chore(deps): bump softprops/action-gh-release from 2 to 3 - #2

…fo in task_queue

…v at startup

… 3 languages

…ring i18n in format_success

…d analysis_tools.py

- search_tools.py: 8 calls fixed — translated templates to English
- analysis_tools.py: 4 calls fixed — translated status messages to English
- Add audit_i18n.py script for detecting i18n issues
- Add _fill_translations.py helper for bulk translation

- Documented Zed crash analysis (0xc0000409)
- Memory leak investigations (MCP server, ScopedKeyValueStore)
- GitHub issue #60475 filing
- Timeline of 5 crashes with Windows Event Log data

- TaskQueue: add auto cleanup loop (60s cycle), dedup by task name,
  release func/args/kwargs references after task completion,
  fix cleanup_old_results to also clear _pending_names
- HeartbeatService: add _on_monitor_done callback to release
  asyncio Task reference, ensure _running flag is reset in
  _monitor() finally block
- Add has_pending() API and change submit/submit_sync to return
  Optional[str] (None if duplicate task name detected)

- _ping_lm_studio now detects both instruct and embedding models
- _embedding_rerank uses lm_studio_embedding_model (not instruct model)
- Reranker models (bge-reranker) are excluded from embedding-rerank
- Prevents Invalid model identifier error on /v1/embeddings

- _query_lm_studio now tries /v1/chat/completions first,
  falls back to /v1/completions for base/reranker models
- _ping_lm_studio prioritizes reranker > instruct > any
  non-embedding model for LLM-reranking
- bge-reranker-v2-m3 is now selected for reranking
- Embedding-rerank still uses dedicated embedding model

…rerank

…llback to first model

…LM final)

… TTL

…call

…iming, LM Studio live reload, detailed telemetry

…mbed for embedding

- Add lm_studio_reranker_model field for bge-reranker-v2-m3-m3
- Fix _ping_lm_studio to detect 3 model types: embed/reranker/LLM
- Remove len(chunks) <= 1 guard that skipped all reranking
- Add _cross_encoder_rerank method for Stage 2
- Three-stage pipeline with graceful fallback chain
- Fix _check_llm_available cache initialization
- Reduce chunk preview length to 400 for faster embedding
- Add per-stage timing telemetry with model names
- Add LLM stage timeout (4s) to prevent long hangs

phi-4-mini-instruct on this hardware runs at ~7 tok/s with
prompt processing at ~6ms/tok. Full pipeline with all three
models now completes: embed(346ms) -> reranker(402ms) -> LLM(10368ms)
= 11s total with complete JSON score generation

…cisions

…ecomposition)

Created src/core/indexing/index_pipeline.py:
- IndexPipeline.process_file() handles parse -> embed pipeline
- Extracted ~150 lines of indexer.py (AST chunking, embedding, SymbolIndex)
- indexer.py _index_single_file now delegates to self._pipeline.process_file()
- Clear separation: IndexPipeline = core logic, Indexer = orchestration + write

Next: extract IndexStatusReporter (get_status), then Watchdog.
indexer.py: 1700 -> ~1100 lines in 2 phases. Target: ~800 lines as facade.

…ject decomposition)

Created src/core/indexing/index_status.py:
- IndexStatusReporter manages get_status, stale file scan, cache
- Indexer.get_status now delegates to self._status_reporter.get_status()
- -100 lines from indexer.py, clear separation of concerns

Created src/core/indexing/index_project_runner.py:
- Extracted 360 lines (index_project + nested _parse_worker_file + _notify_progress)
- 3-phase pipeline: parallel parse -> batch embed -> write + prune + BM25 + IVF
- Indexer.index_project now delegates to self._project_runner.run()

indexer.py: 1049 -> ~650 lines remaining.
~500 lines left: move_chunks_metadata, apply_file_move, _parse_file_only,
  _infer_module_name, _infer_layer, _escape_file_path_for_lance, index_file

Tests: 501/501 PASS

- FileMoveManager: move_chunks_metadata + apply_file_move (-100 lines)
- IndexerUtils: calculate_file_hash, escape_path, infer_module/layer
- _parse_file_only delegated to IndexPipeline
- IndexPipeline now uses shared _parse_chunks (no code duplication)

indexer.py: 1700 -> 675 lines (clean facade)
indexing/ package: 17 files, 6,871 lines total

Tests: 501/501 PASS

All modules now explicitly declare their public API surface.
This prevents  from leaking private imports
through shim files.

Added via AST-aware script: each module's __all__ lists only
module-level classes and functions (not methods).

layer.py re-exports: IntelligenceStore, JobHistoryStore, BackgroundJob

Tests: 501/501 PASS

…key, add noqa for intentional redefs

- F841: removed unused escaped_new (file_move_manager) and current_hash (indexer)
- F401: removed unused get_log_summary import (system_tools)
- F601: fixed repeated 'status' key -> 'indexer_status' (system_tools)
- F811: added # noqa: F811 for intentional redefs in server.py + server_factory.py
- E402: moved import to top in base.py
- E741: renamed 'l' -> 'line' in zed_config.py

…tools hardcode

- Fix hardcoded total_intel=14->12, total_diag=3->6 in server_tools.py
- Fix inline tools docstring (7->6) and comment (7->6)
- Fix server_factory.py: .env encoding='utf-8' + get_running_loop()
- Remove duplicate intel_get_telemetry registration
- Update AGENTS.md, SKILL.md, MCP_TOOLS.md: 36->37 tools
- Update README (en/ru/zh): 59->37, fix breakdowns
- Update ARCHITECTURE, FAQ, GRACEFUL_DEGRADATION, HANDFOFF, ZED_WINDOWS_QUIRKS

…file(abs_path) + add invariant tests

SYM-INDEX-PARTIAL: _parse_file_only вызывал pg.remove_file(rel_path_str)
напрямую (строка 336), что приводило к path mismatch — PropertyGraph
хранит file_path в абсолютном формате, а rel_path_str относительный.

Фикс: self._symbol_index.remove_file(str(full_path)) — через адаптер
с правильной нормализацией пути.

Тесты:
- test_symbol_index_invariant.py (13 тестов): инварианты чистого SymbolIndex
  (консистентность _definitions/_references, reindex, thread-safety, GC)
- test_sym_index_partial.py (7 тестов): SymbolIndexAdapter + PropertyGraph
  (атомарность remove/add, cross-file refs, path mismatch detection)

… to separate files

graph_adapter.py reduced from 1308 to 1089 lines (-220).

- graph_rag_adapter.py (169 lines): GraphRAGAdapter (query_impact, query_feature,
  query_dependencies, query_tests, query_hotspots, query_similar_bugs)
- composition_adapter.py (93 lines): CompositionAdapter (единая точка входа
  для SymbolIndexAdapter + GraphRAGAdapter)
- graph_adapter.py: only SymbolIndexAdapter remains (facade, ~970 lines)
- Все импорты сохранены, циклических зависимостей нет

cypher_engine.py reduced from 1271 to 46 lines (facade with re-exports).

New files:
- cypher_lexer.py (188 lines): TokenType, Token, CypherLexer
- cypher_ast.py (97 lines): all AST node types + internal AST helpers
- cypher_parser.py (458 lines): CypherParser (recursive descent)
- cypher_sql.py (401 lines): CypherToSQL (AST -> SQL transpiler)
- cypher_executor.py (115 lines): CypherExecutor + query_graph()
- cypher_engine.py (46 lines): facade re-exporting all public API

All 47 existing Cypher tests pass without changes.

multi_provider.py reduced from 1190 to 1038 lines (-152).

New files:
- reranker_scoring.py (130 lines): cosine_similarity, validate_scores,
  parse_scores_json, apply_scores — standalone functions
- search_result_reranker.py (130 lines): SearchResultReranker class

…ти, маркер

P0-4: logger до определения в main.py. Перенесён наверх модуля.
P0-1: extension.toml version 2.7.1 → 3.2.3 (синхронизация с pyproject).
P0-2: extension.toml args run_mcp.py → -m src.main (файла не существовало).
P0-3: 6 missing deps добавлены в pyproject.toml (pandas, onnxruntime,
      openvino, transformers, tokenizers, python-dateutil).
P0-5: "MSCodeBase" in path → маркерный файл __mscodebase_ext__.marker.
      install.py создаёт маркер при установке расширения.

P1-5: pylance>=8.0.0 удалён из requirements.txt — не импортируется,
      тянется транзитивно через lancedb.
P1-8: .github/dependabot.yml добавлен — weekly schedule для pip + actions.
P1-7: mypi continue-on-error: true убран — тип-чекер стал gatekeeper.

Bumps [softprops/action-gh-release](https://github.com/softprops/action-gh-release) from 2 to 3.
- [Release notes](https://github.com/softprops/action-gh-release/releases)
- [Changelog](https://github.com/softprops/action-gh-release/blob/master/CHANGELOG.md)
- [Commits](softprops/action-gh-release@v2...v3)

---
updated-dependencies:
- dependency-name: softprops/action-gh-release
  dependency-version: '3'
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <[email protected]>