code-graph-rag 0.0.820__tar.gz → 0.0.845__tar.gz

{code_graph_rag-0.0.820/code_graph_rag.egg-info → code_graph_rag-0.0.845}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-graph-rag
-Version: 0.0.820
+Version: 0.0.845
 Summary: The ultimate RAG for your monorepo. Query, understand, and edit multi-language codebases with the power of AI and knowledge graphs
 License-Expression: MIT
 Keywords: rag,retrieval-augmented-generation,knowledge-graph,code-analysis,tree-sitter,mcp,mcp-server,llm,graph-database,semantic-search,codebase,memgraph,developer-tools,monorepo

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845}/README.md

@@ -51,9 +51,12 @@ Code-Graph-RAG parses a multi-language codebase with Tree-sitter, builds a knowl
 ## Latest News 🔥
 
 <!-- SECTION:latest_news -->
-- **Graph Indexing**: The name property is now indexed for improved graph read-path lookups. SQL routines are also indexed to resolve calls that name their target in a string.
-- **Duplicate Code Detection**: An AST-based duplicate code detection system has been introduced, alongside fixes for duplicate detection across multiple repositories.
-- **Agentic QA**: A new agentic QA benchmark harness and indexing-time benchmark have been added.
+- **Index Consistency**: Ensure incremental deletes match clean index states.
+- **Patch Reporting**: Distinguish between unverifiable and verified patches in reports.
+- **Windows Teardown**: Clear the read-only bit when removing a temp repo for successful Windows teardown.
+- **TypeScript Grading**: Grade TypeScript inheritance against the tsc oracle for improved accuracy.
+- **Exclude Set Handling**: Record the exclusion set to ensure `--exclude` flags are correctly applied across runs.
+- **MCP Indexing**: Honour `.cgrignore` and `.gitignore` in MCP index and update processes, matching the CLI.
 <!-- /SECTION:latest_news -->
 
 See [NEWS.md](NEWS.md) for the full history.

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845/code_graph_rag.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-graph-rag
-Version: 0.0.820
+Version: 0.0.845
 Summary: The ultimate RAG for your monorepo. Query, understand, and edit multi-language codebases with the power of AI and knowledge graphs
 License-Expression: MIT
 Keywords: rag,retrieval-augmented-generation,knowledge-graph,code-analysis,tree-sitter,mcp,mcp-server,llm,graph-database,semantic-search,codebase,memgraph,developer-tools,monorepo

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845}/codebase_rag/cli.py

@@ -406,6 +406,7 @@ def _delete_hash_cache(repo_path: Path) -> None:
         cache_path.unlink(missing_ok=True)
     (repo_path / cs.DIR_MTIMES_FILENAME).unlink(missing_ok=True)
     (repo_path / cs.PARSER_FINGERPRINT_FILENAME).unlink(missing_ok=True)
+    (repo_path / cs.EXCLUSION_STATE_FILENAME).unlink(missing_ok=True)
 
 
 def _resolve_and_validate_repo(repo_path: str | None) -> Path:

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845}/codebase_rag/constants/cli.py

@@ -494,6 +494,23 @@ PATCH_NOT_AN_IDENTIFIER = "{path}:{line}:{col} is not a whole identifier"
 PATCH_PARSE_FAILED = "{path} no longer parses after the patch"
 PATCH_FORMAT_DRIFT = "{path} applied, but {tool} would reformat it"
 PATCH_OK = "{path}: {count} edit(s) applied"
+# `parses is None` means no grammar was available, so the patch was checked
+# by nothing. Reported apart from PATCH_OK because reporting both as OK is
+# what turned an unverifiable write into an apparently verified one; the
+# write itself is still allowed, since refusing would make edits a silent
+# no-op on a base install where Rust and Go grammars are absent (#1580).
+# The one wording for "checked by nothing", shared so the two messages that
+# carry it cannot drift apart and a caller can match on either.
+PATCH_UNVERIFIED_FRAGMENT = "unverified (no parser for it)"
+PATCH_UNVERIFIED = "{path}: {count} edit(s) applied, " + PATCH_UNVERIFIED_FRAGMENT
+# Both facts, because either alone is misleading here: drift on its own reads
+# as "parsed fine, just reformat it", and a language with no grammar is
+# exactly the case that also has a formatter installed (Rust, Go).
+PATCH_UNVERIFIED_DRIFT = (
+    "{path}: {count} edit(s) applied, "
+    + PATCH_UNVERIFIED_FRAGMENT
+    + "; {tool} would also reformat it"
+)
 # Edit transactions (issue #1528).
 EDIT_NOT_A_FILE = "Staged path is not a regular file: {path}"
 EDIT_RESERVED_PATH = "Path is cgr state, not part of the tree: {path}"

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845}/codebase_rag/constants/core.py

@@ -129,6 +129,12 @@ HASH_CACHE_FILENAME = ".cgr-hash-cache.json"
 DIR_MTIMES_FILENAME = ".cgr-dir-mtimes.json"
 PARSER_FINGERPRINT_FILENAME = ".cgr-parser-fingerprint"
 DELOMBOK_STATE_FILENAME = ".cgr-delombok-state.json"
+# The exclusion set the last run indexed under, covering both the excludes and
+# the unignores (which come from `!` lines in .cgrignore/.gitignore and from
+# interactive setup; there is no --unignore flag). Nothing on disk changes when
+# only the CLI --exclude flags do, so without this the sync check cannot tell
+# that the eligible set moved (issue #1606).
+EXCLUSION_STATE_FILENAME = ".cgr-exclusion-state.json"
 # Recorded edit transactions for `cgr edits show|undo` (issue #1528).
 EDIT_HISTORY_FILENAME = ".cgr-edit-history.json"
 EDIT_LOCK_FILENAME = ".cgr-edit-lock"
@@ -144,6 +150,7 @@ CGR_STATE_FILENAMES: frozenset[str] = frozenset(
         DIR_MTIMES_FILENAME,
         PARSER_FINGERPRINT_FILENAME,
         DELOMBOK_STATE_FILENAME,
+        EXCLUSION_STATE_FILENAME,
         EDIT_HISTORY_FILENAME,
         EDIT_LOCK_FILENAME,
     }

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845}/codebase_rag/constants/graph.py

@@ -443,6 +443,7 @@ CYPHER_DELETE_MODULE = (
 # projects, and a path-only delete would take the sibling's node (issue #897).
 CYPHER_DELETE_FILE = "MATCH (f:File {absolute_path: $path}) DETACH DELETE f"
 CYPHER_DELETE_FOLDER = "MATCH (f:Folder {absolute_path: $path}) DETACH DELETE f"
+CYPHER_DELETE_PACKAGE = "MATCH (p:Package {absolute_path: $path}) DETACH DELETE p"
 CYPHER_DELETE_CALLS = "MATCH ()-[r:CALLS]->() DELETE r"
 # Removes external import-target Module nodes that no module imports anymore
 # (e.g. an imported name that was renamed/removed on an incremental rebuild).
@@ -487,6 +488,13 @@ CYPHER_ALL_MODULE_PATHS_INTERNAL = (
 CYPHER_ALL_FOLDER_PATHS = (
     "MATCH (f:Folder) RETURN f.path AS path, f.absolute_path AS absolute_path"
 )
+# Package nodes are pruned against the CURRENT structure, not the disk: a
+# directory whose indicator file (__init__.py, Cargo.toml, ...) went away
+# still exists, as a Folder (issue #1570).
+CYPHER_ALL_PACKAGE_PATHS = (
+    "MATCH (p:Package) RETURN p.path AS path, p.absolute_path AS absolute_path, "
+    "p.qualified_name AS qualified_name"
+)
 
 # Rehydrate the in-memory function registry on an incremental run: returns
 # every definition node's qualified name and label so call/instantiation

{code_graph_rag-0.0.820 → code_graph_rag-0.0.845}/codebase_rag/constants/languages.py

@@ -445,9 +445,33 @@ IGNORE_PATTERNS = frozenset(
         "venv",
     }
 )
+# Machine-generated artefacts: never hand-edited, never first-party. Matched
+# against the whole FILENAME with `str.endswith`, not against `Path.suffix`,
+# because two entries here are not suffixes in pathlib's sense:
+# `Path("jquery.min.js").suffix` is ".js" and `Path("notes.py~").suffix` is
+# ".py~", so a `Path.suffix` test silently matched neither (issue #1636).
+# Use `path_utils.is_ignored_filename`; do not re-derive the rule.
 IGNORE_SUFFIXES = frozenset(
-    {".tmp", "~", ".pyc", ".pyo", ".o", ".a", ".so", ".dll", ".class"}
+    {
+        ".tmp",
+        "~",
+        ".pyc",
+        ".pyo",
+        ".o",
+        ".a",
+        ".so",
+        ".dll",
+        ".class",
+        # Vendored bundles that generated API docs (jazzy, YARD, JSDoc,
+        # Sphinx) ship alongside the source they document. Their symbols are
+        # whatever the minifier emitted, and they outnumber the real ones.
+        ".min.js",
+        ".min.css",
+    }
 )
+# `str.endswith` takes a tuple, and the repository walk tests every filename,
+# so build it once rather than per file. Derived, never edited separately.
+IGNORE_FILENAME_ENDINGS = tuple(sorted(IGNORE_SUFFIXES))
 
 # pathspec style for .cgrignore / --exclude patterns (#495).
 GITWILDMATCH_STYLE = "gitignore"