PR-E3: intra-JVM CALLS invariant guard (item 8)

build_ast_graph.py

@@ -145,6 +145,7 @@ class CallResolutionStats:
     phantom_chained: int = 0
     phantom_other: int = 0
     callee_unresolved: int = 0
+    skipped_cross_service: int = 0
 
 
 @dataclass
@@ -252,6 +253,7 @@ class GraphTables:
     methods_by_type: dict[str, list[MemberEntry]] = field(default_factory=dict)
     parse_errors: int = 0
     skipped_files: int = 0
+    pass3_skipped_cross_service: int = 0
 
 
 # ---------- file walk (see `path_filtering.iter_java_source_files`) ----------
@@ -1042,6 +1044,7 @@ def _resolve_and_emit_call(
     recv_type, strat, conf = _resolve_receiver_type(call, scope=scope, member=member, ast=ast, tables=tables)
 
     if strat == "chained_receiver":
+        # Chained-receiver phantoms have no microservice attribution, so they cannot violate cross-service CALLS invariants.
         pid = _phantom_method_id(
             tables, receiver_fqn=None, receiver_expr=call.receiver_expr,
             callee=call.callee_simple, arg_count=call.arg_count,
@@ -1053,6 +1056,7 @@ def _resolve_and_emit_call(
         return
 
     if recv_type is None:
+        # Unresolved-receiver phantoms also carry empty microservice attribution.
         pid = _phantom_method_id(
             tables, receiver_fqn=None, receiver_expr=call.receiver_expr,
             callee=call.callee_simple, arg_count=call.arg_count,
@@ -1067,6 +1071,22 @@ def _resolve_and_emit_call(
         recv_type, call.callee_simple, call.arg_count, tables, ast,
     )
 
+    # Guard relies on `_lookup_method_candidates` returning a same-ms candidate when one exists; revisit if pass3 scopes lookups per-microservice.
+    if member.microservice:
+        same_ms = [c for c in candidates if c.microservice == member.microservice]
+        if same_ms and len(same_ms) != len(candidates):
+            for c in candidates:
+                if c.microservice and c.microservice != member.microservice:
+                    log.warning(
+                        "skipping cross-microservice CALLS edge %s -> %s "
+                        "(caller=%s, callee=%s)",
+                        f"{member.parent_fqn}#{member.decl.signature}",
+                        f"{c.parent_fqn}#{c.decl.signature}",
+                        member.microservice, c.microservice,
+                    )
+                    stats.skipped_cross_service += 1
+            candidates = same_ms
+
     # Compute the call-shape strategy / confidence override BEFORE the
     # empty-candidates check so they are preserved even when the callee cannot
     # be located on the resolved receiver type (B3 fix).
@@ -1106,11 +1126,12 @@ def _resolve_and_emit_call(
         return
 
     if len(candidates) == 1:
+        candidate = candidates[0]
         ref_arity: int | None = None
         if call.arg_count < 0:
-            ref_arity = len(candidates[0].decl.parameters)
+            ref_arity = len(candidate.decl.parameters)
         _emit_call_edge(
-            tables, stats, src_id=member.node_id, dst_id=candidates[0].node_id, call=call,
+            tables, stats, src_id=member.node_id, dst_id=candidate.node_id, call=call,
             confidence=edge_conf, strategy=edge_strat, resolved=True,
             edge_arg_count=ref_arity,
         )
@@ -1165,11 +1186,13 @@ def pass3_calls(tables: GraphTables, asts: dict[str, JavaFileAst], *, verbose: b
     pct_chained = 100.0 * stats.phantom_chained / max(1, stats.total)
     pct_callee_unres = 100.0 * stats.callee_unresolved / max(1, stats.total)
     pct_phantom_recv = 100.0 * stats.phantom_other / max(1, stats.total)
+    tables.pass3_skipped_cross_service = int(stats.skipped_cross_service)
     msg = (
         f"Call resolution: {stats.total} sites, {stats.phantom_chained} chained phantoms "
         f"({pct_chained:.1f}%), {stats.callee_unresolved} unresolved callee "
         f"({pct_callee_unres:.1f}%), {stats.phantom_other} phantom receiver "
-        f"({pct_phantom_recv:.1f}%), strategies: {dict(stats.by_strategy)}"
+        f"({pct_phantom_recv:.1f}%), {stats.skipped_cross_service} skipped cross-service, "
+        f"strategies: {dict(stats.by_strategy)}"
     )
     log.info(msg)
     if verbose:
@@ -1781,7 +1804,8 @@ def _micro_factor(member: MemberEntry | None) -> float:
     "async_producers_from_brownfield_pct DOUBLE, "
     "http_calls_match_breakdown STRING, "
     "async_calls_match_breakdown STRING, "
-    "cross_service_calls_total INT64"
+    "cross_service_calls_total INT64, "
+    "pass3_skipped_cross_service INT64"
     ")"
 )
 
@@ -2158,7 +2182,8 @@ def _write_meta(conn: kuzu.Connection, tables: GraphTables, source_root: Path) -
         "async_producers_from_brownfield_pct: $async_producers_from_brownfield_pct, "
         "http_calls_match_breakdown: $http_calls_match_breakdown, "
         "async_calls_match_breakdown: $async_calls_match_breakdown, "
-        "cross_service_calls_total: $cross_service_calls_total})",
+        "cross_service_calls_total: $cross_service_calls_total, "
+        "pass3_skipped_cross_service: $pass3_skipped_cross_service})",
         {
             "k": "graph",
             "ov": ONTOLOGY_VERSION,
@@ -2183,6 +2208,7 @@ def _write_meta(conn: kuzu.Connection, tables: GraphTables, source_root: Path) -
             "http_calls_match_breakdown": json.dumps(http_match),
             "async_calls_match_breakdown": json.dumps(async_match),
             "cross_service_calls_total": int(call_stats.cross_service_calls_total),
+            "pass3_skipped_cross_service": int(tables.pass3_skipped_cross_service),
         },
     )
 

kuzu_queries.py

@@ -329,7 +329,26 @@ def _rows(self, query: str, params: dict[str, Any] | None = None) -> list[dict[s
     # ---- meta ----
 
     def meta(self) -> dict[str, Any]:
-        _META_FULL = (
+        _META_PR_E3 = (
+            "MATCH (m:GraphMeta) RETURN m.key AS key, m.ontology_version AS ontology_version, "
+            "m.built_at AS built_at, m.source_root AS source_root, "
+            "m.counts_json AS counts_json, m.parse_errors AS parse_errors, "
+            "m.routes_total AS routes_total, m.exposes_total AS exposes_total, "
+            "m.routes_by_framework AS routes_by_framework, "
+            "m.routes_resolved_pct AS routes_resolved_pct, "
+            "m.routes_from_brownfield_pct AS routes_from_brownfield_pct, "
+            "m.routes_by_layer AS routes_by_layer, "
+            "m.http_calls_total AS http_calls_total, m.async_calls_total AS async_calls_total, "
+            "m.http_calls_by_strategy AS http_calls_by_strategy, m.async_calls_by_strategy AS async_calls_by_strategy, "
+            "m.http_calls_resolved_pct AS http_calls_resolved_pct, m.async_calls_resolved_pct AS async_calls_resolved_pct, "
+            "m.http_clients_from_brownfield_pct AS http_clients_from_brownfield_pct, "
+            "m.async_producers_from_brownfield_pct AS async_producers_from_brownfield_pct, "
+            "m.http_calls_match_breakdown AS http_calls_match_breakdown, "
+            "m.async_calls_match_breakdown AS async_calls_match_breakdown, "
+            "m.cross_service_calls_total AS cross_service_calls_total, "
+            "m.pass3_skipped_cross_service AS pass3_skipped_cross_service"
+        )
+        _META_PRE_E3 = (
             "MATCH (m:GraphMeta) RETURN m.key AS key, m.ontology_version AS ontology_version, "
             "m.built_at AS built_at, m.source_root AS source_root, "
             "m.counts_json AS counts_json, m.parse_errors AS parse_errors, "
@@ -361,19 +380,23 @@ def meta(self) -> dict[str, Any]:
             "m.counts_json AS counts_json, m.parse_errors AS parse_errors"
         )
         rows: list[dict[str, Any]]
-        meta_mode = "full"
+        meta_mode = "pr_e3"
         try:
-            rows = self._rows(_META_FULL)
+            rows = self._rows(_META_PR_E3)
         except Exception:
-            meta_mode = "pr_a2"
+            meta_mode = "pre_e3"
             try:
-                rows = self._rows(_META_PR_A2)
+                rows = self._rows(_META_PRE_E3)
             except Exception:
-                meta_mode = "legacy"
+                meta_mode = "pr_a2"
                 try:
-                    rows = self._rows(_META_LEGACY)
-                except Exception as e:
-                    return {"error": f"{e}"}
+                    rows = self._rows(_META_PR_A2)
+                except Exception:
+                    meta_mode = "legacy"
+                    try:
+                        rows = self._rows(_META_LEGACY)
+                    except Exception as e:
+                        return {"error": f"{e}"}
         if not rows:
             return {"error": "no GraphMeta node"}
         row = rows[0]
@@ -398,6 +421,7 @@ def meta(self) -> dict[str, Any]:
         http_calls_match_breakdown: dict[str, Any] = {}
         async_calls_match_breakdown: dict[str, Any] = {}
         cross_service_calls_total = 0
+        pass3_skipped_cross_service = 0
         if meta_mode != "legacy":
             rfw_raw = row.get("routes_by_framework") or "{}"
             try:
@@ -409,7 +433,7 @@ def meta(self) -> dict[str, Any]:
             routes_total = int(row.get("routes_total") or 0)
             exposes_total = int(row.get("exposes_total") or 0)
             routes_resolved_pct = float(row.get("routes_resolved_pct") or 0.0)
-        if meta_mode == "full":
+        if meta_mode in ("pr_e3", "pre_e3"):
             routes_from_brownfield_pct = float(row.get("routes_from_brownfield_pct") or 0.0)
             rbl_raw = row.get("routes_by_layer") or "{}"
             try:
@@ -453,6 +477,7 @@ def meta(self) -> dict[str, Any]:
             if not isinstance(async_calls_match_breakdown, dict):
                 async_calls_match_breakdown = {}
             cross_service_calls_total = int(row.get("cross_service_calls_total") or 0)
+            pass3_skipped_cross_service = int(row.get("pass3_skipped_cross_service") or 0)
         return {
             "ontology_version": int(row.get("ontology_version") or 0),
             "built_at": int(row.get("built_at") or 0),
@@ -476,6 +501,7 @@ def meta(self) -> dict[str, Any]:
             "http_calls_match_breakdown": http_calls_match_breakdown,
             "async_calls_match_breakdown": async_calls_match_breakdown,
             "cross_service_calls_total": cross_service_calls_total,
+            "pass3_skipped_cross_service": pass3_skipped_cross_service,
             "db_path": self.db_path,
         }
 

plans/PLAN-POST-TIER1B-FOLLOWUPS.md

@@ -19,7 +19,7 @@ Links go to the review comment so the original context survives.
 | 5 | [PR-D2 #13](https://github.com/HumanBean17/java-enterprise-codebase-rag/pull/13#issuecomment-4378995637) post-D3 follow-up 1 | README: document `anchor`-fills-from-builtin behaviour for partial brownfield overrides | low (doc) |
 | 6 | [PR-D2 #13](https://github.com/HumanBean17/java-enterprise-codebase-rag/pull/13#issuecomment-4378995637) post-D3 follow-up 2 | Proposal §6: add `channel` field to the `OutgoingCallDecl` schema sketch as durable | low (doc) |
 | 7 | [PR-D1 #12](https://github.com/HumanBean17/java-enterprise-codebase-rag/pull/12#issuecomment-4378723605) obs 2 | Second copy of strategy ladder still in `graph_enrich.py:720-724` (annotation/spel/constant_ref) — known consolidation candidate | medium (tech debt) |
-| 8 | [PR-E1 #19 review reply](https://github.com/HumanBean17/java-enterprise-codebase-rag/pull/19#issuecomment-4380659734) | `pass3_calls` doesn't enforce the intra-JVM invariant for `CALLS` edges. Today no cross-microservice CALLS edge is emitted on any fixture (verified on `cross_service_smoke`: 9 CALLS edges, 0 cross), but the cleanliness is incidental — no `caller.microservice == callee.microservice` guard exists. FQN collisions across services or brownfield supertype overrides could in principle break the invariant silently. | low-to-medium (invariant) |
+| 8 | [PR-E1 #19 review reply](https://github.com/HumanBean17/java-enterprise-codebase-rag/pull/19#issuecomment-4380659734) | `pass3_calls` doesn't enforce the intra-JVM invariant for `CALLS` edges. Today no cross-microservice CALLS edge is emitted on any fixture (verified on `cross_service_smoke`: 9 CALLS edges, 0 cross), but the cleanliness is incidental — no `caller.microservice == callee.microservice` guard exists. FQN collisions across services or brownfield supertype overrides could in principle break the invariant silently. **✅ shipped in PR-E3** | low-to-medium (invariant) |
 
 ## Recommended PR boundaries
 
@@ -313,7 +313,9 @@ is surgical, not over-eager.
    (verified by extending one existing graph_meta assertion in
    `tests/test_kuzu_meta.py` or equivalent).
 5. Pass3 verbose log mentions the new counter.
-6. `260+` (current baseline + 1 new test) tests still pass.
+ 6. `260+` (current baseline + 1 new test) tests still pass.
+
+Status: shipped as PR-E3 in [#22](https://github.com/HumanBean17/java-enterprise-codebase-rag/pull/22).
 
 ### Risk
 

tests/fixtures/fqn_collision_smoke/svc-x/pom.xml

@@ -0,0 +1,8 @@
+<project xmlns="http://maven.apache.org/POM/4.0.0"
+         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
+  <modelVersion>4.0.0</modelVersion>
+  <groupId>smoke</groupId>
+  <artifactId>svc-x</artifactId>
+  <version>1.0.0</version>
+</project>

tests/fixtures/fqn_collision_smoke/svc-x/src/main/java/com/example/Caller.java

@@ -0,0 +1,7 @@
+package com.example;
+
+public class Caller {
+    public void run() {
+        new SharedDto().process();
+    }
+}

tests/fixtures/fqn_collision_smoke/svc-x/src/main/java/com/example/SharedDto.java

@@ -0,0 +1,6 @@
+package com.example;
+
+public class SharedDto {
+    public void process() {
+    }
+}

tests/fixtures/fqn_collision_smoke/svc-y/pom.xml

@@ -0,0 +1,8 @@
+<project xmlns="http://maven.apache.org/POM/4.0.0"
+         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
+  <modelVersion>4.0.0</modelVersion>
+  <groupId>smoke</groupId>
+  <artifactId>svc-y</artifactId>
+  <version>1.0.0</version>
+</project>

tests/fixtures/fqn_collision_smoke/svc-y/src/main/java/com/example/SharedDto.java

@@ -0,0 +1,6 @@
+package com.example;
+
+public class SharedDto {
+    public void process() {
+    }
+}

tests/test_call_invariant.py

@@ -0,0 +1,53 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+import kuzu
+
+from build_ast_graph import GraphTables, pass1_parse, pass2_edges, pass3_calls, write_kuzu
+from kuzu_queries import KuzuGraph
+
+
+_FQN_COLLISION_FIXTURE = Path(__file__).resolve().parent / "fixtures" / "fqn_collision_smoke"
+_CROSS_SERVICE_FIXTURE = Path(__file__).resolve().parent / "fixtures" / "cross_service_smoke"
+
+
+def _build(root: Path, db_path: Path) -> None:
+    """Build through pass3 only (no routes); sufficient for `pass3_skipped_cross_service` assertions."""
+    tables = GraphTables()
+    asts = pass1_parse(root, tables, verbose=False)
+    pass2_edges(tables, asts, verbose=False)
+    pass3_calls(tables, asts, verbose=False)
+    write_kuzu(db_path, tables, source_root=root, verbose=False)
+
+
+def _scalar(db_path: Path, query: str) -> int:
+    conn = kuzu.Connection(kuzu.Database(str(db_path), read_only=True))
+    r = conn.execute(query)
+    return int(r.get_next()[0] or 0) if r.has_next() else 0
+
+
+def test_call_invariant_blocks_cross_microservice_edges(tmp_path: Path) -> None:
+    db = tmp_path / "fqn_collision.kuzu"
+    _build(_FQN_COLLISION_FIXTURE, db)
+    cross_calls = _scalar(
+        db,
+        "MATCH (a:Symbol)-[:CALLS]->(b:Symbol) "
+        "WHERE a.microservice <> '' AND b.microservice <> '' "
+        "AND a.microservice <> b.microservice "
+        "RETURN count(*)",
+    )
+    assert cross_calls == 0
+    assert KuzuGraph(str(db)).meta()["pass3_skipped_cross_service"] >= 1
+
+
+def test_call_invariant_inert_on_clean_fixtures(tmp_path: Path) -> None:
+    db = tmp_path / "cross_service_smoke.kuzu"
+    _build(_CROSS_SERVICE_FIXTURE, db)
+    assert KuzuGraph(str(db)).meta()["pass3_skipped_cross_service"] == 0
+
+
+def test_call_invariant_inert_on_bank_chat_system(tmp_path: Path, corpus_root: Path) -> None:
+    db = tmp_path / "bank_chat.kuzu"
+    _build(corpus_root, db)
+    assert KuzuGraph(str(db)).meta()["pass3_skipped_cross_service"] == 0