codegraphcontext 0.6.1__tar.gz → 0.6.2__tar.gz

{codegraphcontext-0.6.1/src/codegraphcontext.egg-info → codegraphcontext-0.6.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codegraphcontext
-Version: 0.6.1
+Version: 0.6.2
 Summary: An MCP server that indexes local code into a graph database to provide context to AI assistants.
 Author-email: Shashank Shekhar Singh <shashankshekharsingh1205@gmail.com>
 License: MIT License

{codegraphcontext-0.6.1 → codegraphcontext-0.6.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "codegraphcontext"
-version = "0.6.1"
+version = "0.6.2"
 description = "An MCP server that indexes local code into a graph database to provide context to AI assistants."
 authors = [{ name = "Shashank Shekhar Singh", email = "shashankshekharsingh1205@gmail.com" }]
 readme = "README.md"

{codegraphcontext-0.6.1 → codegraphcontext-0.6.2}/src/codegraphcontext/core/database_embedded_kuzu.py

@@ -798,21 +798,32 @@ class EmbeddedSessionWrapper:
         translated_query, translated_params = self._translate_query(query, parameters)
         debug_log(f"Translated Query: {translated_query[:200]}")
         try:
-            # Force loop fallback for relationship writes inside UNWIND to avoid Kuzu query planner bugs
-            # which can incorrectly bind/corrupt relationship endpoints across rows in the batch.
-            # Only force it for `UNWIND $param AS row`, the single shape the
-            # recovery path below can rewrite into a per-row loop. Read queries
-            # that unwind a bound list (`WITH relationships(p) AS rels UNWIND
-            # rels AS r`, used by find_all_callers/find_all_callees and the
-            # visualizer) matched this guard too, but no `$`-UNWIND for the
-            # fallback to latch onto — so the fabricated exception fell through
-            # every handler and surfaced to the caller as a hard failure.
+            # Force the per-row loop fallback ONLY for UNWIND batches containing a
+            # node-MERGE. Kùzu's MERGE pipeline mis-binds when a merged node's key
+            # value repeats NON-adjacently across the batch: in
+            # [(A,P), (B,X), (C,P)] the C row binds to X's node instead of
+            # re-matching P (reproduced on Kùzu 0.11.3; see #1605). Relationship-only
+            # MERGEs (MATCH … MATCH … MERGE (a)-[r]->(b)) do not exhibit the bug —
+            # verified with interleaved duplicate endpoint keys and duplicate pairs —
+            # so they now run batched, which removes the bulk of the per-row planner
+            # overhead (~2.4x on a full index of the Python sample project).
+            # Only `UNWIND $param AS row` shapes are guarded — that is the single
+            # shape the recovery path below can rewrite into a per-row loop. Read
+            # queries that unwind a bound list (`WITH relationships(p) AS rels
+            # UNWIND rels AS r`) must not match, or the fabricated exception would
+            # surface to the caller as a hard failure.
+            # Node-only UNWIND writes (`MERGE (n:Label {…}) SET n += row`, no
+            # relationship pattern) have always batched through the SET-expansion
+            # path and the per-row rewriter cannot handle their bare `SET n += row`,
+            # so the guard additionally requires a relationship pattern.
+            _has_node_merge = re.search(r"MERGE\s*\(\s*\w+\s*:", query)
             if (
                 re.search(r"UNWIND\s+\$\w+\s+AS\s+\w+", query)
                 and ("-[" in query or "]->" in query)
+                and _has_node_merge
                 and not getattr(self, "_skip_unwind_fallback", False)
             ):
-                raise Exception("unordered_map::at (forced fallback to avoid relationship UNWIND planner bugs)")
+                raise Exception("unordered_map::at (forced fallback: node-MERGE inside UNWIND mis-binds repeated keys)")
 
             # 2. Execute under the lock. _write_lock (name kept for backward
             # compat) now serializes ALL access, reads included: kuzu.Connection

{codegraphcontext-0.6.1 → codegraphcontext-0.6.2}/src/codegraphcontext/tools/indexing/persistence/writer.py

@@ -1138,8 +1138,23 @@ class GraphWriter:
         batch_size = 500
         backend = get_backend_type(self.driver, self._db_manager)
         def _work(session):
-            internal_batch = [r for r in inheritance_batch if r.get("resolved_parent_file_path") != "__external__"]
-            external_batch = [r for r in inheritance_batch if r.get("resolved_parent_file_path") == "__external__"]
+            # Dedupe identical rows first: parsers can emit the same
+            # (child, parent) record more than once, and while Neo4j's MERGE
+            # absorbs the repeat, the embedded backends' rewritten
+            # node-MERGE+rel-MERGE shape has been observed writing a duplicate
+            # INHERITS edge for it. Deduping is correct on every backend.
+            seen_rows: set = set()
+            deduped_batch: List[Dict[str, Any]] = []
+            for r in inheritance_batch:
+                key = (r.get("child_name"), r.get("path"), r.get("parent_name"),
+                       r.get("resolved_parent_file_path"))
+                if key in seen_rows:
+                    continue
+                seen_rows.add(key)
+                deduped_batch.append(r)
+
+            internal_batch = [r for r in deduped_batch if r.get("resolved_parent_file_path") != "__external__"]
+            external_batch = [r for r in deduped_batch if r.get("resolved_parent_file_path") == "__external__"]
 
             labels = ("Class", "Trait", "Interface", "Struct", "Enum", "Union", "Record", "Mixin", "Extension", "Module", "Object", "Variable")
 

{codegraphcontext-0.6.1 → codegraphcontext-0.6.2}/src/codegraphcontext/tools/indexing/pipeline.py

@@ -201,10 +201,32 @@ async def run_tree_sitter_index_async(
             # it to this file's path string turned a single recoverable write
             # failure into an AttributeError that aborted the whole job.
             failed_path = file_data.get("path")
-            write_failures.append({"path": failed_path, "error": str(exc)})
-            error_logger(f"Failed to write {failed_path} to the graph: {exc}")
-            file_data["error"] = str(exc)
-            file_data["parse_failed"] = True
+            # Retry once before giving up: writes are MERGE-idempotent, and a
+            # transient failure under runner load silently costs the graph the
+            # whole file's edges (observed as LadybugDB intermittently writing
+            # ~51 fewer CONTAINS edges in the parity run, #1612).
+            retried_ok = False
+            if "error" not in file_data:
+                try:
+                    await asyncio.sleep(0.2)
+                    await asyncio.to_thread(
+                        writer.add_file_to_graph,
+                        file_data,
+                        repo_name,
+                        imports_map,
+                        repo_path_str=resolved_repo_path_str,
+                    )
+                    retried_ok = True
+                    warning_logger(
+                        f"Write for {failed_path} succeeded on retry after: {exc}"
+                    )
+                except Exception as retry_exc:  # noqa: BLE001
+                    exc = retry_exc
+            if not retried_ok:
+                write_failures.append({"path": failed_path, "error": str(exc)})
+                error_logger(f"Failed to write {failed_path} to the graph: {exc}")
+                file_data["error"] = str(exc)
+                file_data["parse_failed"] = True
 
     if write_failures:
         warning_logger(

{codegraphcontext-0.6.1 → codegraphcontext-0.6.2}/src/codegraphcontext/tools/languages/cpp.py

@@ -519,11 +519,7 @@ class CppTreeSitterParser:
                 if lambda_node is None or lambda_node.type != 'lambda_expression':
                     continue
 
-                params_node = lambda_node.child_by_field_name('declarator')
-                if params_node:
-                    params_node = params_node.child_by_field_name('parameters')
                 name = self._get_node_text(node)
-                params_node = lambda_node.child_by_field_name('parameters')
 
                 context, context_type, _ = self._get_parent_context(assignment_node)
                 class_context, _, _ = self._get_parent_context(assignment_node, types=('class_specifier',))
@@ -532,7 +528,15 @@ class CppTreeSitterParser:
                     "name": name,
                     "line_number": node.start_point[0] + 1,
                     "end_line": assignment_node.end_point[0] + 1,
-                    "args": [p for p in [self._get_node_text(p) for p in params_node.children if p.type == 'identifier'] if p] if params_node else [],
+                    # lambda_expression carries the same declarator→parameters
+                    # field chain as a function_definition, so the shared
+                    # extractor handles parameter_declaration unwrapping
+                    # (pointers/refs included). The old inline version first
+                    # clobbered its correctly-walked parameter_list with a
+                    # nonexistent field and then filtered for bare identifiers
+                    # over parameter_declaration nodes — every lambda got
+                    # args: [] (#1527, case 5).
+                    "args": self._extract_function_params(lambda_node),
                     
                     "docstring": None,
                     "cyclomatic_complexity": 1,

{codegraphcontext-0.6.1 → codegraphcontext-0.6.2/src/codegraphcontext.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codegraphcontext
-Version: 0.6.1
+Version: 0.6.2
 Summary: An MCP server that indexes local code into a graph database to provide context to AI assistants.
 Author-email: Shashank Shekhar Singh <shashankshekharsingh1205@gmail.com>
 License: MIT License