- Comparisons set __CT_RET directly instead of $() subshell fork
- Logical NOT (!) inlined same way
- fs.exists() inlined to [[ -e ]] in conditions
- fs.read uses $(<file) instead of $(cat), fs.write/append use printf
- fs.exists/remove/mkdir avoid unnecessary $() wrapping
- Dict .get()/.has()/.len() inlined for local dict vars
- __ct_print simplified: removed unnecessary local variable
- String concat ..= uses native += append
- Skip __CT_RET self-assignment in return statements

- Remove redundant echo in @awk functions (leaked to stdout)
- Replace range() subshell fork with C-style for ((i=0; i<N; i++))
- Inline string methods (upper/lower/len/trim/replace/contains/starts/ends/substr/index) for simple variable receivers
- Direct array subscript for .get() and .set() instead of function calls

- Fix numeric/string comparison operator selection in expr and stmt codegen
- Use ${VAR:-} for env var reads to prevent unbound variable errors
- Add {{ and }} as literal brace escapes in string interpolation
- Fix DCE to discover transitive class dependencies through method callees
- Fix &&/|| condition codegen to prevent __CT_RET variable conflicts

Add string concatenation assignment operator (..=) across the full
pipeline: lexer, parser, and backend codegen for variables, this.field,
dict fields, and object fields.

Fix __ct_last_instance set after construct call instead of before.
Fix IR builder treating field symbols as known function calls.
Fix resolver _current_func corruption during constructor resolution.

Update bootstrap/ → compiler/ in install paths.

- Add configure_file for content.in → content binary in compiler/meson.build
- Update spec: bootstrap/ → compiler/ in %files

Replace the bootstrap mixin-based compiler with a new IR-based pipeline:
Source → Lexer → Parser → Resolver → TypeChecker → IR Builder → Optimizer → Backend

New architecture (compiler/):
- semantics/: scope resolution, type system, type checker
- ir/: intermediate representation with stable node IDs
- optimizer/: DCE via call graph BFS, CSE, constant folding
- backend/bash/: bash code generation from IR
- backend/awk/: AWK code generation sharing same IR
- symbols/: symbol table with LSP-ready serialization

Key improvements:
- Explicit shell command detection (IRCall.is_shell_cmd)
- Namespace prefixing only in bash backend, not in resolver
- DCE reduced from 580-line fixed-point to ~40-line BFS
- CSE uses stable node_id instead of Python id()
- Shell commands in assignments use || true for set -e safety
- No local keyword in global scope (when/foreach)
- Coproc uses exec to prevent orphan child processes

All 400 tests pass. Verified with real-world unified-theme-switcher service.

Duplicate coproc fds with exec instead of copying fd numbers.
Bash auto-closes coproc fds when the process exits, causing
race conditions on slower systems. Save PID before coproc
cleanup and close original write fd after dup to prevent hangs.

- Add foreach line in proc syntax for process handle iteration
- Fix DCE not tracking class methods used via objects
- Fix CSE precomputing is_empty/is_number (they echo, don't set __CT_RET)
- Fix proc.read() leaking echo to stdout when assigning to variable
- Fix if/while condition with shell calls resolving incorrectly
- Support hyphenated shell commands in lexer (e.g. dbus-monitor)

- namespace X { } for symbol isolation (multi-file merge)
- using X / using h = X / using X { a, b } (Vala-style)
- busing for sourcing .sh scripts with named access
- Remove legacy import system
- Rewrite cli.ct: action callbacks, aliases, FloatFlag, auto help/version
- Add cli examples: cli_demo, cli_flags, cli_subcommands, cli_categories
- Fix parent class field initialization in child constructors
- Fix var.field.method() dispatch for inherited array/dict fields
- Recursive directory scan for content build/run/test
- Full build-lib with metadata, directory support, --install
- Meson custom_target to compile cli.sh during build
- RPM spec: split content-lib-cli subpackage

- json.unmarshal(str, Class) deserializes JSON into class instance
- json.marshal(obj) serializes class instance to JSON string
- reflect module: fields, get, set, type, class_name, create
- Class metadata generation for runtime introspection
- Fix DCE namespace detection inside function bodies
- Update telegram bot example with Message class, /json, /info commands

- DCE: propagate parent methods to child classes so inherited proxy
  functions are generated (StringFlag.with_short etc.)
- stmt_codegen: return this.field.method() now checks field type
  (array/dict) instead of always assuming string
- stmt_codegen: fix missing f-prefix on two RET_VAR f-strings
- cli.ct: exit(0) after --help/--version to prevent double output
- Add TestInheritance tests covering all fixed bugs

Fix spaces before parentheses in CT source code within tests.

Unknown function calls now compile directly as shell commands:
grep ("pattern") instead of shell.exec ("grep pattern").
Added mixed pipe support (CT func | shell cmd) with fd 3 redirect.
Removed ngrep builtin (grep can be called directly now).

- CLAUDE.md: expanded @awk section with new features
- README.md: updated @awk supported features list
- LANGUAGE_SPEC.md: added AWK extras, dict.field syntax

AWK codegen:
- Add string interpolation: "Hello {name}" -> ("Hello " name "!")
- Add env.VAR support via ENVIRON["VAR"]
- Add time.now() via systime()
- Add str.chr() via sprintf("%c", n)
- Add assert_eq() for equality assertions
- Add print() as statement
- Add method calls as statements (arr.push, dict.set)

Bash codegen:
- Fix dict.field = value for dict variables (was using wrong storage)
- Fix dict.field read to use dict["field"] syntax

Support assigning to object fields from outside the class using
obj.field = value syntax with all compound operators (=, +=, -=, *=, /=, ..=).
Also support typed parameters in functions to enable field assignment
on passed objects.

- env.VAR = value generates export VAR="value"
- Supports interpolation: env.GREETING = "Hello, {name}!"
- Add 2 tests for env.set functionality
- Update all documentation files

- Access env vars directly: env.HOME, env.USER, env.PATH
- Generates ${VAR} without fork (no shell.capture needed)
- Type is 'any', can be explicitly typed: port: int = env.PORT
- Add 4 tests for env variables including int arithmetic
- Update all documentation files

- Add TypeAnnotation AST node for type annotations
- Add ClassField dataclass for typed class fields
- Extend Parameter, FunctionDecl, Assignment with type_annotation field
- Add type parsing: parse_type(), parse_type_annotation(), parse_function_type()
- Support types: string, int, float, bool, any, void, T[], dict[K,V], classes, (A) => B
- Add type checking at compile time with configurable error handling
- Add --no-type-check and --warn-types CLI flags
- Fix callback variable calls (prioritize callback_vars, use $() capture)
- Document closure limitation (lambdas can't capture outer variables)
- Add 28 new tests for typing and callbacks (279 total)

- User decorators (@obj.method(args)) for handler registration
- Callbacks: pass functions/lambdas as arguments
- Fix UTF-8 urlencode with LC_ALL=C
- Fix dict field access in class methods
- Refactor telegram_echobot to use decorators
- Add tests for callbacks and user decorators
- Update documentation (LANGUAGE_SPEC, README)

- Split methods.py into methods/ directory with separate modules
- Add awk_builtin field to Method for unified AWK generation
- Replace hardcoded method dispatch in awk_codegen with generate_awk()
- Add RET_VAR/RET_ARR constants, replace hardcoded __CT_RET
- Migrate all codegen files to use indented() context manager

- Import method definitions from methods.py for consistency
- Define bash implementations in STRING_IMPLS, ARRAY_IMPLS, DICT_IMPLS
- Generate stdlib functions from these definitions
- Use context manager (with self.indented()) for cleaner code
- Reduce code duplication and improve maintainability

- Add proper type tracking for function and method parameters
- DCE now tracks array-returning methods (keys, split, slice)
- Fix split/slice to return arrays via __CT_RET_ARR
- Add foreach support for inline split expressions
- Detect object parameters by unknown method calls
- Add comprehensive tests for parameter passing

New test classes:
- TestFunctionParameterPassing (11 tests)
- TestClassInstancePassing (4 tests)

- String interpolation with operators: {a == b} now generates proper
  bash comparison instead of invalid ${a == b}
- Double method call in conditions: CSE precompute now correctly reuses
  temp variable for CallExpr to avoid calling method twice
- arr.push() with method call: detect side effects and call method
  separately to avoid subshell state isolation
- charAt() with newline: use printf X marker to preserve trailing newlines
- this.field.method() and obj.field.method(): handle nested MemberAccess
  in return statements and assignments
- Namespace vs variable collision: check if identifier is known variable
  before treating as namespace
- return func() with arrays: use _generate_call_arg for proper nameref

The dead code elimination was not tracking string/array method usage
when methods were called on class fields like this.source.len().
Now properly calls _check_method() when field class is unknown.

- json.parse() now creates a dict from JSON string
- json.stringify() converts dict back to JSON
- Dict methods (.get, .set, .has, .del, .keys) work on parsed JSON
- Added 4 integration tests for JSON functionality
- Updated documentation

- Fix benchmark numbers (47s/0.09s, ~530x speedup)
- Add full project structure with all codegen mixins
- Replace "Управление потоком" with "Условия и циклы"
- Remove stale internal reference

Made bench_heavy_awk.ct and bench_heavy_bash.ct identical except for
@awk decorator. Fixed compound assignment with method calls that was
causing different results.

New features:
- .map() / .filter() for arrays with lambda support
- @validate decorator for argument validation
- fs.open() / f.read() / f.write() / f.close() file handles
- with X in Y { } context manager for auto-closing resources

All features work in both Bash and @awk codegen.
Added 18 new pytest tests for the new functionality.
Updated documentation (LANGUAGE_SPEC.md, README.md, README_ru.md).