codespine 0.9.3__tar.gz → 0.9.5__tar.gz

{codespine-0.9.3 → codespine-0.9.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codespine
-Version: 0.9.3
+Version: 0.9.5
 Summary: Local Java code intelligence indexer backed by a graph database
 Author: CodeSpine contributors
 License: MIT License

{codespine-0.9.3 → codespine-0.9.5}/codespine/__init__.py

@@ -1,4 +1,4 @@
 """CodeSpine package."""
 
 __all__ = ["__version__"]
-__version__ = "0.9.3"
+__version__ = "0.9.5"

{codespine-0.9.3 → codespine-0.9.5}/codespine/db/store.py

@@ -527,15 +527,27 @@ class GraphStore:
         rows = [{"source_id": r["source_id"], "target_id": r["target_id"],
                   "confidence": float(r["confidence"]), "reason": r["reason"]}
                  for r in records]
-        op = "CREATE" if create_mode else "MERGE"
-        self.execute(
-            f"""
-            UNWIND $rows AS row
-            MATCH (src:Method {{id: row.source_id}}), (dst:Method {{id: row.target_id}})
-            {op} (src)-[:CALLS {{confidence: row.confidence, reason: row.reason}}]->(dst)
-            """,
-            {"rows": rows},
-        )
+        if create_mode:
+            self.execute(
+                """
+                UNWIND $rows AS row
+                MATCH (src:Method {id: row.source_id}), (dst:Method {id: row.target_id})
+                CREATE (src)-[:CALLS {confidence: row.confidence, reason: row.reason}]->(dst)
+                """,
+                {"rows": rows},
+            )
+        else:
+            # Properties are SET, not part of the MERGE pattern — ensures at most
+            # one CALLS edge per (src, dst) pair regardless of confidence value.
+            self.execute(
+                """
+                UNWIND $rows AS row
+                MATCH (src:Method {id: row.source_id}), (dst:Method {id: row.target_id})
+                MERGE (src)-[r:CALLS]->(dst)
+                SET r.confidence = row.confidence, r.reason = row.reason
+                """,
+                {"rows": rows},
+            )
 
     def add_reference(self, rel: str, src_label: str, src_id: str, dst_label: str, dst_id: str, confidence: float) -> None:
         if rel not in {"REFERENCES_TYPE", "IMPLEMENTS", "OVERRIDES"}:
@@ -671,28 +683,30 @@ class GraphStore:
             self.clear_file(f_id)
         self._recycle_conn()
 
-        # 2–5. Write all nodes with CREATE (clear_file above guarantees absence).
+        # 2. Upsert file record
         with self.transaction():
             self.upsert_files_batch(
                 [{"id": f_id, "path": path, "project_id": project_id,
                   "is_test": is_test, "hash": digest}],
-                create_mode=True,
             )
         self._recycle_conn()
 
+        # 3. Upsert classes (typically very few per file)
         if classes:
             with self.transaction():
-                self.upsert_classes_batch(classes, create_mode=True)
+                self.upsert_classes_batch(classes)
             self._recycle_conn()
 
+        # 4. Upsert methods in sub-batches of 200
         for i in range(0, len(methods), self._FILE_METHOD_SUB_BATCH):
             with self.transaction():
-                self.upsert_methods_batch(methods[i: i + self._FILE_METHOD_SUB_BATCH], create_mode=True)
+                self.upsert_methods_batch(methods[i: i + self._FILE_METHOD_SUB_BATCH])
             self._recycle_conn()
 
+        # 5. Upsert symbols in sub-batches of 200
         for i in range(0, len(symbols), self._FILE_SYMBOL_SUB_BATCH):
             with self.transaction():
-                self.upsert_symbols_batch(symbols[i: i + self._FILE_SYMBOL_SUB_BATCH], create_mode=True)
+                self.upsert_symbols_batch(symbols[i: i + self._FILE_SYMBOL_SUB_BATCH])
             self._recycle_conn()
 
         # 6. Write call edges in sub-batches of 500 (normalise key names to match add_calls_batch)
@@ -705,14 +719,14 @@ class GraphStore:
                 for rec in batch
             ]
             with self.transaction():
-                self.add_calls_batch(normalised, create_mode=True)
+                self.add_calls_batch(normalised)
             self._recycle_conn()
 
         # 7. Write type relations (IMPLEMENTS, OVERRIDES, REFERENCES_TYPE)
         for i in range(0, len(type_rels), self._FILE_REL_SUB_BATCH):
             batch = type_rels[i: i + self._FILE_REL_SUB_BATCH]
             with self.transaction():
-                self.add_references_batch(batch, create_mode=True)
+                self.add_references_batch(batch)
             self._recycle_conn()
 
     def clear_file_by_path(self, project_id: str, project_path: str, file_path: str) -> None:

{codespine-0.9.3 → codespine-0.9.5}/codespine/indexer/call_resolver.py

@@ -5,7 +5,7 @@ from typing import Iterator
 
 from codespine.noise.blocklist import MIN_FUZZY_NAME_LEN, NOISE_METHOD_NAMES
 
-MAX_FUZZY_TARGETS = 12
+MAX_FUZZY_TARGETS = 6  # reduced from 12 — keeps precision, halves low-confidence edge fan-out
 
 
 def _simple_type_name(type_name: str | None) -> str:

{codespine-0.9.3 → codespine-0.9.5}/codespine/indexer/engine.py

@@ -221,7 +221,7 @@ class JavaIndexer:
         calls_resolved = 0
         type_relationships = 0
         file_batch_size = max(1, int(getattr(SETTINGS, "index_file_batch_size", 64)))
-        edge_batch_size = max(1, int(getattr(SETTINGS, "edge_write_batch_size", 2000)))
+        edge_batch_size = max(1, int(getattr(SETTINGS, "edge_write_batch_size", 5000)))
 
         if not full:
             method_catalog, class_catalog, fqcn_to_class_ids, class_methods = (
@@ -462,42 +462,54 @@ class JavaIndexer:
                     with self.store.transaction():
                         self.store.clear_files_batch(clear_sub)
                     self.store._recycle_conn()
-            # Always CREATE — full clears via clear_project, incremental clears
-            # per-file above, so nodes are guaranteed absent in both paths.
             with self.store.transaction():
-                self.store.upsert_files_batch(file_rows, create_mode=True)
+                self.store.upsert_files_batch(file_rows)
             self.store._recycle_conn()
             with self.store.transaction():
-                self.store.upsert_classes_batch(class_rows, create_mode=True)
+                self.store.upsert_classes_batch(class_rows)
             self.store._recycle_conn()
             _METHOD_SUB_BATCH = 200
             for method_sub in self._chunked(method_rows, _METHOD_SUB_BATCH):
                 with self.store.transaction():
-                    self.store.upsert_methods_batch(method_sub, create_mode=True)
+                    self.store.upsert_methods_batch(method_sub)
                 self.store._recycle_conn()
             _SYMBOL_SUB_BATCH = 200
             for symbol_sub in self._chunked(symbol_rows, _SYMBOL_SUB_BATCH):
                 with self.store.transaction():
-                    self.store.upsert_symbols_batch(symbol_sub, create_mode=True)
+                    self.store.upsert_symbols_batch(symbol_sub)
                 self.store._recycle_conn()
 
         self._emit(progress, "resolve_calls_start")
-        call_rows: list[dict] = []
-        for src, dst, confidence, reason in resolve_calls(method_catalog, method_calls, method_context, class_catalog):
-            call_rows.append(
-                {
-                    "source_id": src,
-                    "target_id": dst,
-                    "confidence": confidence,
-                    "reason": reason,
-                }
+        # Deduplicate (src, dst) pairs — the same pair can appear many times when
+        # a method calls another method multiple times at different call sites.
+        # Keep the highest-confidence resolution to avoid N writes per pair.
+        best_calls: dict[tuple[str, str], tuple[float, str]] = {}
+        for src, dst, confidence, reason in resolve_calls(
+            method_catalog, method_calls, method_context, class_catalog
+        ):
+            key = (src, dst)
+            if key not in best_calls or confidence > best_calls[key][0]:
+                best_calls[key] = (confidence, reason)
+
+        # Stream writes in batches — never hold the full set in RAM.
+        call_buf: list[dict] = []
+        for (src, dst), (confidence, reason) in best_calls.items():
+            call_buf.append(
+                {"source_id": src, "target_id": dst,
+                 "confidence": confidence, "reason": reason}
             )
-        for call_chunk in self._chunked(call_rows, edge_batch_size):
+            if len(call_buf) >= edge_batch_size:
+                with self.store.transaction():
+                    self.store.add_calls_batch(call_buf)
+                calls_resolved += len(call_buf)
+                self.store._recycle_conn()
+                self._emit(progress, "resolve_calls_progress", calls_resolved=calls_resolved)
+                call_buf = []
+        if call_buf:
             with self.store.transaction():
-                self.store.add_calls_batch(call_chunk, create_mode=True)
-            calls_resolved += len(call_chunk)
+                self.store.add_calls_batch(call_buf)
+            calls_resolved += len(call_buf)
             self.store._recycle_conn()
-            self._emit(progress, "resolve_calls_progress", calls_resolved=calls_resolved)
         self._emit(progress, "resolve_calls_done", calls_resolved=calls_resolved)
 
         self._emit(progress, "resolve_types_start")
@@ -509,7 +521,7 @@ class JavaIndexer:
         )
         for rel_chunk in self._chunked(type_rows, edge_batch_size):
             with self.store.transaction():
-                self.store.add_references_batch(rel_chunk, create_mode=True)
+                self.store.add_references_batch(rel_chunk)
             type_relationships += len(rel_chunk)
             self.store._recycle_conn()
         self._emit(progress, "resolve_types_done", type_relationships=type_relationships)

{codespine-0.9.3 → codespine-0.9.5}/codespine.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: codespine
-Version: 0.9.3
+Version: 0.9.5
 Summary: Local Java code intelligence indexer backed by a graph database
 Author: CodeSpine contributors
 License: MIT License

{codespine-0.9.3 → codespine-0.9.5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "codespine"
-version = "0.9.3"
+version = "0.9.5"
 description = "Local Java code intelligence indexer backed by a graph database"
 readme = "README.md"
 requires-python = ">=3.10"