graphlint 0.3.0__tar.gz → 0.3.2__tar.gz

{graphlint-0.3.0 → graphlint-0.3.2}/CHANGELOG.md

@@ -1,243 +1,275 @@
-# Changelog
-
-All notable changes to this project will be documented in this file.
-
-## [0.3.0] - 2026-07-23
-
-### Added
-- Experimental Rust language analysis via `tree-sitter-rust` (opt-in via `[rust]`
-  optional dependency group)
-- `query --public-as-entry` option: treats all public functions/methods as entry
-  points for dead-code reachability analysis
-- `pub_api` entry rule template (disabled by default) for projects that want
-  public API surface as the primary entry
-
-
-## [0.2.0] - 2026-07-22
-
-### Added
-- Multi-language architecture: `LanguageAdapter` abstract base class and
-  `LanguageRegistry` for extension-based routing of source files to language
-  backends, laying the foundation for future multi-language support
-- `language/python/` package: Python-specific adapter, parser, entry detector,
-  imports analyzer, decorators resolver, and constants — decoupled from the
-  core framework
-- Per-language `constants.py`: configurable public API dunders, special
-  methods, and test-file detection
-- Type annotation awareness in AST visitor: automatically resolves unpacked
-  variable types (e.g., loop bindings, tuple unpacking) for more accurate
-  read/write edge tracking
-
-### Changed
-- **[Breaking]** Analyzer modules reorganized: `parser.py`, `entry_detect.py`,
-  `imports.py`, `decorators.py`, `visitor.py` moved into `language/python/` —
-  direct imports of old module paths will break
-- Entry point rules now serve as initialization template only: rules are
-  written to `.graphlint/config.json` on first build, after which the config
-  file is the single source of truth
-- Default entry strategy simplified: entry rules are no longer hardcoded in
-  the indexer; they come solely from runtime configuration
-- `GraphBuilder` no longer owns entry detection — entry points are resolved
-  through the language adapter
-- `WarningCollector` simplified: language-specific constant tables moved to
-  `language/python/constants.py`
-
-## [0.1.12] - 2026-07-21
-
-### Added
-- `query --fail-on <warn-types>` option: returns exit code `2` when matching
-  warning types are found in query results, enabling CI pipeline integration
-- `graphlint config` subcommands now return exit code `1` on error (previously
-  always returned `0`)
-
-## [0.1.11] - 2026-07-21
-
-### Changed
-- Entry point detection now uses a unified pattern-driven architecture — all
-  rules (built-in and custom) share the same `_detect_custom` code path
-- Built-in detectors replaced with declarative `ast_pattern` expressions using
-  the existing prefix syntax (`function_call:`, `decorator:`, `if_name_main`,
-  `file_match:`, `test_file`)
-- `ast_pattern` supports OR combinations with ` | ` separator (e.g.
-  `class_instantiation:FastAPI | function_call:uvicorn.run`)
-- `ast_pattern` field is no longer required for entry rules — rules without
-  it are silently skipped instead of failing validation
-- Decorator matching uses `fnmatch` instead of substring `in`, providing
-  precise glob matching and eliminating false positives from partial name
-  overlaps
-- `function_call:` and `class_instantiation:` patterns now support `fnmatch`
-  globs (was exact match only)
-
-### Fixed
-- `file_match:` pattern now correctly checks the file path at rule level
-  instead of a non-existent `node.filename` attribute, fixing the bug where
-  `file_match:` rules never matched
-- `graphlint config copy-from` no longer picks up irrelevant query/analyzer
-  parameters, fixing the config copy command exception
-
-## [0.1.10] - 2026-07-09
-
-### Fixed
-- Assignment target context resolution now correctly distinguishes class fields
-  (`self.xxx` / `cls.xxx`), method-local variables, and module-level variables,
-  preventing misclassification of `self.x = y` inside methods as local variables
-- Class fields assigned across multiple methods (e.g. `__init__` + `set_x`) are
-  now deduplicated to a single node, eliminating duplicate field nodes
-
-### Changed
-- Attribute-write edges (`self.xxx = value`) now also emit a `read` edge on
-  the parent object (`self`), ensuring proper usage tracking of the containing
-  class instance
-
-## [0.1.9] - 2026-07-08
-
-### Fixed
-- FOREIGN KEY constraint failure on incremental rebuilds — prebuilt edge
-  remapping now correctly writes remapped node IDs into edge objects
-- `_insert_nodes` node-to-file key collision when multiple changed files
-  share the same qualified name and line number
-
-### Changed
-- Incremental rebuild now parses only changed files instead of all files
-- Filesystem scan merged into a single `os.walk` pass instead of two
-- Redundant file hash computation eliminated in both full and incremental paths
-
-### Added
-- `component_members` table in SQLite schema for future incremental
-  connectivity analysis
-
-## [0.1.8] - 2026-07-07
-
-### Added
-- Agent prompt auto-update: `install` now scans all installed prompts on start,
-  replaces outdated content when its version is below the library version,
-  and prints the list of updated agents
-- `install` / `uninstall` interactive selection: support Ctrl+C to abort and
-  empty input to cancel; prompt texts are now i18n-localized
-- `graphlint prompt` subcommand to copy the agent prompt to the system clipboard
-- Documentation for agent integration (`docs/*/guide/agent-integration.md`)
-
-### Fixed
-- Query deadlock: `_build_edge_mapping` now avoids `result` future list
-  mutation under concurrent executor shutdown, preventing infinite waits
-
-## [0.1.7] - 2026-07-06
-
-### Fixed
-- `_resolve_symbol` scope filtering no longer drops cross-class `call` edges
-  when a method resolves a same-named call via a different class (registry
-  / dispatch pattern) — all suffix-matched candidates are now included
-- Entry point `file_pattern` matching (`**/*.py` via fnmatch) now also
-  matches root-level Python files, not only files in subdirectories
-
-## [0.1.6] - 2026-07-05
-
-### Changed
-- `ThreadPoolExecutor` replaced with `ProcessPoolExecutor` for edge building
-- Incremental builds on file change now trigger a full rebuild instead of loading prebuilt edges
-- Module-level usage expansion in connected-component analysis uses a Counter to distinguish real references from synthetic edges
-- Dead-code merge loop uses O(1) `comp_by_id` dict lookup instead of O(C) linear scan
-
-### Removed
-- `_quick_changed_check()` — mtime-based fast-path stamp was unreliable on Windows NTFS and could trigger unnecessary incremental builds that produced stale edges
-- Ghost-node insertion workaround and stale-edge skip logic in `_do_insert_edges`
-
-## [0.1.5] - 2026-07-05
-
-### Added
-- `global` keyword support: `visit_Global` handler tracks names declared as `global X` and resolves references to module-level qualified names (`module.X`) instead of local ones, preventing false local node creation
-- `visit_AugAssign` handler for augmented assignment (`X += 1`, `X *= 2`, etc.), generating both read and write references to the target
-- Scope isolation for `global` declarations — `_global_names` is saved/restored on function entry/exit, preventing leakage between functions
-
-### Fixed
-- `global X` in functions/methods no longer creates spurious local variable nodes for globally-declared names
-- Augmented assignment (`X += 1`) was silently dropped during AST traversal — now correctly produces both read and write edges
-
-## [0.1.4] - 2026-07-04
-
-### Fixed
-- Module-level dead code detection: removed synthetic `read` edges from pseudo-node 0 that masked all top-level symbols as "used", enabling detection of genuinely dead module-level functions and variables
-- Module-level write references no longer create unresolvable edges — variable definitions are not treated as write operations, preventing silent edge drops
-- Module-level read/call references now correctly create edges from pseudo-node 0, fixing a false positive where module-level used variables (e.g., `WARN_TYPE_VALUES` read by `VALID_WARN_TYPES = WARN_TYPE_VALUES`) were flagged as dead code
-- Public API dunders (`__all__`, `__version__`, etc.) no longer appear as isolated components — they stay anchored to their file via synthetic containment edges
-- Python special methods (`__init__`, `__enter__`, `__exit__`, etc.) no longer generate redundant dead-code warnings when their parent class is already flagged as unreachable
-- Unreachable classes with special method overloads now produce only the class-level `dead_code` warning, not per-method duplicates
-
-## [0.1.3] - 2026-07-04
-
-### Added
-- I18n support for CLI parameter help text: `PARAM_DEFS` help strings now go through the i18n system via `help_key`, enabling full language switching for `--help` output
-
-### Changed
-- `ParamDef` dataclass: added `help_key` field for i18n key lookup
-- `cli.py:_add_arg()` now resolves help text via `_t(help_key)` when available
-- Updated `en.py` and `zh_CN.py` with all `help.param.*` translation entries
-
-## [0.1.2] - 2026-07-04
-
-### Added
-- SQLite composite indexes (`idx_edges_source_type`, `idx_edges_target_type`) to accelerate edge-type filtered lookups
-- SQLite single-column index (`idx_warnings_node`) to speed up warnings JOIN queries
-- SQLite sorted indexes (`idx_snapshots_warnings`, `idx_snapshots_nodes`) for faster list_graphs pagination
-- PRAGMA optimizations on Database init: `cache_size=-8000` (8 MB), `mmap_size=268435456` (256 MB), `temp_store=MEMORY`
-- `_quick_changed_check()` — mtime-based fast-path stamp that short-circuits the entire build when no files have changed
-- `_update_scan_stamp()` — persists `{path: mtime_ns}` snapshot inside IndexLock after each successful build
-- `warnings_summary` cache in `QueryEngine.list_graphs()` — lazily computed then shallow-copied, eliminating repeated `GROUP BY` scans
-- `_precompute_edge_counts()` — single-pass edge count per component, replacing O(n × m) per-component iteration
-
-### Changed
-- Eliminated second AST traversal: edge building is now driven by `ReferenceInfo` collected during the single `ASTVisitor` pass — removed `_walk`, `_read_edges`, `_proc_call`, `_proc_assign`, `_proc_annassign`, `_target_ids`, `_read_target_expr` (~400 lines). In a large codebase (700+ `.py` files, 1000+ classes, 14,000+ functions), full rebuild time improved from ~2500s to ~200s.
-- `_build_file_edges_worker` now consumes `ParseResult.references` directly instead of re-walking the AST tree
-- Module pseudo-node (id=0) connectivity tightened: liveness only propagates through `read`/`call` edges, not `write` edges — preventing false reachability of write-only variables and fixing false dead-code reports for module-level used variables
-- Connected component BFS no longer unconditionally seeds node 0, avoiding spurious reachability
-- Module-level synthetic edges recreated for all files each build instead of only for unchanged files
-- Optimized multiple AST I/O bottlenecks in source file reading and tree reuse
-- `_symbol_index` / `_suffix_index` from `dict` to `defaultdict(list)`, removing all `.setdefault()` calls
-- `_resolve_symbol` now accepts a per-worker `resolve_cache` dict — repeated `(qname, scope)` pairs are resolved once per file
-- `call_graph` prebuilt once in `find_connected_components` and reused across `compute_entry_reachability` calls, eliminating redundant edge re-traversal
-- `_insert_nodes` uses a pre-built `(qualified_name, line_start) → file_path` map instead of O(n²) linear scan via `_node_path()`
-- Full rebuild (`incremental=False`) now uses per-table `DELETE FROM` instead of per-file `_delete_old` + selective cleanup
-- Incremental build edge insertion now uses global `DELETE FROM edges` instead of per-file partial delete, eliminating dangling old node-ID references
-- `load_prebuilt_edges` accepts optional `sql_to_mem` mapping, simplifying caller in `indexer.py`
-- Agent injection prompt strengthened to warn when graphlint CLI is unavailable
-
-### Removed
-- `_node_path()` helper function (replaced by pre-built node-to-file map)
-
-### Fixed
-- Windows compatibility: `os.unlink()` with `missing_ok=True` replaced by `if exists()` guard
-
-## [0.1.1] - 2026-07-04
-
-### Added
-- Parallel edge building via ThreadPoolExecutor for changed files (`performance.parallel_workers`)
-- Multi-level suffix index for improved symbol resolution beyond single-level short names
-- Auto-build retry with full index rebuild on FOREIGN KEY constraint failure
-
-### Changed
-- Agent injection prompt wording strengthened to "Always use" with fallback instruction when CLI is unavailable
-- Entry detection methods now reuse pre-parsed source (`pr.source`) to reduce redundant file I/O
-- Refactored AST walk functions to module level to enable parallel execution
-
-### Fixed
-- FOREIGN KEY constraint violations in incremental mode by sorting nodes parent-first before DB insert
-- Incremental rebuild no longer rebuilds edges for unchanged files
-
-## [0.1.0] - 2026-07-01
-
-### Added
-- Initial release
-- AST-based Python source code parsing (classes, functions, methods, variables, fields)
-- Dependency graph construction with 5 edge types (read, write, call, inherit, decorate)
-- 8 built-in entry point detection rules (main, FastAPI, Flask, Django, Click, Typer, Celery, pytest)
-- Custom entry point rules via AST pattern matching
-- 11 warning types (unused_import, dynamic_import, circular_ref, syntax_error, write_only, deprecated_usage, dead_code, type_mismatch, unresolved_ref, unused_variable, file_too_large)
-- Incremental indexing with SHA256-based change detection
-- Adaptive output volume strategy (full, index, truncated)
-- Internationalization (English and Simplified Chinese)
-- CLI interface (`graphlint query`, `graphlint build`, `graphlint config`)
-- Python API (`query()`, `build()`, `configure()`)
-- SQLite persistence with cross-platform file locking
-- Parallel file parsing via ProcessPoolExecutor
-- Configuration management with dot-notation key access
-- Comprehensive test suite (unit, integration, performance)
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+## [0.3.2] - 2026-07-25
+
+### Changed
+- First-time builds now run a full rebuild; all subsequent queries use
+  incremental mode (previously every query forced a full rebuild)
+- Incremental rebuild restructured into 6 clearly delineated phases: Remove →
+  Parse → Downstream Reconnect → Rebuild → Persist → Verify
+
+### Fixed
+- Incremental rebuild: empty `set()` passed as `changed_files` was treated as
+  falsy and replaced with all files, causing unnecessary full re-parse
+- Incremental rebuild: unchanged child nodes whose parent received a new ID in
+  a changed file had stale `parent_node_id` values in the database (in-memory
+  remapping was not written back)
+- Incremental rebuild: removed files were not included in downstream
+  reconnection, causing FOREIGN KEY violations when prebuilt edges referenced
+  deleted symbols
+
+## [0.3.1] - 2026-07-24
+
+### Added
+- Cargo workspace detection: `[[bin]]` path entries in `Cargo.toml` are now
+  recognized as Rust language entry points
+
+### Fixed
+- SQLite `IN` clause batching: queries with >999 variables are now split into
+  batches to avoid the SQLite variable limit, preventing crashes on large
+  codebases
+- Dotted imports (`import X.Y.Z`) are now correctly detected as used, fixing
+  false-positive `unused_import` warnings for multi-level imports
+- Relaxed `tree-sitter-rust` version bound to allow Rust grammar v15+
+
+## [0.3.0] - 2026-07-23
+
+### Added
+- Experimental Rust language analysis via `tree-sitter-rust` (opt-in via `[rust]`
+  optional dependency group)
+- `query --public-as-entry` option: treats all public functions/methods as entry
+  points for dead-code reachability analysis
+- `pub_api` entry rule template (disabled by default) for projects that want
+  public API surface as the primary entry
+
+
+## [0.2.0] - 2026-07-22
+
+### Added
+- Multi-language architecture: `LanguageAdapter` abstract base class and
+  `LanguageRegistry` for extension-based routing of source files to language
+  backends, laying the foundation for future multi-language support
+- `language/python/` package: Python-specific adapter, parser, entry detector,
+  imports analyzer, decorators resolver, and constants — decoupled from the
+  core framework
+- Per-language `constants.py`: configurable public API dunders, special
+  methods, and test-file detection
+- Type annotation awareness in AST visitor: automatically resolves unpacked
+  variable types (e.g., loop bindings, tuple unpacking) for more accurate
+  read/write edge tracking
+
+### Changed
+- **[Breaking]** Analyzer modules reorganized: `parser.py`, `entry_detect.py`,
+  `imports.py`, `decorators.py`, `visitor.py` moved into `language/python/` —
+  direct imports of old module paths will break
+- Entry point rules now serve as initialization template only: rules are
+  written to `.graphlint/config.json` on first build, after which the config
+  file is the single source of truth
+- Default entry strategy simplified: entry rules are no longer hardcoded in
+  the indexer; they come solely from runtime configuration
+- `GraphBuilder` no longer owns entry detection — entry points are resolved
+  through the language adapter
+- `WarningCollector` simplified: language-specific constant tables moved to
+  `language/python/constants.py`
+
+## [0.1.12] - 2026-07-21
+
+### Added
+- `query --fail-on <warn-types>` option: returns exit code `2` when matching
+  warning types are found in query results, enabling CI pipeline integration
+- `graphlint config` subcommands now return exit code `1` on error (previously
+  always returned `0`)
+
+## [0.1.11] - 2026-07-21
+
+### Changed
+- Entry point detection now uses a unified pattern-driven architecture — all
+  rules (built-in and custom) share the same `_detect_custom` code path
+- Built-in detectors replaced with declarative `ast_pattern` expressions using
+  the existing prefix syntax (`function_call:`, `decorator:`, `if_name_main`,
+  `file_match:`, `test_file`)
+- `ast_pattern` supports OR combinations with ` | ` separator (e.g.
+  `class_instantiation:FastAPI | function_call:uvicorn.run`)
+- `ast_pattern` field is no longer required for entry rules — rules without
+  it are silently skipped instead of failing validation
+- Decorator matching uses `fnmatch` instead of substring `in`, providing
+  precise glob matching and eliminating false positives from partial name
+  overlaps
+- `function_call:` and `class_instantiation:` patterns now support `fnmatch`
+  globs (was exact match only)
+
+### Fixed
+- `file_match:` pattern now correctly checks the file path at rule level
+  instead of a non-existent `node.filename` attribute, fixing the bug where
+  `file_match:` rules never matched
+- `graphlint config copy-from` no longer picks up irrelevant query/analyzer
+  parameters, fixing the config copy command exception
+
+## [0.1.10] - 2026-07-09
+
+### Fixed
+- Assignment target context resolution now correctly distinguishes class fields
+  (`self.xxx` / `cls.xxx`), method-local variables, and module-level variables,
+  preventing misclassification of `self.x = y` inside methods as local variables
+- Class fields assigned across multiple methods (e.g. `__init__` + `set_x`) are
+  now deduplicated to a single node, eliminating duplicate field nodes
+
+### Changed
+- Attribute-write edges (`self.xxx = value`) now also emit a `read` edge on
+  the parent object (`self`), ensuring proper usage tracking of the containing
+  class instance
+
+## [0.1.9] - 2026-07-08
+
+### Fixed
+- FOREIGN KEY constraint failure on incremental rebuilds — prebuilt edge
+  remapping now correctly writes remapped node IDs into edge objects
+- `_insert_nodes` node-to-file key collision when multiple changed files
+  share the same qualified name and line number
+
+### Changed
+- Incremental rebuild now parses only changed files instead of all files
+- Filesystem scan merged into a single `os.walk` pass instead of two
+- Redundant file hash computation eliminated in both full and incremental paths
+
+### Added
+- `component_members` table in SQLite schema for future incremental
+  connectivity analysis
+
+## [0.1.8] - 2026-07-07
+
+### Added
+- Agent prompt auto-update: `install` now scans all installed prompts on start,
+  replaces outdated content when its version is below the library version,
+  and prints the list of updated agents
+- `install` / `uninstall` interactive selection: support Ctrl+C to abort and
+  empty input to cancel; prompt texts are now i18n-localized
+- `graphlint prompt` subcommand to copy the agent prompt to the system clipboard
+- Documentation for agent integration (`docs/*/guide/agent-integration.md`)
+
+### Fixed
+- Query deadlock: `_build_edge_mapping` now avoids `result` future list
+  mutation under concurrent executor shutdown, preventing infinite waits
+
+## [0.1.7] - 2026-07-06
+
+### Fixed
+- `_resolve_symbol` scope filtering no longer drops cross-class `call` edges
+  when a method resolves a same-named call via a different class (registry
+  / dispatch pattern) — all suffix-matched candidates are now included
+- Entry point `file_pattern` matching (`**/*.py` via fnmatch) now also
+  matches root-level Python files, not only files in subdirectories
+
+## [0.1.6] - 2026-07-05
+
+### Changed
+- `ThreadPoolExecutor` replaced with `ProcessPoolExecutor` for edge building
+- Incremental builds on file change now trigger a full rebuild instead of loading prebuilt edges
+- Module-level usage expansion in connected-component analysis uses a Counter to distinguish real references from synthetic edges
+- Dead-code merge loop uses O(1) `comp_by_id` dict lookup instead of O(C) linear scan
+
+### Removed
+- `_quick_changed_check()` — mtime-based fast-path stamp was unreliable on Windows NTFS and could trigger unnecessary incremental builds that produced stale edges
+- Ghost-node insertion workaround and stale-edge skip logic in `_do_insert_edges`
+
+## [0.1.5] - 2026-07-05
+
+### Added
+- `global` keyword support: `visit_Global` handler tracks names declared as `global X` and resolves references to module-level qualified names (`module.X`) instead of local ones, preventing false local node creation
+- `visit_AugAssign` handler for augmented assignment (`X += 1`, `X *= 2`, etc.), generating both read and write references to the target
+- Scope isolation for `global` declarations — `_global_names` is saved/restored on function entry/exit, preventing leakage between functions
+
+### Fixed
+- `global X` in functions/methods no longer creates spurious local variable nodes for globally-declared names
+- Augmented assignment (`X += 1`) was silently dropped during AST traversal — now correctly produces both read and write edges
+
+## [0.1.4] - 2026-07-04
+
+### Fixed
+- Module-level dead code detection: removed synthetic `read` edges from pseudo-node 0 that masked all top-level symbols as "used", enabling detection of genuinely dead module-level functions and variables
+- Module-level write references no longer create unresolvable edges — variable definitions are not treated as write operations, preventing silent edge drops
+- Module-level read/call references now correctly create edges from pseudo-node 0, fixing a false positive where module-level used variables (e.g., `WARN_TYPE_VALUES` read by `VALID_WARN_TYPES = WARN_TYPE_VALUES`) were flagged as dead code
+- Public API dunders (`__all__`, `__version__`, etc.) no longer appear as isolated components — they stay anchored to their file via synthetic containment edges
+- Python special methods (`__init__`, `__enter__`, `__exit__`, etc.) no longer generate redundant dead-code warnings when their parent class is already flagged as unreachable
+- Unreachable classes with special method overloads now produce only the class-level `dead_code` warning, not per-method duplicates
+
+## [0.1.3] - 2026-07-04
+
+### Added
+- I18n support for CLI parameter help text: `PARAM_DEFS` help strings now go through the i18n system via `help_key`, enabling full language switching for `--help` output
+
+### Changed
+- `ParamDef` dataclass: added `help_key` field for i18n key lookup
+- `cli.py:_add_arg()` now resolves help text via `_t(help_key)` when available
+- Updated `en.py` and `zh_CN.py` with all `help.param.*` translation entries
+
+## [0.1.2] - 2026-07-04
+
+### Added
+- SQLite composite indexes (`idx_edges_source_type`, `idx_edges_target_type`) to accelerate edge-type filtered lookups
+- SQLite single-column index (`idx_warnings_node`) to speed up warnings JOIN queries
+- SQLite sorted indexes (`idx_snapshots_warnings`, `idx_snapshots_nodes`) for faster list_graphs pagination
+- PRAGMA optimizations on Database init: `cache_size=-8000` (8 MB), `mmap_size=268435456` (256 MB), `temp_store=MEMORY`
+- `_quick_changed_check()` — mtime-based fast-path stamp that short-circuits the entire build when no files have changed
+- `_update_scan_stamp()` — persists `{path: mtime_ns}` snapshot inside IndexLock after each successful build
+- `warnings_summary` cache in `QueryEngine.list_graphs()` — lazily computed then shallow-copied, eliminating repeated `GROUP BY` scans
+- `_precompute_edge_counts()` — single-pass edge count per component, replacing O(n × m) per-component iteration
+
+### Changed
+- Eliminated second AST traversal: edge building is now driven by `ReferenceInfo` collected during the single `ASTVisitor` pass — removed `_walk`, `_read_edges`, `_proc_call`, `_proc_assign`, `_proc_annassign`, `_target_ids`, `_read_target_expr` (~400 lines). In a large codebase (700+ `.py` files, 1000+ classes, 14,000+ functions), full rebuild time improved from ~2500s to ~200s.
+- `_build_file_edges_worker` now consumes `ParseResult.references` directly instead of re-walking the AST tree
+- Module pseudo-node (id=0) connectivity tightened: liveness only propagates through `read`/`call` edges, not `write` edges — preventing false reachability of write-only variables and fixing false dead-code reports for module-level used variables
+- Connected component BFS no longer unconditionally seeds node 0, avoiding spurious reachability
+- Module-level synthetic edges recreated for all files each build instead of only for unchanged files
+- Optimized multiple AST I/O bottlenecks in source file reading and tree reuse
+- `_symbol_index` / `_suffix_index` from `dict` to `defaultdict(list)`, removing all `.setdefault()` calls
+- `_resolve_symbol` now accepts a per-worker `resolve_cache` dict — repeated `(qname, scope)` pairs are resolved once per file
+- `call_graph` prebuilt once in `find_connected_components` and reused across `compute_entry_reachability` calls, eliminating redundant edge re-traversal
+- `_insert_nodes` uses a pre-built `(qualified_name, line_start) → file_path` map instead of O(n²) linear scan via `_node_path()`
+- Full rebuild (`incremental=False`) now uses per-table `DELETE FROM` instead of per-file `_delete_old` + selective cleanup
+- Incremental build edge insertion now uses global `DELETE FROM edges` instead of per-file partial delete, eliminating dangling old node-ID references
+- `load_prebuilt_edges` accepts optional `sql_to_mem` mapping, simplifying caller in `indexer.py`
+- Agent injection prompt strengthened to warn when graphlint CLI is unavailable
+
+### Removed
+- `_node_path()` helper function (replaced by pre-built node-to-file map)
+
+### Fixed
+- Windows compatibility: `os.unlink()` with `missing_ok=True` replaced by `if exists()` guard
+
+## [0.1.1] - 2026-07-04
+
+### Added
+- Parallel edge building via ThreadPoolExecutor for changed files (`performance.parallel_workers`)
+- Multi-level suffix index for improved symbol resolution beyond single-level short names
+- Auto-build retry with full index rebuild on FOREIGN KEY constraint failure
+
+### Changed
+- Agent injection prompt wording strengthened to "Always use" with fallback instruction when CLI is unavailable
+- Entry detection methods now reuse pre-parsed source (`pr.source`) to reduce redundant file I/O
+- Refactored AST walk functions to module level to enable parallel execution
+
+### Fixed
+- FOREIGN KEY constraint violations in incremental mode by sorting nodes parent-first before DB insert
+- Incremental rebuild no longer rebuilds edges for unchanged files
+
+## [0.1.0] - 2026-07-01
+
+### Added
+- Initial release
+- AST-based Python source code parsing (classes, functions, methods, variables, fields)
+- Dependency graph construction with 5 edge types (read, write, call, inherit, decorate)
+- 8 built-in entry point detection rules (main, FastAPI, Flask, Django, Click, Typer, Celery, pytest)
+- Custom entry point rules via AST pattern matching
+- 11 warning types (unused_import, dynamic_import, circular_ref, syntax_error, write_only, deprecated_usage, dead_code, type_mismatch, unresolved_ref, unused_variable, file_too_large)
+- Incremental indexing with SHA256-based change detection
+- Adaptive output volume strategy (full, index, truncated)
+- Internationalization (English and Simplified Chinese)
+- CLI interface (`graphlint query`, `graphlint build`, `graphlint config`)
+- Python API (`query()`, `build()`, `configure()`)
+- SQLite persistence with cross-platform file locking
+- Parallel file parsing via ProcessPoolExecutor
+- Configuration management with dot-notation key access
+- Comprehensive test suite (unit, integration, performance)

{graphlint-0.3.0/graphlint.egg-info → graphlint-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: graphlint
-Version: 0.3.0
+Version: 0.3.2
 Summary: A static analysis tool for identifying dead code in a codebase. It provides Python API and CLI for use by agents.
 Author-email: Yutong Zou <yutong.zou.24@alumni.ucl.ac.uk>
 License-Expression: MIT
@@ -28,7 +28,7 @@ Provides-Extra: dev
 Requires-Dist: pytest>=7; extra == "dev"
 Requires-Dist: pytest-cov>=4; extra == "dev"
 Provides-Extra: rust
-Requires-Dist: tree-sitter<0.25,>=0.22; extra == "rust"
+Requires-Dist: tree-sitter<0.27,>=0.22; extra == "rust"
 Requires-Dist: tree-sitter-rust<0.25,>=0.22; extra == "rust"
 Dynamic: license-file
 
@@ -64,6 +64,7 @@ Install with Rust support: `pip install graphlint[rust]` (adds `tree-sitter` and
 - **Configurable entry templates** — add custom entry rules via `ast_pattern` prefixes including `function_call:`, `function_def:`, `decorator:`, `file_match:`, `visibility:pub` (Rust), `trait_impl:` (Rust), `macro_def:` (Rust), and more
 - **`--public-as-entry` flag** — treat all public items (Rust `pub`) as entry points for library-crate analysis
 - **Warning detection** — 11 warning types including circular references, unused imports, write-only variables, and more
+- **Incremental updates** — after initial full scan, only changed files are re-indexed; subsequent queries leverage cached data to minimize computation overhead
 - **Python API + CLI** — integrate into any Tool, CI pipeline, or let agents self-analyze and self-clean
 
 ## Installation
@@ -220,7 +221,7 @@ Full documentation is available in the [docs/](https://github.com/AngelosZou/gra
 - **Python dynamic imports** — due to Python's dynamic import mechanisms (`importlib`, `getattr`, metaclasses, etc.), the default entry templates may produce false positives in codebases that rely heavily on runtime dispatch. Users should tune the `entry_rules` configuration to match their project's conventions.
 - **Rust macro expansion** — tree-sitter parses unexpanded source; procedural macros and `macro_rules!` bodies appear as opaque token trees. Some macro-generated call paths may be missed. `#[derive]` attributes are partially recognized via implicit `inherit` edges.
 - **`--public-as-entry` scope** — this flag only applies to languages with `public` visibility declarations (Rust `pub`). It has no effect on Python files. Toggling this flag triggers a full re-index. For long-term library-crate analysis, prefer enabling the `rust_pub_api` entry rule via `graphlint config` to persist the setting.
-- **Large codebase build time** — on a large codebase with 700+ `.py` files, 1,000+ classes, and 14,000+ functions, a full rebuild takes approximately 200 seconds (actual performance depends on hardware). Small projects (~60 files) complete in ~1 second. **Best practice:** run `query` before making changes to plan your work, and avoid invoking `query` during refactoring to prevent unnecessary index rebuilds.
+- **Large codebase build time** — on a large codebase with 700+ `.py` files, 1,000+ classes, and 14,000+ functions, a full rebuild takes approximately 200 seconds (actual performance depends on hardware). Small projects (~60 files) complete in ~1 second. This cost is one-time, after the initial full scan, subsequent queries use incremental updates.
 
 ## License