?","B!","c","D!"], r)
 ```
 Constraints:
 - Map splat keys must be strings; otherwise, a clean error is thrown.
 - Named splats cannot duplicate parameters already assigned (by positional or named arguments).
 - Named splats must follow all positional arguments and positional splats.
 - Ellipsis parameters (variadic) remain positional-only and cannot be assigned by name.
 ## Trailing-lambda rule interaction
 If a call is immediately followed by a block `{ ... }`, it is treated as an extra last argument and bound to the last parameter. However, if the last parameter is already assigned by a named argument or a named splat, using a trailing block is an error:
 ```lyng
    fun f(x, onDone) { onDone(x) }
    f(x: 1) { 42 }   // ERROR
    f(1) { 42 }      // OK
 ```
 [tutorial]: tutorial.md
 [List]: List.md
@@ -0,0 +1,194 @@
 # Delegation in Lyng
 Delegation is a powerful pattern that allows you to outsource the logic of properties (`val`, `var`) and functions (`fun`) to another object. This enables code reuse, separation of concerns, and the implementation of common patterns like lazy initialization, observable properties, and remote procedure calls (RPC) with minimal boilerplate.
 ## The `by` Keyword
 Delegation is triggered using the `by` keyword in a declaration. The expression following `by` is evaluated once when the member is initialized, and the resulting object becomes the **delegate**.
 ```lyng
 val x by MyDelegate()
 var y by MyDelegate()
 fun f by MyDelegate()
 ```
 ## The Unified Delegate Model
 A delegate object can implement any of the following methods to intercept member access. All methods receive the `thisRef` (the instance containing the member) and the `name` of the member.
 ```lyng
 interface Delegate {
    // Called when a 'val' or 'var' is read
    fun getValue(thisRef, name)
    // Called when a 'var' is assigned
    fun setValue(thisRef, name, newValue)
    // Called when a 'fun' is invoked
    fun invoke(thisRef, name, args...)
    // Optional: Called once during initialization to "bind" the delegate
    // Can be used for validation or to return a different delegate instance
    fun bind(name, access, thisRef) = this
 }
 ```
 ### Delegate Access Types
 The `bind` method receives an `access` parameter of type `DelegateAccess`, which can be one of:
 - `DelegateAccess.Val`
 - `DelegateAccess.Var`
 - `DelegateAccess.Callable` (for `fun`)
 ## Usage Cases and Examples
 ### 1. Lazy Initialization
 The classic `lazy` pattern ensures a value is computed only when first accessed and then cached. In Lyng, `lazy` is implemented as a class that follows this pattern. While classes typically start with an uppercase letter, `lazy` is an exception to make its usage feel like a native language feature.
 ```lyng
 class lazy(val creator) : Delegate {
    private var value = Unset
    override fun bind(name, access, thisRef) {
        if (access != DelegateAccess.Val) throw "lazy delegate can only be used with 'val'"
        this
    }
    override fun getValue(thisRef, name) {
        if (value == Unset) {
            // calculate value using thisRef as this:
            value = with(thisRef) creator()
        }
        value
    }
 }
 // Usage:
 val expensiveData by lazy {
    println("Performing expensive computation...")
    42
 }
 println(expensiveData) // Computes and prints 42
 println(expensiveData) // Returns 42 immediately
 ```
 ### 2. Observable Properties
 Delegates can be used to react to property changes.
 ```lyng
 class Observable(initialValue, val onChange) {
    private var value = initialValue
    fun getValue(thisRef, name) = value
    fun setValue(thisRef, name, newValue) {
        val oldValue = value
        value = newValue
        onChange(name, oldValue, newValue)
    }
 }
 class User {
    var name by Observable("Guest") { name, old, new ->
        println("Property %s changed from %s to %s"(name, old, new))
    }
 }
 val u = User()
 u.name = "Alice" // Prints: Property name changed from Guest to Alice
 ```
 ### 3. Function Delegation (Proxies)
 You can delegate an entire function to an object. This is particularly useful for implementing decorators or RPC clients.
 ```lyng
 object LoggerDelegate {
    fun invoke(thisRef, name, args...) {
        println("Calling function: " + name + " with args: " + args)
        // Logic here...
        "Result of " + name
    }
 }
 fun remoteAction by LoggerDelegate
 println(remoteAction(1, 2, 3))
 // Prints: Calling function: remoteAction with args: [1, 2, 3]
 // Prints: Result of remoteAction
 ```
 ### 4. Stateless Delegates (Shared Singletons)
 Because `getValue`, `setValue`, and `invoke` receive `thisRef`, a single object can act as a delegate for multiple properties across many instances without any per-property memory overhead.
 ```lyng
 object Constant42 {
    fun getValue(thisRef, name) = 42
 }
 class Foo {
    val a by Constant42
    val b by Constant42
 }
 val f = Foo()
 assertEquals(42, f.a)
 assertEquals(42, f.b)
 ```
 ### 5. Local Delegation
 Delegation is not limited to class members; you can also use it for local variables inside functions.
 ```lyng
 fun test() {
    val x by LocalProxy(123)
    println(x)
 }
 ```
 ### 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:
 1.  **Validate usage**: Throw an error if the delegate is used with the wrong member type (e.g., `lazy` on a `var`).
 2.  **Initialize state**: Set up internal state based on the property name or the containing instance.
 3.  **Substitute itself**: Return a different object that will act as the actual delegate.
 ```lyng
 class ValidatedDelegate() {
    fun bind(name, access, thisRef) {
        if (access == DelegateAccess.Var) {
            throw "This delegate cannot be used with 'var'"
        }
        this
    }
    fun getValue(thisRef, name) = "Validated"
 }
 ```
 ## Summary
 Delegation in Lyng combines the elegance of Kotlin-style properties with the flexibility of dynamic function interception. By unifying `val`, `var`, and `fun` delegation into a single model, Lyng provides a consistent and powerful tool for meta-programming and code reuse.
@@ -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
@@ -0,0 +1,720 @@
 # 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
 - 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
 All snippets below use idiomatic Kotlin and rely on Lyng public APIs. They work on JVM and other Kotlin Multiplatform targets supported by `lynglib`.
 Note: all Lyng APIs shown are `suspend`, because script evaluation is coroutine‑friendly and can suspend.
 ### 1) Add Lyng to your build
 Add the repository where you publish Lyng artifacts and the dependency on the core library `lynglib`.
 Gradle Kotlin DSL (build.gradle.kts):
 ```kotlin
 repositories {
    // Your standard repos
    mavenCentral()
    // If you publish to your own Maven (example: Gitea packages). Adjust URL/token as needed.
    maven(url = uri("https://gitea.sergeych.net/api/packages/SergeychWorks/maven"))
 }
 dependencies {
    // Multiplatform: place in appropriate source set if needed
    implementation("net.sergeych:lynglib:1.0.0-SNAPSHOT")
 }
 ```
 If you use Kotlin Multiplatform, add the dependency in the `commonMain` source set (and platform‑specific sets if you need platform APIs).
 ### 2) Preferred runtime: `EvalSession`
 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
    val result = scope.eval("1 + 2 * 3")
    println("Lyng result: $result")
 }
 ```
 You can also pre‑compile a script and execute it multiple times on the same scope:
 ```kotlin
 val script = Compiler.compile("""
    // any Lyng code
    val x = 40 + 2
    x
 """)
 val run1 = script.execute(scope)
 val run2 = script.execute(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) Preferred: bind extern globals from Kotlin
 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
 import net.sergeych.lyng.bridge.*
 import net.sergeych.lyng.obj.ObjInt
 import net.sergeych.lyng.obj.ObjString
 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())
    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
    )
 }
 ```
 Usage from Lyng:
 ```lyng
 import my.api
 assertEquals(42, globalFun(41))
 assertEquals("initial", globalProp)
 globalProp = "changed"
 assertEquals("changed", globalProp)
 assertEquals("1.0.0", globalVersion)
 ```
 Minimal rule of thumb:
 - 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
    ObjInt(x.value + 1)
 }
 // A void function (returns Lyng Void)
 scope.addVoidFn("log") {
    val items = args.list // List<Obj>
    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
 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)
 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)
 ```
 In Lyng:
 ```lyng
 val instance = MyClass()
 println(instance.version) // -> "1.0.0"
 instance.count = 5
 println(instance.count)   // -> 5
 ```
 ### 6) Add Kotlin‑backed properties
 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
 myClass.addProperty(
    name = "value",
    getter = { 
        // Return current value as a Lyng object
        ObjInt(internalValue) 
    },
    setter = { newValue ->
        // newValue is passed as a Lyng object (the first and only argument)
        internalValue = (newValue as ObjInt).value
    }
 )
 // 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)
 ```
 Usage in Lyng:
 ```lyng
 val instance = MyClass()
 println(instance.value) // -> 10
 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 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"))
 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.
 ### 8) Execute scripts with parameters; call Lyng functions from Kotlin
 There are two convenient patterns.
 1) Evaluate a Lyng call expression directly:
 ```kotlin
 // Suppose Lyng defines: fun add(a, b) = a + b
 val session = EvalSession()
 val scope = session.getScope()
 session.eval("fun add(a, b) = a + b")
 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
 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
 // Create a child scope with arguments (as Lyng Objs)
 val callScope = scope.createChildScope(
    args = Arguments(listOf(ObjInt(20), ObjInt(22)))
 )
 val resultObj = addFn.execute(callScope)
 val result = resultObj.toKotlin(scope) // -> 42
 ```
 If you need to pass complex data (lists, maps), construct the corresponding Lyng `Obj` types (`ObjList`, `ObjMap`, etc.) and pass them in `Arguments`.
 ### 9) Create your own packages and import them in Lyng
 Lyng supports packages that are imported from scripts. You can register packages programmatically via `ImportManager` or by providing source texts that declare `package ...`.
 Key concepts:
 - `ImportManager` holds package registrations and lazily builds `ModuleScope`s when first imported.
 - Every `Scope` has `currentImportProvider` and (if it’s an `ImportManager`) a convenience `importManager` to register packages.
 Register a Kotlin‑built package:
 ```kotlin
 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 ->
    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
 session.eval("""
    import my.tools.*
    val v = triple(14)
    status = "busy"
 """)
 val v = session.eval("v").toKotlin(scope) // -> 42
 ```
 Register a package from Lyng source text:
 ```kotlin
 val pkgText = """
    package math.extra
    fun sqr(x) = x * x
 """.trimIndent()
 scope.importManager.addTextPackages(pkgText)
 session.eval("""
    import math.extra.*
    val s = sqr(12)
 """)
 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 `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, prefer a fresh `EvalSession()` per request. Use `Scope.new()` / `Script.newScope()` when you specifically need low-level raw scopes or modules.
 ```kotlin
 // 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
 - All values that cross the boundary must be Lyng `Obj` instances. Convert Kotlin values explicitly (e.g., `ObjInt`, `ObjReal`, `ObjString`).
 - Use `toKotlin(scope)` to get Kotlin values back. Collections convert to Kotlin collections recursively.
 - Most public API in Lyng is suspendable. If you are not already in a coroutine, wrap calls in `runBlocking { ... }` on the JVM for quick tests.
 - 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.
@@ -0,0 +1,267 @@
 # Exceptions handling
 Exceptions are widely used in modern programming languages, so
 they are implemented also in Lyng and in the most complete way.
 # Exception classes
 Exceptions are throwing instances of some class that inherits `Exception`
 across the code. Below is the list of built-in exceptions. Note that
 only objects that inherit `Exception` can be thrown. For example:
    assert( IllegalArgumentException() is Exception)
    >>> void
 # Try statement: catching exceptions
 There general pattern is:
 ```
 try_statement = try_clause, [catch_clause, ...], [finally_clause] 
 try_clause = "try", "{", statements, "}"
 catch_clause = "catch", [(full_catch | shorter_catch)], "{", statements "}"
 full_catch = "(", catch_var, ":", exception_class [, excetpion_class...], ")
 shorter_catch = "(", catch_var, ")"
 finally_clause = "{", statements, "}"
 ```
 Let's in details.
 ## Full catch block:
    val result = try {
        throw IllegalArgumentException("the test")
    }
    catch( x: IndexOutOfBoundsException, IllegalArgumentException) {
        x.message
    }
    catch(x: Exception) {
        "bad"
    }
    assertEquals(result, "the test")
    >>> void
 Because our exception is listed in a first catch block, it is processed there.
 The full form allow a single catch block to process exceptions with specified classes and bind actual caught object to
 the given variable. This is most common and well known form, implemented like this or similar in many other languages,
 like Kotlin, Java or C++.
 ## Shorter form
 When you want to catch _all_ the exceptions, you should write `catch(e: Exception)`,
 but it is somewhat redundant, so there is simpler variant:
    val sample2 = try {
        throw IllegalArgumentException("sample 2")
    }
    catch(x) {
        x.message
    }
    assertEquals( sample2, "sample 2" )
    >>> void
 But well most likely you will find default variable `it`, like in Kotlin, more than enough
 to catch all exceptions to, then you can write it even shorter:
    val sample2 = try {
        throw IllegalArgumentException("sample 3")
    }
    catch {
        it.message
    }
    assertEquals( sample2, "sample 3" )
    >>> void
 You can even check the type of the `it` and create more convenient and sophisticated processing logic. Such approach is
 used, for example, in Scala.
 ## finally block
 If `finally` block present, it will be executed after body (until first exception)
 and catch block, if any will match. finally statement is executed even if the
 exception will be thrown and not caught locally. It does not alter try/catch block result:
    try {
    }
    finally {
        println("called finally")
    }
    >>> called finally
    >>> void
 - and yes, there could be try-finally block, no catching, but perform some guaranteed cleanup.
 # Conveying data with exceptions
 The simplest way is to provide exception string and `Exception` class:
    try {
        throw Exception("this is my exception")
    }
    catch {
        it.message
    }
    >>> "this is my exception"
 This way, in turn, can also be shortened, as it is overly popular:
    try {
        throw "this is my exception"
    }
    catch {
        it.message
    }
    >>> "this is my exception"
 The trick, though, works with strings only, and always provide `Exception` instances, which is good for debugging but
 most often not enough.
 # Exception class
 Serializable class that conveys information about the exception. Important members and methods are:
 | name              | description                                            |
 |-------------------|--------------------------------------------------------|
 | message           | String message                                           |
 | 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:
 ```lyng
 class StackTraceEntry(
    val sourceName: String,
    val line: Int,
    val column: Int,
    val sourceString: String
 )
 ```
 - `sourceString` is a line extracted from sources. Note that it _is serialized and printed_, so if you want to conceal it, catch all exceptions and filter out sensitive information.
 # Custom error classes
 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 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 internal exception caught                  |
 | NotFoundException          |                                                       |
 | IllegalOperationException  |                                                       |
 | UnsetException             | access to uninitialized late-init val                 |
 | NotImplementedException    | used by `TODO()`                                      |
 | SyntaxError                |                                                       |
 ### Symbol resolution errors
 For compatibility, `SymbolNotFound` is an alias of `SymbolNotDefinedException`. You can catch either name in examples and tests.
 Example:
 ```lyng
 try {
    nonExistingMethod()
 }
 catch(e: SymbolNotFound) {
    // handle
 }
 ```
@@ -0,0 +1,62 @@
 # Lyng formatter (core, CLI, and IDE)
 This document describes the Lyng code formatter included in this repository. The formatter lives in the core library (`:lynglib`), is available from the CLI (`lyng fmt`), and is used by the IntelliJ plugin.
 ## Core library
 Package: `net.sergeych.lyng.format`
 - `LyngFormatConfig`
  - `indentSize` (default 4)
  - `useTabs` (default false)
  - `continuationIndentSize` (default 8)
  - `maxLineLength` (default 120)
  - `applySpacing` (default false)
  - `applyWrapping` (default false)
 - `LyngFormatter`
  - `reindent(text, config)` — recomputes indentation from scratch (braces, `else/catch/finally` alignment, continuation indent under `(` `)` and `[` `]`), idempotent.
  - `format(text, config)` — runs `reindent` and, depending on `config`, optionally applies:
    - a safe spacing pass (commas/operators/colons/keyword parens; member access `.` remains tight; no changes to strings/comments), and
    - a controlled wrapping pass for long call arguments (no trailing commas).
 Both passes are designed to be idempotent. Extensive tests live under `:lynglib/src/commonTest/.../format`.
 ## CLI formatter
 ```
 lyng fmt [--check] [--in-place|-i] [--spacing] [--wrap] <file1.lyng> [file2.lyng ...]
 ```
 - Defaults: indent-only; spacing and wrapping are OFF unless flags are provided.
 - `--check` prints files that would change and exits with code 2 if any changes are detected.
 - `--in-place`/`-i` rewrites files in place (default if not using `--check`).
 - `--spacing` enables the safe spacing pass (commas/operators/colons/keyword parens).
 - `--wrap` enables controlled wrapping of long call argument lists (respects `maxLineLength`, no trailing commas).
 Examples:
 ```
 # check formatting without modifying files
 lyng fmt --check docs/samples/fs_sample.lyng
 # format in place with spacing rules enabled
 lyng fmt --spacing -i docs/samples/fs_sample.lyng
 # format in place with spacing + wrapping
 lyng fmt --spacing --wrap -i src/**/*.lyng
 ```
 ## IntelliJ plugin
 - Indentation: always enabled, idempotent; the plugin computes per-line indent via the core formatter.
 - Spacing/wrapping: optional and OFF by default.
  - Settings/Preferences → Lyng Formatter provides toggles:
    - "Enable spacing normalization (commas/operators/colons/keyword parens)"
    - "Enable line wrapping (120 cols) [experimental]"
  - Reformat Code applies: indentation first, then spacing, then wrapping if toggled.
 ## Design notes
 - Single source of truth: The core formatter is used by CLI and IDE to keep behavior consistent.
 - Stability first: Spacing/wrapping are gated by flags/toggles; indentation from scratch is always safe and idempotent.
 - Non-destructive: The formatter carefully avoids changing string/char literals and comment contents.
@@ -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.
@@ -0,0 +1,32 @@
 # Plugin for IntelliJ IDEA
 [//]: # (excludeFromIndex)
 We introduce the alpha version of the plugin for IntelliJ IDEA 2024.3.x+ IDE variants. It is compatible with 2025.x and
 should be compatible with other IDEA flavors, notably [OpenIDE](https://openide.ru/). It supports the following features:
 - syntax highlighting (2 stage, fast and more accurate that analyses in background)
 - 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
 > this plugin over using a TextMate bundle.
 ### Install
 - From ZIP: download the archive below, then in IntelliJ IDEA open Settings/Preferences → Plugins →
  gear icon → Install Plugin from Disk… and select the downloaded ZIP. Restart IDE if prompted.
 - Alternatively, if/when the plugin is published to a marketplace, you will be able to install it
  directly from the “Marketplace” tab (not yet available).
 ### [Download plugin v0.0.2-SNAPSHOT](https://lynglang.com/distributables/lyng-idea-0.0.2-SNAPSHOT.zip)
 Your ideas and bugreports are welcome on the [project gitea page](https://gitea.sergeych.net/SergeychWorks/lyng/issues)
@@ -0,0 +1,179 @@
 # Json support
 Since 1.0.5 we start adding JSON support. Versions 1,0,6* support serialization of the basic types, including lists and
 maps, and simple classes. Multiple inheritance may produce incorrect results, it is work in progress.
 ## Serialization in Lyng
    // in lyng
    assertEquals("{\"a\":1}", {a: 1}.toJsonString())
    void
    >>> void
 Simple classes serialization is supported:
    import lyng.serialization
    class Point(foo,bar) {
        val t = 42
    }
    // val is not serialized
    assertEquals( "{\"foo\":1,\"bar\":2}", Point(1,2).toJsonString() )
    >>> void
 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
    class Point2(foo,bar) {
        var reason = 42
        // but we override json serialization:
        fun toJsonObject() {
            { "custom": true }
        }
    }
    // var is serialized instead
    assertEquals( "{\"custom\":true}", Point2(1,2).toJsonString() )
    >>> void
 Custom serialization of user classes is possible by overriding `toJsonObject` method. It must return an object which is
 serializable to Json. Most often it is a map, but any object is accepted, that makes it very flexible:
    import lyng.serialization
    class Point2(foo,bar) {
        var reason = 42
        // but we override json serialization:
        fun toJsonObject() {
            { "custom": true }
        }
    }
    class Custom {
        fun toJsonObject() {
            "full freedom"
        }
    }
    // var is serialized instead
    assertEquals( "\"full freedom\"", Custom().toJsonString() )
    >>> void
 Please note that `toJsonString` should be used to get serialized string representation of the object. Don't call
 `toJsonObject` directly, it is not intended to be used outside the serialization library.
 ## Kotlin side interfaces
 The "Batteries included" principle is also applied to serialization.
 - `Obj.toJson()` provides Kotlin `JsonElement`
 - `Obj.toJsonString()` provides Json string representation
 - `Obj.decodeSerializableWith()` and `Obj.decodeSerializable()` allows to decode Lyng classes as Kotlin objects using
  `kotlinx.serialization`:
 ```kotlin
 /**
 * Decodes the current object into a deserialized form using the provided deserialization strategy.
 * It is based on [Obj.toJson] and uses existing Kotlin Json serialization, without string representation
 * (only `JsonElement` to carry information between Kotlin and Lyng serialization worlds), thus efficient.
 *
 * @param strategy The deserialization strategy that defines how the object should be decoded.
 * @param scope An optional scope used during deserialization to define the context. Defaults to a new instance of Scope.
 * @return The deserialized object of type T.
 */
 suspend fun <T> Obj.decodeSerializableWith(strategy: DeserializationStrategy<T>, scope: Scope = Scope()): T =
    Json.decodeFromJsonElement(strategy, toJson(scope))
 /**
 * Decodes a serializable object of type [T] using the provided decoding scope. The deserialization uses
 * [Obj.toJson] and existing Json based serialization ithout using actual string representation, thus
 * efficient.
 *
 * @param T The type of the object to be decoded. Must be a reified type.
 * @param scope The scope used during decoding. Defaults to a new instance of [Scope].
 */
 suspend inline fun <reified T> Obj.decodeSerializable(scope: Scope = Scope()) =
    decodeSerializableWith<T>(serializer<T>(), scope)
 ```
 Note that lyng-2-kotlin deserialization with `kotlinx.serialization` uses JsonElement as information carrier without
 formatting and parsing actual Json strings. This is why we use `Json.decodeFromJsonElement` instead of
 `Json.decodeFromString`. Such an approach gives satisfactory performance without writing and supporting custom
 `kotlinx.serialization` codecs.
 ### Pitfall: JSON objects and Map<String, Any?>
 Kotlin serialization does not support `Map<String, Any?>` as a serializable type, more general, it can't serialize `Any`. This in particular means that you can deserialize Kotlin `Map<String, T>` as long as `T` is `@Serializable` in Kotlin: 
 ```kotlin
@Serializable
 data class TestJson2(
    val value: Int,
    val inner: Map<String,Int>
 )
@Test
 fun deserializeMapWithJsonTest() = runTest {
    val session = EvalSession()
    val x = session.eval("""
        import lyng.serialization
        { value: 1, inner: { "foo": 1, "bar": 2 }}
    """.trimIndent()).decodeSerializable<TestJson2>()
    // That works perfectly well:
    assertEquals(TestJson2(1, mapOf("foo" to 1, "bar" to 2)), x)
 }
 ```
 But what if your map has objects of different types? The approach of using polymorphism is partially applicable, but what to do with `{ one: 1, two: "two" }`?
 The answer is pretty simple: use `JsonObject` in your deserializable object. This class is capable of holding any JSON types and structures and is sort of a silver bullet for such cases:
 ~~~kotlin
@Serializable
 data class TestJson3(
    val value: Int,
    val inner: JsonObject
 )
@Test
 fun deserializeAnyMapWithJsonTest() = runTest {
    val session = EvalSession()
    val x = session.eval("""
        import lyng.serialization
        { value: 12, inner: { "foo": 1, "bar": "two" }}
    """.trimIndent()).decodeSerializable<TestJson3>()
    assertEquals(TestJson3(12, JsonObject(mapOf("foo" to JsonPrimitive(1), "bar" to Json.encodeToJsonElement("two")))), x)
 }
 ~~~
 # List of supported types
 | Lyng type | JSON type | notes       |
 |-----------|-----------|-------------|
 | `Int`     | number    |             |
 | `Real`    | number    |             |
 | `String`  | string    |             |
 | `Bool`    | boolean   |             |
 | `null`    | null      |             |
 | `Instant` | string    | ISO8601 (1) |
 | `List`    | array     | (2)         |
 | `Map`     | object    | (2)         |
 (1)
 : ISO8601 flavor 1970-05-06T06:00:00.000Z in used; number of fractional digits depends on the truncation
 on [Instant](time.md), see `Instant.truncateTo...` functions.
 (2)
 : List may contain any objects serializable to Json.
 (3)
 : Map keys must be strings, map values may be any objects serializable to Json.
@@ -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`.
@@ -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).
@@ -0,0 +1,236 @@
 ### lyng.io.fs — async filesystem access for Lyng scripts
 This module provides a uniform, suspend-first filesystem API to Lyng scripts, backed by Kotlin Multiplatform implementations.
 - JVM/Android/Native: Okio `FileSystem.SYSTEM` (non-blocking via coroutine dispatcher)
 - JS/Node: Node filesystem (currently via Okio node backend; a native `fs/promises` backend is planned)
 - JS/Browser and Wasm: in-memory virtual filesystem for now
 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 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.
 ---
 #### 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")
 }
 ```
 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 {
        maven("https://gitea.sergeych.net/api/packages/SergeychWorks/maven")
    }
 ```
 This brings in:
 - `:lynglib` (Lyng engine)
 - Okio (`okio`, `okio-fakefilesystem`, and `okio-nodefilesystem` for JS)
 - Kotlin coroutines
 ---
 #### Install the module into a Lyng session
 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:
 ```kotlin
 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 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).
 ---
 #### Using from Lyng scripts
 ```lyng
 val p = Path("/tmp/hello.txt")
 // Text I/O
 p.writeUtf8("Hello Lyng!\n")
 println(p.readUtf8())
 // Binary I/O
 val data = Buffer.fromHex("deadbeef")
 p.writeBytes(data)
 // Existence and directories
 assertTrue(p.exists())
 Path("/tmp/work").mkdirs()
 // Listing
 for (entry in Path("/tmp").list()) {
    println(entry)
 }
 // Globbing
 val txts = Path("/tmp").glob("**/*.txt").toList()
 // Copy / Move / Delete
 Path("/tmp/a.txt").copy("/tmp/b.txt", overwrite=true)
 Path("/tmp/b.txt").move("/tmp/c.txt", overwrite=true)
 Path("/tmp/c.txt").delete()
 // Streaming large files (does not load whole file into memory)
 var bytes = 0
 val it = Path("/tmp/big.bin").readChunks(1_048_576) // 1MB chunks
 val iter = it.iterator()
 while (iter.hasNext()) {
    val chunk = iter.next()
    bytes = bytes + chunk.size()
 }
 // Text chunks and lines
 for (s in Path("/tmp/big.txt").readUtf8Chunks(64_000)) {
    // process each string chunk
 }
 for (ln in Path("/tmp/big.txt").lines()) {
    // process line by line
 }
 ```
 ---
 #### API (Lyng class `Path`)
 Constructor:
 - `Path(path: String)` — creates a Path object
 - `Paths(path: String)` — alias
 File and directory operations (all suspend under the hood):
 - `name`: name, `String`
 - `segments`: list of parsed path segments (directories)
 - `parent`: parent directory, `Path?`; null if root
 - `exists(): Bool`
 - `isFile(): Bool` — true if the path points to a regular file (cached metadata)
 - `isDirectory(): Bool` — true if the path points to a directory (cached metadata)
 - `size(): Int?` — size in bytes or null if unknown (cached metadata)
 - `createdAt(): Instant?` — creation time as Lyng `Instant`, or null (cached metadata)
 - `createdAtMillis(): Int?` — creation time in epoch milliseconds, or null (cached metadata)
 - `modifiedAt(): Instant?` — last modification time as Lyng `Instant`, or null (cached metadata)
 - `modifiedAtMillis(): Int?` — last modification time in epoch milliseconds, or null (cached metadata)
 - `list(): List<Path>` — children of a directory
 - `readBytes(): Buffer`
 - `writeBytes(bytes: Buffer)`
 - `appendBytes(bytes: Buffer)`
 - `readUtf8(): String`
 - `writeUtf8(text: String)`
 - `appendUtf8(text: String)`
 - `metadata(): Map` — keys: `isFile`, `isDirectory`, `size`, `createdAtMillis`, `modifiedAtMillis`, `isSymlink`
 - `mkdirs(mustCreate: Bool = false)`
 - `move(to: Path|String, overwrite: Bool = false)`
 - `delete(mustExist: Bool = false, recursively: Bool = false)`
 - `copy(to: Path|String, overwrite: Bool = false)`
 - `glob(pattern: String): List<Path>` — supports `**`, `*`, `?` (POSIX-style)
 Streaming readers for big files:
 - `readChunks(size: Int = 65536): Iterator<Buffer>` — iterate fixed-size byte chunks
 - `readUtf8Chunks(size: Int = 65536): Iterator<String>` — iterate text chunks by character count
 - `lines(): Iterator<String>` — line iterator built on `readUtf8Chunks`
 Notes:
 - Iterators implement Lyng iterator protocol. If you break early from a loop, the runtime will attempt to call `cancelIteration()` when available.
 - Current implementations chunk in memory. The public API is stable; internals will evolve to true streaming on all platforms.
 - Attribute accessors (`isFile`, `isDirectory`, `size`, `createdAt*`, `modifiedAt*`) cache a metadata snapshot inside the `Path` instance to avoid repeated filesystem calls during a sequence of queries. `metadata()` remains available for bulk access.
 ---
 #### Access policy (security)
 Access control is enforced by `FsAccessPolicy`. You pass a policy at installation time. The module wraps the filesystem with a secured decorator that consults the policy for each primitive operation.
 Main types:
 - `FsAccessPolicy` — your policy implementation
 - `PermitAllAccessPolicy` — allows all operations (default for testing)
 - `AccessOp` (sealed) — operations the policy can decide on:
  - `ListDir(path)`
  - `CreateFile(path)`
  - `OpenRead(path)`
  - `OpenWrite(path)`
  - `OpenAppend(path)`
  - `Delete(path)`
  - `Rename(from, to)`
  - `UpdateAttributes(path)` — defaults to write-level semantics
 Minimal denying policy example (imports omitted for brevity):
 ```kotlin
 val denyWrites = object : FsAccessPolicy {
    override suspend fun check(op: AccessOp, ctx: AccessContext): AccessDecision = when (op) {
        is AccessOp.OpenRead, is AccessOp.ListDir -> AccessDecision(Decision.Allow)
        else -> AccessDecision(Decision.Deny, reason = "read-only policy")
    }
 }
 createFs(denyWrites, scope)
 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)`).
 ---
 #### Errors and exceptions
 Policy denials are surfaced as Lyng runtime errors, not raw Kotlin exceptions:
 - Internally, a denial throws `AccessDeniedException`. The module maps it to `ObjIllegalOperationException` wrapped into an `ExecutionError` visible to scripts.
 Examples (Lyng):
 ```lyng
 import lyng.io.fs
 val p = Path("/protected/file.txt")
 try {
    p.writeUtf8("x")
    fail("expected error")
 } catch (e) {
    // e is an ExecutionError; message contains the policy reason
 }
 ```
 Other I/O failures (e.g., not found, not a directory) are also raised as Lyng errors (`ObjIllegalStateException`, `ObjIllegalArgumentException`, etc.) depending on context.
 ---
 #### Platform notes
 - JVM/Android/Native: synchronous Okio calls are executed on `Dispatchers.IO` (JVM/Android) or `Dispatchers.Default` (Native) to avoid blocking the main thread.
 - NodeJS: currently uses Okio’s Node backend. For heavy I/O, a native `fs/promises` backend is planned to fully avoid event-loop blocking.
 - Browser/Wasm: uses an in-memory filesystem for now. Persistent backends (IndexedDB or File System Access API) are planned.
 ---
 #### Roadmap
 - Native NodeJS backend using `fs/promises`
 - Browser persistent storage (IndexedDB)
 - Streaming readers/writers over real OS streams
 - Attribute setters and richer metadata
 If you have specific needs (e.g., sandboxing, virtual roots), implement a custom `FsAccessPolicy` or ask us to add a helper.
@@ -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
@@ -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
@@ -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`.
@@ -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
@@ -0,0 +1,239 @@
 # 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
 Requirements:
 - JDK 17+ (for Gradle and the JVM distribution)
 - GNU zip utilities (for packaging the native executable)
 - upx tool (executable in-place compression)
 The repository provides convenience scripts in `bin/` for local builds and installation into `~/bin`.
 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`)
 1) Build the native binary:
 ```
 ./gradlew :lyng:linkReleaseExecutableLinuxX64
 ```
 2) Install and package locally:
 ```
 bin/local_release
 ```
 What this does:
 - Copies the built executable to `~/bin/lyng` for easy use in your shell.
 - Produces `distributables/lyng-linuxX64.zip` containing the `lyng` executable.
 ### Option B: JVM distribution (`jlyng` launcher)
 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: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
 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
 - Run a script by file name and pass arguments to `ARGV`:
 ```
 lyng path/to/script.lyng arg1 arg2
 ```
 - Run a script whose name starts with `-` using `--` to stop option parsing:
 ```
 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
 ```
 - Print version/help:
 ```
 lyng --version
 lyng --help
 ```
 ### 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:
    #!/usr/bin/env lyng
    println("Hello, world!")
 ### Formatting source: `fmt` subcommand
 Format Lyng files with the built-in formatter.
 Basic usage:
 ```
 lyng fmt [OPTIONS] FILE...
 ```
 Options:
 - `--check` — Check-only mode. Prints file paths that would change and exits with code 2 if any changes are needed, 0 otherwise.
 - `-i, --in-place` — Write formatted content back to the source files (off by default).
 - `--spacing` — Apply spacing normalization.
 - `--wrap`, `--wrapping` — Enable line wrapping.
 Semantics and exit codes:
 - Default behavior is to write formatted content to stdout. When multiple files are provided, the output is separated with `--- <path> ---` headers.
 - `--check` and `--in-place` are mutually exclusive; using both results in an error and exit code 1.
 - `--check` exits with 2 if any file would change, with 0 otherwise.
 - Other errors (e.g., I/O issues) result in a non-zero exit code.
 Examples:
 ```
 # Print formatted content to stdout
 lyng fmt src/file.lyng
 # Format multiple files to stdout with headers
 lyng fmt src/a.lyng src/b.lyng
 # Check mode: list files that would change; exit 2 if changes are needed
 lyng fmt --check src/**/*.lyng
 # In-place formatting
 lyng fmt -i src/**/*.lyng
 # Enable spacing normalization and wrapping
 lyng fmt --spacing --wrap src/file.lyng
 ```
 ## 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`.
 - The interpreter recognizes shebang lines (`#!`) at the beginning of a script file and ignores them at runtime, so you can make Lyng scripts directly executable on Unix-like systems.
@@ -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.
@@ -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
@@ -9,17 +9,37 @@ Same as in C++.
 | **Highest**<br>0 | power, not, calls, indexing, dot,... |
 |        1         | `%` `*` `/`                          |
 |        2         | `+` `-`                              |
-|        3         | bit shifts (NI)                      |
+|        3         | bit shifts `<<` `>>`                 |
 |        4         | `<=>` (1)                            |
 |        5         | `<=` `>=` `<` `>`                    |
 |        6         | `==` `!=`                            |
-|        7         | bitwise and `&` (NI)                 |
+|        7         | bitwise and `&`                      |
-|        9         | bitwise or `\|` (NI)                 |
+|        8         | bitwise xor `^`                      |
 |        9         | bitwise or `\|`                      |
 |        10        | `&&`                                 |
 |  11<br/>lowest   | `\|\|`                               |
-(NI)
+Bitwise operators
-: not yet implemented.
+: available only for `Int` values. For mixed `Int`/`Real` numeric expressions, bitwise operators are not defined.
 Bitwise NOT `~x`
 : unary operator that inverts all bits of a 64‑bit signed integer (`Int`). It follows two's‑complement rules, so
  `~x` is numerically equal to `-(x + 1)`. Examples: `~0 == -1`, `~1 == -2`, `~(-1) == 0`.
 Examples:
 ```
 5 & 3    // -> 1
 5 | 3    // -> 7
 5 ^ 3    // -> 6
 ~0       // -> -1
 1 << 3   // -> 8
 8 >> 3   // -> 1
 ```
 Notes:
 - Shifts operate on 64-bit signed integers (`Int` is 64-bit). Right shift `>>` is arithmetic (sign-propagating).
 - Shift count is masked to the range 0..63, similar to the JVM/Kotlin behavior (e.g., `1 << 65` equals `1 << 1`).
 (1)
 : Shuttle operator: `a <=> b` returns 0 if a == b, negative Int if a < b and positive Int otherwise. It is necessary to
@@ -40,8 +60,13 @@ but:
 ## Round and range
-The following functions return its argument if it is `Int`,
+The following functions return the argument unchanged if it is `Int`.
-or transformed `Real` otherwise.
+
 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                                            |
 |----------------|--------------------------------------------------------|
@@ -52,6 +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                                                 |
@@ -71,7 +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:
@@ -82,6 +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
@@ -0,0 +1,329 @@
 # Multithreading/parallel execution
 [//]: # (topMenu)
 Lyng is built to me multithreaded where possible (e.g. all targets byt JS and wasmJS as for now)
 and cooperatively parallel (coroutine based) everywhere.
 In Lyng, every function, every lambda are _coroutines_. It means, you can have as many of these as you want without risking running out of memory on threads stack, or get too many threads.
 Depending on the platform, these coroutines may be executed on different CPU and cores, too, truly in parallel. Where not, like Javascript browser, they are still executed cooperatively. You should not care about the platform capabilities, just call `launch`:
    // track coroutine call: 
    var xIsCalled = false
    // launch coroutine in parallel
    val x = launch { 
        // wait 10ms to let main code to be executed
        delay(10)
        // now set the flag
        xIsCalled = true
        // and return something useful:
        "ok"
    }
    // corouine is launhed, but not yet executed
    // due to delay call:
    assert(!xIsCalled)
    // now we wait for it to be executed:
    assertEquals( x.await(), "ok")
    // now glag should be set:
    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, 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:
    var counter = 0
    (1..50).map { 
        launch {
            // slow increment:
            val c = counter
            delay(100)
            counter = c + 1
        }
    }.forEach { (it as Deferred).await() }
    assert(counter < 50) { "counter is "+counter }
    >>> void
 The obviously wrong result is not 4, as all coroutines capture the counter value, which is 1, then sleep for 5ms, then save 1 + 1 as result. May some coroutines will pass, so it will be 1 or 2, most likely.
 Using [Mutex] makes it all working:
    var counter = 0
    val mutex = Mutex()
    (1..4).map { 
        launch {
            // slow increment:
            mutex.withLock {
                val c = counter ?: 0
                counter = c + 1
            }
        }
    }.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.
 ## Completable deferred
 Sometimes it is convenient to manually set completion status of some deferred result. This is when [CompletableDeferred] is used:
    // this variable will be completed later:
    val done = CompletableDeferred()
    // complete it ater delay
    launch { 
        delay(10)
        // complete it setting the result:
        done.complete("ok")
    }
    // now it is still not completed: coroutine is delayed
    // (ot not started on sinthe-threaded platforms):
    assert(!done.isCompleted)
    assert(done.isActive)
    // then we can just await it as any other deferred:
    assertEquals( done.await(), "ok")
    // and as any other deferred it is now complete:
    assert(done.isCompleted)
 ## True parallelism
 Cooperative, coroutine-based parallelism is automatically available on all platforms. Depending on the platform, though, the coroutines could be dispatched also in different threads; where there are multiple cores and/or CPU available, it means the coroutines could be exuted truly in parallel, unless [Mutex] is used:
 | platofrm   | multithreaded |
 |------------|---------------|
 | JVM        | yes           |
 | Android    | yes           |
 | Javascript | NO            |
 | wasmJS     | NO            |
 | IOS        | yes           |
 | MacOSX     | yes           |
 | Linux      | yes           |
 | Windows    | yes           |
 So it is important to always use [Mutex] where concurrent execution could be a problem (so called Race Conditions, or RC).
 ## Yield
 When the coroutine is executed, on the single-threaded environment all other coroutines are suspended until active one will wait for something. Sometimes, it is undesirable; the coroutine may perform long calculations or some other CPU consuming task. The solution is to call `yield()` periodically. Unlike `delay()`, yield does not pauses the coroutine for some specified time, but it just makes all other coroutines to be executed. In other word, yield interrupts current coroutines and out it to the end of the dispatcher list of active coroutines. It is especially important on Javascript and wasmJS targets as otherwise UI thread could be blocked. 
 Usage example:
    fun someLongTask() { // ...
        do {
            // execute step
            if( done ) break
            yield()
        } while(true)
    }
 # Data exchange for coroutines
 ## Flow
 Flow is an async cold sequence; it is named after kotlin's Flow as it resembles it closely. The cold means the flow is only evaluated when iterated (collected, in Kotlin terms), before it is inactive. Sequence means that it is potentially unlimited, as in our example of glorious Fibonacci number generator:
    // Fibonacch numbers flow! 
    val f = flow {
        println("Starting generator")
        var n1 = 0
        var n2 = 1
        emit(n1)
        emit(n2)
        while(true) {
            val n = n1 + n2
            emit(n)
            n1 = n2
            n2 = n
        }
    }
    val correctFibs = [0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1597, 2584, 4181, 6765]
    println("Generation starts")
    assertEquals( correctFibs, f.take(correctFibs.size))
    >>> Generation starts
    >>> Starting generator
    >>> void
 Great: the generator is not executed until collected bu the `f.take()` call, which picks specified number of elements from the flow, can cancel it.
 Important difference from the channels or like, every time you collect the flow, you collect it anew:
    var isStarted = false
    val f = flow {
            emit("start")
            isStarted = true
            (1..4).forEach { emit(it) }
    }
    // flow is not yet started, e.g. not got execited, 
    // that is called 'cold':
    assertEquals( false, isStarted )
    // let's collect flow:
    val result = []
    for( x in f ) result += x
    println(result)
    assertEquals( true, isStarted)
    // let's collect it once again, it should be the same:
    println(f.toList())
    // and again:
    assertEquals( result, f.toList() )
    >>> [start,1,2,3,4]
    >>> [start,1,2,3,4]
    >>> void
 Notice that flow's lambda is not called until actual collection is started. Cold flows are
 better in terms of resource consumption.
 Flows allow easy transforming of any [Iterable]. See how the standard Lyng library functions use it:
    fun Iterable.filter(predicate) {
        val list = this
        flow {
            for( item in list ) {
                if( predicate(item) ) {
                    emit(item)
                }
            }
        }
    }
 ## 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
 ## Scope frame pooling (JVM)
 Lyng includes an optional optimization for function/method calls on JVM: scope frame pooling, toggled by the runtime flag `PerfFlags.SCOPE_POOL`.
 - Default: `SCOPE_POOL` is OFF on JVM.
 - Rationale: the current `ScopePool` implementation is not thread‑safe. Lyng targets multi‑threaded execution on most platforms, therefore we keep pooling disabled by default until a thread‑safe design is introduced.
 - When safe to enable: single‑threaded runs (e.g., micro‑benchmarks or scripts executed on a single worker) where no scopes are shared across threads.
 - How to toggle at runtime (Kotlin/JVM tests):
  - `PerfFlags.SCOPE_POOL = true` to enable.
  - `PerfFlags.SCOPE_POOL = false` to disable.
 - Expected effect (from our JVM micro‑benchmarks): in deep call loops, enabling pooling reduced total time by about 1.38× in a dedicated pooling benchmark; mileage may vary depending on workload.
 Future work: introduce thread‑safe pooling (e.g., per‑thread pools or confinement strategies) before considering enabling it by default in multi‑threaded environments.
 ### Closures inside coroutine helpers (launch/flow)
 Closures executed by `launch { ... }` and `flow { ... }` use **compile‑time resolution** just like any other Lyng code:
 - **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:
 - 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: compile-time resolution](scopes_and_closures.md)
@@ -0,0 +1,623 @@
 This document explains how to enable and measure the performance optimizations added to the Lyng interpreter. The focus is JVM‑first with safe, flag‑guarded rollouts and quick A/B testing. Other targets (JS/Wasm/Native) keep conservative defaults until validated.
 [//]: # (excludeFromIndex)
 ## Overview
 Optimizations are controlled by runtime‑mutable flags in `net.sergeych.lyng.PerfFlags`, initialized from platform‑specific static defaults `net.sergeych.lyng.PerfDefaults` (KMP `expect/actual`).
 - JVM/Android defaults are aggressive (e.g. `RVAL_FASTPATH=true`).
 - Non‑JVM defaults are conservative (e.g. `RVAL_FASTPATH=false`).
 All flags are `var` and can be flipped at runtime (e.g., from tests or host apps) for A/B comparisons.
 ### Workload presets (JVM‑first)
 To simplify switching between recommended flag sets for different workloads, use `net.sergeych.lyng.PerfProfiles`:
 ```
 val snap = PerfProfiles.apply(PerfProfiles.Preset.BENCH)  // or BASELINE / BOOKS
 // ... run workload ...
 PerfProfiles.restore(snap)  // restore previous flags
 ```
 - BASELINE: restores platform defaults from `PerfDefaults` (good rollback point).
 - BENCH: expression‑heavy micro‑bench focus (aggressive R‑value and PIC optimizations on JVM).
 - BOOKS: documentation workloads (prefers simpler paths; disables some PIC/arg builder features shown neutral/negative for this load in A/B).
 ## Key flags
 - `LOCAL_SLOT_PIC` — Runtime cache in `LocalVarRef` to avoid repeated name→slot lookups per frame (ON JVM default).
 - `EMIT_FAST_LOCAL_REFS` — Compiler emits `FastLocalVarRef` for identifiers known to be locals/params (ON JVM default).
 - `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 (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.
 - `METHOD_PIC_SIZE_4` — Increases Method PIC size from 2 to 4 entries (JVM-first tuning; OFF by default).
 - `PIC_ADAPTIVE_2_TO_4` — Adaptive growth of Field/Method PICs from 2→4 entries per-site when miss rate >20% over ≥256 accesses (JVM-first; OFF by default).
 - `INDEX_PIC` — Enables polymorphic inline cache for indexing (e.g., `a[i]`) and related fast paths. Defaults to follow `FIELD_PIC` on init; can be toggled independently.
 - `INDEX_PIC_SIZE_4` — Increases Index PIC size from 2 to 4 entries (JVM-first tuning). Default: ON for JVM; OFF elsewhere by default.
 - `PIC_DEBUG_COUNTERS` — Enable lightweight hit/miss counters via `PerfStats` (OFF by default).
 - `PRIMITIVE_FASTOPS` — Fast paths for `(ObjInt, ObjInt)` arithmetic/comparisons and `(ObjBool, ObjBool)` logic (ON JVM default).
 - `RVAL_FASTPATH` — Bypass `ObjRecord` in pure expression evaluation via `ObjRef.evalValue` (ON JVM default, OFF elsewhere).
 See `src/commonMain/kotlin/net/sergeych/lyng/PerfFlags.kt` and `PerfDefaults.*.kt` for details and platform defaults.
 ## Where optimizations apply
 - Locals: `FastLocalVarRef`, `LocalVarRef` per‑frame cache (PIC).
 - Calls: small‑arity zero‑alloc paths (0–8 args; optionally 0–12 with `ARG_SMALL_ARITY_12`), pooled builder (JVM), and child frame pooling (optional).
 - Properties/methods: Field/Method PICs with receiver shape `(classId, layoutVersion)` and handle‑aware caches; configurable 2→4 entries under flags.
 - Expressions: R‑value fast paths in hot nodes (`UnaryOpRef`, `BinaryOpRef`, `ElvisRef`, logical ops, `RangeRef`, `IndexRef` read, `FieldRef` receiver eval, `ListLiteralRef` elements, `CallRef` callee, `MethodCallRef` receiver, assignment RHS).
 - Primitives: Direct boolean/int ops where safe.
 ### Compiler constant folding (conservative)
 - The compiler folds a safe subset of literal‑only expressions at compile time to reduce runtime work:
  - Integer arithmetic: `+ - * / %` (division/modulo only when divisor ≠ 0).
  - Bitwise integer ops: `& ^ | << >>`.
  - Comparisons and equality for ints/strings/chars: `== != < <= > >=`.
  - Boolean logic for literal booleans: `|| &&` and unary `!`.
  - String concatenation of literal strings: `"a" + "b"`.
 - Non‑literal operands or side‑effecting constructs are not folded.
 - Semantics remain unchanged; tests verify parity.
 ## Running JVM micro‑benchmarks
 Each benchmark prints timings with `[DEBUG_LOG]` and includes correctness assertions to prevent dead‑code elimination.
 Run individual tests to avoid multiplatform matrices:
 ```
 ./gradlew :lynglib:jvmTest --tests LocalVarBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests CallBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests CallMixedArityBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests CallSplatBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests PicBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests PicInvalidationJvmTest
 ./gradlew :lynglib:jvmTest --tests PicAdaptiveABTest
 ./gradlew :lynglib:jvmTest --tests ArithmeticBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests ExpressionBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests IndexPicABTest
 ./gradlew :lynglib:jvmTest --tests IndexWritePathABTest
 ./gradlew :lynglib:jvmTest --tests CallPoolingBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests MethodPoolingBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests MixedBenchmarkTest
 ./gradlew :lynglib:jvmTest --tests DeepPoolingStressJvmTest
 ```
 Typical output (example):
 ```
 [DEBUG_LOG] [BENCH] mixed-arity x200000 [ARG_BUILDER=ON]: 85.7 ms
 ```
 Lower time is better. Run the same bench with a flag OFF vs ON to compare.
 ### Optional JFR allocation profiling (JVM)
 When running end‑to‑end “book” workloads or heavier benches, you can enable JFR to capture allocation and GC details:
 ```
 ./gradlew :lynglib:jvmTest --tests BookAllocationProfileTest -Dlyng.jfr=true \
  -Dlyng.profile.warmup=1 -Dlyng.profile.repeats=3 -Dlyng.profile.shuffle=true
 ```
 - Dumps are saved to `lynglib/build/jfr_*.jfr` if the JVM supports Flight Recorder.
 - The test also records GC counts/time and median time/heap deltas to `lynglib/build/book_alloc_profile.txt`.
 ## Toggling flags in tests
 Flags are mutable at runtime, e.g.:
 ```kotlin
 runTest {
    PerfFlags.ARG_BUILDER = false
    val r1 = (EvalSession(Scope()).eval(script) as ObjInt).value
    PerfFlags.ARG_BUILDER = true
    val r2 = (EvalSession(Scope()).eval(script) as ObjInt).value
 }
 ```
 Reset flags at the end of a test to avoid impacting other tests.
 ## PIC diagnostics (optional)
 Enable counters:
 ```kotlin
 PerfFlags.PIC_DEBUG_COUNTERS = true
 PerfStats.resetAll()
 ```
 Available counters in `PerfStats`:
 - Field PIC: `fieldPicHit`, `fieldPicMiss`, `fieldPicSetHit`, `fieldPicSetMiss`
 - Method PIC: `methodPicHit`, `methodPicMiss`
 - Index PIC: `indexPicHit`, `indexPicMiss`
 - Locals: `localVarPicHit`, `localVarPicMiss`, `fastLocalHit`, `fastLocalMiss`
 - Primitive ops: `primitiveFastOpsHit`
 Print a summary at the end of a bench/test as needed. Remember to turn counters OFF after the test.
 ## A/B scenarios and guidance (JVM)
 ### Adaptive PIC (fields/methods)
 - Flags: `FIELD_PIC=true`, `METHOD_PIC=true`, `FIELD_PIC_SIZE_4=false`, `METHOD_PIC_SIZE_4=false`, toggle `PIC_ADAPTIVE_2_TO_4` OFF vs ON.
 - Benchmarks: `PicBenchmarkTest`, `MixedBenchmarkTest`, `PicAdaptiveABTest`.
 - Expect wins at sites with >2 receiver shapes; counters should show fewer misses with adaptivity ON.
 ### Index PIC and size
 - Flags: toggle `INDEX_PIC` OFF vs ON; then `INDEX_PIC_SIZE_4` OFF vs ON.
 - Benchmarks: `ExpressionBenchmarkTest` (list indexing) and `IndexPicABTest` (string/map indexing).
 - Expect wins when the same index shape recurs; counters should show higher `indexPicHit`.
 ### Index WRITE paths (Map and List)
 - Flags: toggle `INDEX_PIC` OFF vs ON; then `INDEX_PIC_SIZE_4` OFF vs ON.
 - Benchmark: `IndexWritePathABTest` (Map[String] put, List[Int] set) — writes results to `lynglib/build/index_write_ab_results.txt`.
 - Direct fast‑paths are used on R‑value paths where safe and semantics‑preserving (e.g., optional‑chaining no‑ops on null receivers; bounds exceptions unchanged).
 ## Guidance per flag (JVM)
 - Keep `RVAL_FASTPATH = true` unless debugging a suspected expression‑semantics issue.
 - Use `SCOPE_POOL = true` only for benchmarks or once pooling passes the deep stress tests and broader validation; currently OFF by default.
 - `FIELD_PIC` and `METHOD_PIC` should remain ON; they are validated with invalidation tests.
 - Consider enabling `FIELD_PIC_SIZE_4`/`METHOD_PIC_SIZE_4` for sites with 3–4 receiver shapes; measure first.
 - `PIC_ADAPTIVE_2_TO_4` is useful on polymorphic sites and may outperform fixed size 2 on mixed-shape workloads. Validate with `PicAdaptiveABTest`.
 - `INDEX_PIC` is generally beneficial on JVM; leave ON when measuring index‑heavy workloads.
 - `INDEX_PIC_SIZE_4` is ON by default on JVM as A/B showed consistent wins on String[Int] and Map[String] workloads. You can disable it by setting `PerfFlags.INDEX_PIC_SIZE_4 = false` if needed.
 - `ARG_BUILDER` should remain ON; switch OFF only to get a baseline.
 - `ARG_SMALL_ARITY_12` is experimental and OFF by default. Enable it only if your workload frequently calls functions with 9–12 arguments and A/B shows consistent wins.
 ### Workload‑specific recommendations (JVM)
 - “Books”/documentation loads (BookTest): prefer simpler paths; in A/B these often benefit from the BOOKS preset (e.g., `ARG_BUILDER=false`, `SCOPE_POOL=false`, `INDEX_PIC=false`). Use `PerfProfiles.apply(PerfProfiles.Preset.BOOKS)` before the run and `restore(...)` after.
 - Expression‑heavy benches: use the BENCH preset (PICs and R‑value fast‑paths enabled, `INDEX_PIC_SIZE_4=true`).
 - Always verify with local A/B on your environment; rollback is a flag flip or applying BASELINE.
 ## Notes on correctness & safety
 - Optional chaining semantics are preserved across fast paths.
 - Visibility/mutability checks are enforced even on PIC fast‑paths.
 - `frameId` is regenerated on each pooled frame borrow; stress tests verify no leakage under deep nesting/recursion.
 ## Cross‑platform
 - Non‑JVM defaults keep `RVAL_FASTPATH=false` for now; other low‑risk flags may be ON.
 - Once JVM path is fully validated and measured, add lightweight benches for JS/Wasm/Native and enable flags incrementally.
 ## Range fast iteration (experimental)
 - Flag: `RANGE_FAST_ITER` (default OFF).
 - When enabled and applicable, simple ascending integer ranges (`0..n`, `0..<n`) use a specialized non‑allocating iterator (`ObjFastIntRangeIterator`).
 - Benchmark: `RangeIterationBenchmarkTest` records OFF/ON timings for inclusive, exclusive, reversed, negative, and empty ranges. Semantics are preserved; non‑int or complex ranges fall back to the generic iterator.
 ## Troubleshooting
 - If a benchmark shows regressions, flip related flags OFF to isolate the source (e.g., `ARG_BUILDER`, `RVAL_FASTPATH`, `FIELD_PIC`, `METHOD_PIC`).
 - Use `PIC_DEBUG_COUNTERS` to observe inline cache effectiveness.
 - Ensure tests do not accidentally keep flags ON for subsequent tests; reset after each test.
 ## JVM micro-benchmark results (3× medians; OFF → ON)
 Date: 2025-11-10 23:04 (local)
 | Flag               | Benchmark/Test                              | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |--------------------|----------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | ARG_BUILDER        | CallMixedArityBenchmarkTest                   |           788.02 |          668.79 |  1.18×  | Clear win on mixed arity |
 | ARG_BUILDER        | CallBenchmarkTest (simple calls)              |           423.87 |          425.47 |  1.00×  | Neutral on repeated simple calls |
 | FIELD_PIC          | PicBenchmarkTest::benchmarkFieldGetSetPic     |           113.575 |          106.017 |  1.07×  | Small but consistent win |
 | METHOD_PIC         | PicBenchmarkTest::benchmarkMethodPic          |           251.068 |          149.439 |  1.68×  | Large consistent win |
 | RVAL_FASTPATH      | ExpressionBenchmarkTest                       |           514.491 |          426.800 |  1.21×  | Consistent win in expression chains |
 | PRIMITIVE_FASTOPS  | ArithmeticBenchmarkTest (int-sum)             |           243.420 |          128.146 |  1.90×  | Big win for integer addition |
 | PRIMITIVE_FASTOPS  | ArithmeticBenchmarkTest (int-cmp)             |           210.385 |          168.534 |  1.25×  | Moderate win for comparisons |
 | SCOPE_POOL         | CallPoolingBenchmarkTest                      |           505.778 |          366.737 |  1.38×  | Single-threaded bench; per-thread ThreadLocal pool; default ON on JVM |
 Notes:
 - All results obtained from `[DEBUG_LOG] [BENCH]` outputs with three repeated Gradle test invocations per configuration; medians reported.
 - JVM defaults (current): `ARG_BUILDER=true`, `PRIMITIVE_FASTOPS=true`, `RVAL_FASTPATH=true`, `FIELD_PIC=true`, `METHOD_PIC=true`, `SCOPE_POOL=true` (per‑thread ThreadLocal pool), `REGEX_CACHE=true`.
 ## Concurrency (multi‑core) pooling results (3× medians; OFF → ON)
 Date: 2025-11-10 22:56 (local)
 | Flag       | Benchmark/Test                      | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |------------|--------------------------------------|-----------------:|----------------:|:-------:|-------|
 | SCOPE_POOL | ConcurrencyCallBenchmarkTest (JVM)   |           521.102 |          201.374 |  2.59×  | Multithreaded workload on `Dispatchers.Default` with per‑thread ThreadLocal pool; workers=8, iters=15000/worker. |
 Methodology:
 - The test toggles `PerfFlags.SCOPE_POOL` within a single run and executes the same script across N worker coroutines scheduled on `Dispatchers.Default`.
 - We executed the test three times via Gradle and computed medians from the printed `[DEBUG_LOG]` timings:
  - OFF runs (ms): 532.442 | 521.102 | 474.386 → median 521.102
  - ON runs (ms):  218.683 | 201.374 | 198.737 → median 201.374
 - Speedup = OFF/ON.
 Reproduce:
 ```
 ./gradlew :lynglib:jvmTest --tests "ConcurrencyCallBenchmarkTest" --rerun-tasks
 ```
 ## Next optimization steps (JVM)
 Date: 2025-11-10 23:04 (local)
 - PICs
  - Widen METHOD_PIC to 3–4 entries with tiny LRU; keep invalidation on layout change; re-run `PicInvalidationJvmTest`.
  - Micro fast-path for FIELD_PIC read-then-write pairs (`x = x + 1`) to reuse the resolved slot within one step.
 - Locals and slots
  - Pre-size `Scope` slot structures when compiler knows local/param counts; audit `EMIT_FAST_LOCAL_REFS` coverage.
  - Re-run `LocalVarBenchmarkTest` to quantify gains.
 - RVAL_FASTPATH coverage
  - Cover primitive `ObjList` index reads, pure receivers in `FieldRef`, and assignment RHS where safe; add micro-benches to `ExpressionBenchmarkTest`.
 - Collections and ranges
  - Specialize `(Int..Int)` loops into tight counted loops (no intermediary objects).
  - Add primitive-specialized `ObjList` ops (`map`, `filter`, `sum`, `contains`) under `PRIMITIVE_FASTOPS`.
 - Regex and strings
  - Cache compiled regex for string literals at compile time; add a tiny LRU for dynamic patterns behind `REGEX_CACHE`.
  - Add `RegexBenchmarkTest` for repeated matches.
 - JIT friendliness (Kotlin/JVM)
  - Inline tiny helpers in hot paths, prefer arrays for internal buffers, finalize hot data structures where safe.
 Validation matrix
 - Always re-run: `CallBenchmarkTest`, `CallMixedArityBenchmarkTest`, `PicBenchmarkTest`, `ExpressionBenchmarkTest`, `ArithmeticBenchmarkTest`, `CallPoolingBenchmarkTest`, `DeepPoolingStressJvmTest`, `ConcurrencyCallBenchmarkTest` (3× medians when comparing).
 - Keep full `:lynglib:jvmTest` green after each change.
 ## PIC update (4‑way METHOD_PIC) — JVM (3× medians; OFF → ON)
 Date: 2025-11-11 00:16 (local)
 | Flag      | Benchmark/Test                               | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |-----------|-----------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | FIELD_PIC | PicBenchmarkTest::benchmarkFieldGetSetPic      |          207.578 |         106.481 |  1.95×  | Read→write loop; micro fast‑path groundwork present |
 | METHOD_PIC| PicBenchmarkTest::benchmarkMethodPic           |          273.478 |         182.226 |  1.50×  | 4‑way PIC with move‑to‑front (was 2‑way before) |
 Medians computed from three Gradle runs in this session; see `[DEBUG_LOG] [BENCH]` lines in test output.
 ## Locals/slots capacity (pre‑sizing hints) — JVM (3× medians; OFF → ON)
 Date: 2025-11-11 13:19 (local)
 | Optimization            | Benchmark/Test              | OFF config                         | ON config                          | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |-------------------------|-----------------------------|------------------------------------|------------------------------------|-----------------:|----------------:|:-------:|-------|
 | Locals pre‑sizing + PIC | LocalVarBenchmarkTest       | LOCAL_SLOT_PIC=OFF, FAST_LOCAL=OFF | LOCAL_SLOT_PIC=ON, FAST_LOCAL=ON   |          472.129 |         370.871 |  1.27×  | Compiler hint `params+4`; slot pre‑size; semantics unchanged |
 Methodology:
 - Each configuration executed three times via `:lynglib:jvmTest --tests "…" --rerun-tasks`; medians reported.
 - Locals improvement stacks with per‑thread `SCOPE_POOL` and ARG fast paths.
 ## RVAL fast paths update — JVM (IndexRef and FieldRef) [3× medians; OFF → ON]
 Date: 2025-11-11 13:19 (local)
 New micro-benchmarks have been added to quantify the latest `RVAL_FASTPATH` extensions:
 - Primitive `ObjList` index-read fast path in `IndexRef`.
 - Conservative “pure receiver” evaluation in `FieldRef` (monomorphic, immutable receiver), preserving visibility/mutability checks and optional chaining semantics.
 Benchmarks to run (each 3× OFF → ON):
 - `ExpressionBenchmarkTest::benchmarkListIndexReads`
 - `ExpressionBenchmarkTest::benchmarkFieldReadPureReceiver`
 Reproduce (3× each; collect `[DEBUG_LOG] [BENCH]` lines and compute medians):
 ```
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkListIndexReads" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkListIndexReads" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkListIndexReads" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkFieldReadPureReceiver" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkFieldReadPureReceiver" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkFieldReadPureReceiver" --rerun-tasks
 ```
 Once collected, add medians and speedups to the table below:
 | Flag          | Benchmark/Test                                  | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |---------------|---------------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | RVAL_FASTPATH | ExpressionBenchmarkTest::benchmarkListIndexReads  |           305.243 |          230.942 |  1.32×  | Fast path in `IndexRef` for `ObjList` + `ObjInt` index |
 | RVAL_FASTPATH | ExpressionBenchmarkTest::benchmarkFieldReadPureReceiver |           266.222 |          190.720 |  1.40×  | Pure-receiver evaluation in `FieldRef` (monomorphic, immutable) |
 Notes:
 - Both benches toggle `PerfFlags.RVAL_FASTPATH` within a single run to produce OFF and ON timings under identical conditions.
 - Correctness assertions ensure the loops are not optimized away.
 - All semantics (visibility/mutability checks, optional chaining) remain intact; fast paths only skip interim `ObjRecord` traffic when safe.
 ## ARG_BUILDER — splat fast‑path (3× medians; OFF → ON)
 Date: 2025-11-11 13:12 (local)
 Environment: Gradle 8.7; JVM (JDK as configured by toolchain); single‑threaded test execution; stdout enabled.
 | Flag        | Benchmark/Test                    | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |-------------|-----------------------------------|-----------------:|----------------:|:-------:|-------|
 | ARG_BUILDER | CallSplatBenchmarkTest (splat)    |          613.689 |         463.593 |  1.32×  | Single‑splat fast‑path returns underlying list directly; avoids intermediate copies |
 Inputs (3×):
 - OFF runs (ms): 613.689 | 629.604 | 612.361 → median 613.689
 - ON runs (ms):  453.752 | 463.593 | 468.844 → median 463.593
 Reproduce (3×):
 ```
 ./gradlew :lynglib:jvmTest --tests "CallSplatBenchmarkTest" --rerun-tasks
 ```
 ## Phase A consolidation (JVM) — 3× medians updated
 Date: 2025-11-11 13:48 (local)
 Environment:
 - JDK: OpenJDK 20.0.2.1 (Amazon Corretto 20.0.2.1+10-FR)
 - Gradle: 8.7
 - OS/Arch: macOS 14.8.1 (aarch64)
 ### ARG_BUILDER
 | Benchmark/Test                   | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |----------------------------------|-----------------:|----------------:|:-------:|-------|
 | CallMixedArityBenchmarkTest      |          866.681 |         717.439 |  1.21×  | Small-arity 0–8 fast path + builder; correctness preserved |
 | CallSplatBenchmarkTest (splat)   |          600.880 |         459.706 |  1.31×  | Single-splat fast path returns underlying list; avoids copies |
 Inputs (3×):
 - Mixed arity OFF: 874.088291 | 866.680959 | 858.577125 → median 866.680959
 - Mixed arity ON:  731.308625 | 706.440125 | 717.438542 → median 717.438542
 - Splat OFF: 600.268625 | 607.849416 | 600.879666 → median 600.879666
 - Splat ON:  459.706375 | 449.950166 | 461.815167 → median 459.706375
 ### RVAL_FASTPATH (new coverage)
 | Benchmark/Test                                   | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |--------------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | ExpressionBenchmarkTest::benchmarkListIndexReads |          299.366 |         218.812 |  1.37×  | IndexRef fast path for ObjList + ObjInt |
 | ExpressionBenchmarkTest::benchmarkFieldReadPureReceiver |          268.315 |         186.032 |  1.44×  | Pure-receiver evaluation in FieldRef (monomorphic, immutable) |
 Inputs (3×):
 - ListIndex OFF: 291.344 | 310.717167 | 299.365709 → median 299.365709
 - ListIndex ON:  217.795375 | 221.504166 | 218.812042 → median 218.812042
 - FieldRead OFF: 267.2775 | 274.355208 | 268.315125 → median 268.315125
 - FieldRead ON:  189.599333 | 186.031791 | 182.069167 → median 186.031791
 ### Locals/slots capacity (precise hints)
 | Benchmark/Test             | OFF config                         | ON config                          | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |---------------------------|------------------------------------|------------------------------------|-----------------:|----------------:|:-------:|-------|
 | LocalVarBenchmarkTest     | LOCAL_SLOT_PIC=OFF, FAST_LOCAL=OFF | LOCAL_SLOT_PIC=ON, FAST_LOCAL=ON   |          446.018 |         347.964 |  1.28×  | Precise capacity hints + fast-locals coverage |
 Inputs (3×):
 - Locals OFF: 470.575041 | 441.89625 | 446.017833 → median 446.017833
 - Locals ON:  370.664208 | 345.615541 | 347.964291 → median 347.964291
 Methodology:
 - Each test executed three times via Gradle with stdout enabled; medians computed from `[DEBUG_LOG] [BENCH]` lines.
 - Full JVM tests and stress benches remain green in this cycle.
 ## Phase B — List ops specialization (PRIMITIVE_FASTOPS) — 3× medians (OFF → ON)
 Date: 2025-11-11 13:48 (local)
 Environment:
 - JDK: OpenJDK 20.0.2.1 (Amazon Corretto 20.0.2.1+10-FR)
 - Gradle: 8.7
 - OS/Arch: macOS 14.8.1 (aarch64)
 | Optimization        | Benchmark/Test                          | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |---------------------|------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | PRIMITIVE_FASTOPS   | ListOpsBenchmarkTest::benchmarkSumInts   |          324.805 |         144.908 |  2.24×  | ObjList.sum fast path for int lists; generic fallback preserved |
 | PRIMITIVE_FASTOPS   | ListOpsBenchmarkTest::benchmarkContainsInts |          440.414 |         415.476 |  1.06×  | ObjList.contains fast path when searching ObjInt in int list |
 Inputs (3×):
 - list-sum OFF: 332.863417 | 323.491625 | 324.804083 → median 324.804083
 - list-sum ON:  144.907833 | 148.870792 | 126.418542 → median 144.907833
 - list-contains OFF: 440.413709 | 440.368333 | 441.4365 → median 440.413709
 - list-contains ON:  416.465292 | 412.283291 | 415.475833 → median 415.475833
 Methodology:
 - Each test executed three times via Gradle; medians computed from `[DEBUG_LOG] [BENCH]` lines.
 - Changes are fully guarded by `PerfFlags.PRIMITIVE_FASTOPS`; semantics preserved (null on empty sum; generic fallback on mixed types).
 ### Phase B — Ranges for-in lowering (PRIMITIVE_FASTOPS) — 3× medians (OFF → ON)
 Date: 2025-11-11 13:48 (local)
 Environment:
 - JDK: OpenJDK 20.0.2.1 (Amazon Corretto 20.0.2.1+10-FR)
 - Gradle: 8.7
 - OS/Arch: macOS 14.8.1 (aarch64)
 | Optimization        | Benchmark/Test                          | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |---------------------|------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | PRIMITIVE_FASTOPS   | RangeBenchmarkTest::benchmarkIntRangeForIn |         1705.299 |         788.974 |  2.16×  | Tight counted loop for (Int..Int) for-in; preserves semantics |
 Inputs (3×):
 - range-for-in OFF: 1705.298958 | 1684.357708 | 1735.880917 → median 1705.298958
 - range-for-in ON:  794.178458 | 778.741834 | 788.973625 → median 788.973625
 Methodology:
 - Each configuration executed three times via Gradle; medians computed from `[DEBUG_LOG] [BENCH]` lines.
 - Lowering is guarded by `PerfFlags.PRIMITIVE_FASTOPS` and applies only when the source is an `ObjRange` with int bounds; otherwise falls back to generic iteration.
 ## Phase B — Regex caching (REGEX_CACHE) — 3× medians (OFF → ON)
 Date: 2025-11-11 13:48 (local)
 Environment:
 - JDK: OpenJDK 20.0.2.1 (Amazon Corretto 20.0.2.1+10-FR)
 - Gradle: 8.7
 - OS/Arch: macOS 14.8.1 (aarch64)
 | Flag         | Benchmark/Test                                  | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |--------------|---------------------------------------------------|-----------------:|----------------:|:-------:|-------|
 | REGEX_CACHE  | RegexBenchmarkTest::benchmarkLiteralPatternMatches |          378.246 |         275.890 |  1.37×  | Caches compiled regex for identical literal pattern per iteration |
 | REGEX_CACHE  | RegexBenchmarkTest::benchmarkDynamicPatternMatches |          514.944 |         229.006 |  2.25×  | Two dynamic patterns alternate; cache size sufficient to retain both |
 Inputs (1× here; can extend to 3× on request):
 - regex-literal OFF: 378.245916; ON: 275.889541
 - regex-dynamic OFF: 514.944167; ON: 229.005834
 Methodology:
 - Each benchmark toggles `PerfFlags.REGEX_CACHE` inside a single test and prints `[DEBUG_LOG]` timings for OFF and ON runs under identical conditions. We recorded one set of OFF/ON timings here; we can extend to 3× medians if required for publication.
 - The cache is a tiny size-bounded map (64 entries) activated only when `PerfFlags.REGEX_CACHE` is true. Defaults remain OFF.
 ## JIT tweaks (Round 1) — quick gains snapshot (locals, ranges, list ops)
 Date: 2025-11-11 21:05 (local)
 Scope: fast confirmation of overall gain using current configuration; focused on locals, ranges, and list ops. Each test prints OFF → ON timings in a single run. We executed the benches via Gradle with stdout enabled and single test fork.
 Environment:
 - Gradle: 8.7 (stdout enabled, maxParallelForks=1)
 - JVM: as configured by toolchain for this project
 - OS/Arch: per developer machine (unchanged from prior sections)
 Reproduce:
 ```
 ./gradlew :lynglib:jvmTest --tests LocalVarBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests RangeBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ListOpsBenchmarkTest --rerun-tasks
 ```
 Results (representative runs; OFF → ON):
 - Local variables — LOCAL_SLOT_PIC + EMIT_FAST_LOCAL_REFS
  - Run 1: 468.407 ms → 367.277 ms (≈ 1.28×)
  - Run 2: 447.031 ms → 346.126 ms (≈ 1.29×)
 - Ranges for‑in — PRIMITIVE_FASTOPS (tight counted loop for (Int..Int))
  - 1731.780 ms → 799.023 ms (≈ 2.17×)
 - List ops — PRIMITIVE_FASTOPS
  - sum(int list): 318.943 ms → 148.571 ms (≈ 2.15×)
  - contains(int in int list): 440.013 ms → 412.450 ms (≈ 1.07×)
 Summary: All three areas improved with optimizations ON; no regressions observed in these runs. For publication‑grade stability, run each test 3× and report medians (see sections below for methodology and previous median tables).
 ## Additional tweaks — verification snapshot (Index write fast‑path, List literal pre‑size, Regex LRU)
 Date: 2025-11-11 21:31 (local)
 Scope: Implemented three semantics‑neutral optimizations and verified they are green across targeted and broader JVM benches.
 What changed (guarded by flags where applicable):
 - RVAL_FASTPATH: Index write fast‑path
  - `IndexRef.setAt`: direct path for `ObjList` + `ObjInt` (`list[i] = value`) mirrors the read fast‑path. Optional chaining semantics preserved; bounds exceptions propagate unchanged.
 - RVAL_FASTPATH: List literal pre‑sizing
  - `ListLiteralRef.get`: pre‑counts element entries and uses `ArrayList` with capacity hint; for spreads of `ObjList`, uses `ensureCapacity` before bulk add. Evaluation order unchanged.
 - REGEX_CACHE: LRU‑like behavior
  - `RegexCache`: emulates access‑order LRU within a tiny bounded map (`MAX=64`) by moving accessed entries to the tail; improves alternating‑pattern scenarios. Only active when `PerfFlags.REGEX_CACHE` is true.
 Reproduce quick verification (1× runs):
 ```
 ./gradlew :lynglib:jvmTest --tests ExpressionBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ListOpsBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests RegexBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests PicBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests PicInvalidationJvmTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests LocalVarBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ConcurrencyCallBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests DeepPoolingStressJvmTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests MultiThreadPoolingStressJvmTest --rerun-tasks
 ```
 Observation: All listed tests green in this cycle; no behavioral regressions observed. For the new paths (index write, list literal), performance was neutral‑to‑positive in smoke runs; Regex benches remained positive or neutral with the LRU behavior. For publication‑grade medians, extend to 3× per test as in earlier sections.
 ## Sanity matrix (JVM) — quick OFF→ON runs
 Date: 2025-11-11 21:59 (local)
 Scope: Final Round 1 sanity sweep across JVM micro‑benches and stress tests to confirm that optimizations ON do not regress performance vs OFF in representative scenarios. Each benchmark prints `[DEBUG_LOG] [BENCH]` timings for OFF → ON within a single run. This section records a quick pass confirmation (not 3× medians) and reproduction commands.
 Environment:
 - Gradle: 8.7 (stdout enabled, maxParallelForks=1)
 - JVM: as configured by the project toolchain
 - OS/Arch: macOS 14.x (aarch64)
 Benches covered (all green; no regressions observed in these runs):
 - Calls/Args: `CallBenchmarkTest`, `CallMixedArityBenchmarkTest` (ARG_BUILDER)
 - PICs: `PicBenchmarkTest` (field/method); `PicInvalidationJvmTest` correctness reconfirmed
 - Expressions/Arithmetic: `ExpressionBenchmarkTest`, `ArithmeticBenchmarkTest` (RVAL_FASTPATH, PRIMITIVE_FASTOPS)
 - Ranges: `RangeBenchmarkTest` (PRIMITIVE_FASTOPS counted loop)
 - List ops: `ListOpsBenchmarkTest` (PRIMITIVE_FASTOPS specializations)
 - Regex: `RegexBenchmarkTest` (REGEX_CACHE with LRU behavior)
 - Locals: `LocalVarBenchmarkTest` (LOCAL_SLOT_PIC + FAST_LOCAL)
 - Concurrency/Pooling: `ConcurrencyCallBenchmarkTest`, `DeepPoolingStressJvmTest`, `MultiThreadPoolingStressJvmTest` (SCOPE_POOL per‑thread)
 Reproduce (examples):
 ```
 ./gradlew :lynglib:jvmTest --tests CallBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests CallMixedArityBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests PicBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests PicInvalidationJvmTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ExpressionBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ArithmeticBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests RangeBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ListOpsBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests RegexBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests LocalVarBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests ConcurrencyCallBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests DeepPoolingStressJvmTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests MultiThreadPoolingStressJvmTest --rerun-tasks
 ```
  Summary:
  - All listed tests passed in this sanity sweep.
  - For each benchmark’s OFF → ON printouts examined during this pass, ON was equal or faster than OFF; no ON<OFF regressions were observed.
  - For publication‑grade numbers, use the 3× medians methodology outlined earlier in this document. The existing median tables in previous sections remain representative, and the additional tweaks (Index write, List literal pre‑size, Regex LRU, Field PIC 4‑way + read→write reuse, mixed Int/Real fast‑ops) remained neutral‑to‑positive.
 ## Quick snapshot — IndexRef PIC + negative miss cache (JVM) — 3× medians (OFF → ON)
 Date: 2025-11-11 22:32 (local)
 Scope
 - Confirm that the latest changes — IndexRef read/write PIC (stacked on RVAL_FASTPATH) and safe catch‑and‑cache negative entries for Field/Method PICs — do not regress performance. We collected 3× medians for the two expression sub‑benches that are most sensitive to RVAL paths and cross‑checked PICs and ranges.
 Environment
 - Gradle: 8.7 (stdout enabled, maxParallelForks=1)
 - JVM: project toolchain default
 - OS/Arch: macOS 14.x (aarch64)
 Results (3× medians)
 | Area | Benchmark/Test | OFF median (ms) | ON median (ms) | Speedup | Notes |
 |------|-----------------|-----------------:|----------------:|:-------:|-------|
 | RVAL_FASTPATH | ExpressionBenchmarkTest::benchmarkListIndexReads | 304.282 | 229.168 | 1.33× | IndexRef direct fast‑path for ObjList+ObjInt; 4‑way Index PIC handles polymorphic cases |
 | RVAL_FASTPATH | ExpressionBenchmarkTest::benchmarkFieldReadPureReceiver | 275.122 | 194.876 | 1.41× | Monomorphic, immutable receiver path; preserves visibility/optional semantics |
 Cross‑checks (from the same session, 1× quick)
 - PicBenchmarkTest::benchmarkFieldGetSetPic — OFF 203.701 ms → ON 117.129 ms (≈1.74×)
 - PicBenchmarkTest::benchmarkMethodPic — OFF 280.806 ms → ON 202.613 ms (≈1.39×)
 - RangeBenchmarkTest::benchmarkIntRangeForIn — OFF 1762.425 ms → ON 806.898 ms (≈2.18×)
 Reproduce
 ```
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkListIndexReads" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkListIndexReads" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkListIndexReads" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkFieldReadPureReceiver" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkFieldReadPureReceiver" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests "ExpressionBenchmarkTest.benchmarkFieldReadPureReceiver" --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests PicBenchmarkTest --rerun-tasks
 ./gradlew :lynglib:jvmTest --tests RangeBenchmarkTest --rerun-tasks
 ```
 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.
@@ -0,0 +1,58 @@
 # JVM-only Performance Optimization Plan (Saved)
 [//]: # (excludeFromIndex)
 Date: 2025-11-10 22:14 (local)
 This document captures the agreed next optimization steps so we can restore the plan later if needed.
 ## Objectives
 - Reduce overhead on the call/argument path.
 - Extend and harden PIC performance (fields/methods/locals).
 - Improve R-value fast paths and interpreter hot nodes (loops, ranges, lists).
 - Make scope frame pooling thread-safe on JVM so it can be enabled by default later.
 - Keep semantics correct and all JVM tests green.
 ## Prioritized tasks (now)
 1) Call/argument path: fewer allocs, tighter fast paths
 - Extend small-arity zero-alloc path to 6–8 args; benchmark with `CallMixedArityBenchmarkTest`.
 - Splat handling: fast-path single-list splats; benchmark with `CallSplatBenchmarkTest`.
 - Arg builder micro-optimizations: capacity hints, avoid redundant copies, inline simple branches.
 - Optional-chaining fast return (`SKIP_ARGS_ON_NULL_RECEIVER`) coverage audit, add A/B bench.
 2) Scope frame pooling: per-thread safety on JVM
 - Replace global deque with ThreadLocal pool on JVM (and Android) actuals.
 - Keep `frameId` uniqueness and pool size cap.
 - Verify with `DeepPoolingStressJvmTest`, `CallPoolingBenchmarkTest`, and spot benches.
 - Do NOT flip default yet; keep `SCOPE_POOL=false` unless explicitly approved.
 ## Next tasks (queued)
 3) PICs: cheaper misses, broader hits
 - Method PIC 2→3/4 entries (tiny LRU); validate with `PicInvalidationJvmTest`.
 - Field PIC micro-fast path for read-then-write pairs.
 4) Locals and slots
 - Ensure `EMIT_FAST_LOCAL_REFS` coverage across compiler sites.
 - Pre-size `slots`/`nameToSlot` when local counts are known; re-run `LocalVarBenchmarkTest`.
 5) R-value fast path coverage
 - Cover index reads on primitive lists, pure receivers, assignment RHS where safe.
 - Add benches in `ExpressionBenchmarkTest`.
 6) Collections & ranges
 - Tight counted loop for `(Int..Int)` in `for`.
 - Primitive-specialized `ObjList` ops (`map`, `filter`, `sum`, `contains`) under `PRIMITIVE_FASTOPS`.
 7) Regex and string ops
 - Cache compiled regex for string literals at compile time; tiny LRU for dynamic patterns under a new `REGEX_CACHE` flag.
 8) JIT micro-tweaks
 - Inline tiny helpers; prefer arrays for hot buffers; finalize hot classes where safe.
 ## Validation matrix
 - Always re-run: `CallBenchmarkTest`, `CallMixedArityBenchmarkTest`, `PicBenchmarkTest`, `ExpressionBenchmarkTest`, `ArithmeticBenchmarkTest`, `CallPoolingBenchmarkTest`, `DeepPoolingStressJvmTest`.
 - Use 3× medians where comparing flags; keep `:lynglib:jvmTest` green.
 ## Notes
 - All risky changes remain flag-guarded and JVM-only where applicable.
 - Documentation and perf tables updated after each cycle.
@@ -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
@@ -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`.
@@ -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
 }
@@ -0,0 +1,24 @@
 #!/bin/env lyng
 import lyng.io.fs
 val files = Path("../..").list().toList()
 // most long is longest?
 val longestNameLength = files.maxOf { it.name.length }
 // testdoc
 fun test() {
    22
 }
 val format = "%"+(longestNameLength+1) +"s %d"
 for( f in files )
 {
    var name = f.name
    if( f.isDirectory() )
        name += "/"
    println( format(name, f.size()) )
 }
@@ -1,24 +0,0 @@
 /*
    Count "happy tickets": 6-digit numbers where sum of first three
    digits is equal to those of last three. Just to demonstrate and
    test the Lyng way. It is not meant to be effective.
 */
 fun naiveCountHappyNumbers() {
    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 )
                            if( n1 + n2 + n3 == n4 + n5 + n6 ) count++
    count
 }
 //
 // After all optimizations it takes ~120ms.
 //
 val found = naiveCountHappyNumbers()
 println("Found happy numbers: "+found)
 assert( found == 55252 )
@@ -0,0 +1,36 @@
 /*
    Count "happy tickets": 6-digit numbers where sum of first three
    digits is equal to those of last three. Just to demonstrate and
    test the Lyng way. It is not meant to be effective.
 */
 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 ) {
                            if( n1 + n2 + n3 == n4 + n5 + n6 ) count++
                        }
                    }
                }
            }
        }
    }
    count
 }
 import lyng.time
 //
 // After all optimizations it took ~120ms - before.
 //
 for( r in 1..900 ) {
    val start = Instant.now()
    val found = naiveCountHappyNumbers()
    println("Found happy numbers: %d time %s"(found, Instant.now() - start))
    assert( found == 55252 )
    delay(0.05)
 }
@@ -1,2 +1,3 @@
 #!/bin/env jlyng
 println("Hello, world!");
@@ -0,0 +1,10 @@
 #!/bin/env lyng
 val переменная = "значение"
 val data = { greeting: "привет", переменная:, поле: "содержимое" }
 assertEquals(data["переменная"], "значение")
 assertEquals(data["поле"], "содержимое")
 println(data)
@@ -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)
@@ -1,7 +1,7 @@
 #!/bin/env lyng
 /*
    Calculate the limit of Sum( f(n) )
    until it reaches asymptotic limit 0.00001% change
    return null or found limit
 */
 fun findSumLimit(f) {
@@ -13,7 +13,6 @@ fun findSumLimit(f) {
            println("limit reached after "+n+" rounds")
            break sum
        }
        n++
    }
    else {
        println("limit not reached")
@@ -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))
@@ -48,6 +48,7 @@ $$ \ln(1+x)=x-{\dfrac {x^{2}}{2}}+{\dfrac {x^{3}}{3}}-\cdots =\sum \limits _{n=0
    }
    assert( почти_равны( 0.0005, 0.000501 ) )
-Во многих случаях вычисление $n+1$ члена значительно проще cчитается от предыдущего члена, в нашем случае это можно было бы записать через итератор, что мы вскоре добавим.
+Во многих случаях вычисление $n+1$ члена значительно проще cчитается от предыдущего члена, в нашем случае это можно было бы записать через итератор, см [Iterable] и `flow` в [parallelism].
-(продолжение следует)
+[Iterable]: ../Iterable.md
 [parallelism]: ../parallelism.md
@@ -0,0 +1,19 @@
 # Scopes and Closures: compile-time resolution
 Attention to AI: name lookup in runtime `Scope` is legacy. The bytecode compiler uses **compile-time name/member resolution only**.
 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.
 ## 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.
 ## 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.
 ## 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.
@@ -0,0 +1,63 @@
 # Lyng serialization
 Lyng has builting binary bit-effective serialization format, called Lynon for LYng Object Notation. It is typed, binary, implements caching, automatic compression, variable-length ints, one-bit Booleans an many nice features. 
 It is as simple as:
    import lyng.serialization
    val text = "
        We hold these truths to be self-evident, that all men are created equal, 
        that they are endowed by their Creator with certain unalienable Rights, 
        that among these are Life, Liberty and the pursuit of Happiness.    
        "
    val encodedBits = Lynon.encode(text)
    // decode bits source:
    assertEquals( text, Lynon.decode(encodedBits) )
    // compression was used automatically
    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.
 ## Transient Fields
 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:
 ```lyng
 class MyData(@Transient val tempSecret, val publicData) {
    @Transient var cachedValue = 0
    var persistentValue = 42
    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.
 ## 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:
    buffer.toBitInput()
 this possibly creates extra zero bits at the end, as bit content could be shorter than byte-grained but for the Lynon format it does not make sense. Note that when you serialize [BitBuffer], exact number of bits is written. To convert bit buffer to bytes:
    Lynon.encode("hello").toBuffer()
 (topic is incomplete and under construction)
@@ -0,0 +1,77 @@
 # TextMate bundle
 [//]: # (excludeFromIndex)
 The TextMate-format bundle contains a syntax definition for initial language support in
 popular editors that understand TextMate grammars: TextMate, Visual Studio Code, Sublime Text, etc.
 - [Download TextMate Bundle for Lyng](https://lynglang.com/distributables/lyng-textmate.zip)
 > Note for IntelliJ-based IDEs (IntelliJ IDEA, Fleet, etc.): although you can import TextMate
 > bundles there (Settings/Preferences → Editor → TextMate Bundles), we strongly recommend using the
 > dedicated plugin instead — it provides much better support (formatting, smart enter, background
 > analysis, etc.). See: [IDEA Plugin](#/docs/idea_plugin.md).
 ## Visual Studio Code
 VS Code uses TextMate grammars packaged as extensions. A minimal local extension is easy to set up:
 1) Download and unzip the bundle above. Inside you will find the grammar file (usually
   `*.tmLanguage.json` or `*.tmLanguage` plist).
 2) Create a new folder somewhere, e.g. `lyng-textmate-vscode/` with the following structure:
 ```
 lyng-textmate-vscode/
  package.json
  syntaxes/
    lyng.tmLanguage.json   # copy the grammar file here (rename if needed)
 ```
 3) Put this minimal `package.json` into that folder (adjust file names if needed):
 ```
 {
  "name": "lyng-textmate",
  "displayName": "Lyng (TextMate grammar)",
  "publisher": "local",
  "version": "0.0.1",
  "engines": { "vscode": "^1.70.0" },
  "contributes": {
    "languages": [
      { "id": "lyng", "aliases": ["Lyng"], "extensions": [".lyng"] }
    ],
    "grammars": [
      {
        "language": "lyng",
        "scopeName": "source.lyng",
        "path": "./syntaxes/lyng.tmLanguage.json"
      }
    ]
  }
 }
 ```
 4) Open a terminal in `lyng-textmate-vscode/` and run:
 ```
 code --install-extension .
 ```
   Alternatively, open the folder in VS Code and press F5 to run an Extension Development Host.
 5) Reload VS Code. Files with the `.lyng` extension should now get Lyng highlighting.
 ## Sublime Text 3/4
 1) Download and unzip the bundle.
 2) In Sublime Text, use “Preferences → Browse Packages…”, then copy the unzipped bundle
   to a folder like `Packages/Lyng/`.
 3) Open a `.lyng` file; Sublime should pick up the syntax automatically. If not, use
   “View → Syntax → Lyng”.
 ## TextMate 2
 1) Download and unzip the bundle.
 2) Double‑click the `.tmBundle`/grammar package or drag it onto TextMate to install, or place
   it into `~/Library/Application Support/TextMate/Bundles/`.
 3) Restart TextMate if needed and open a `.lyng` file.
		`@@ -0,0 +1,3 @@`
							`The Lyng programming language.`
							`Copyright (c) 2024-2025 Sergey Chernov real.sergeych@gmail.com`
		`@@ -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.__`
		`@@ -0,0 +1,3 @@`
							`#!/usr/bin/env lyng`

							`println("Hello from lyng!")`
`@@ -1,2 +1,3 @@`
		`#!/bin/env jlyng`

	`println("Hello, world!");`	`println("Hello, world!");`