Merge pull request #1400 from graphistry/issue-1396-worker-d

GFQL: tag-cooccurrence join aggregation cardinality hardening (#1396)

Author: lmeyerov

CHANGELOG.md

@@ -48,6 +48,7 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 ### Tests
 - **GFQL / Cypher two-MATCH reentry varlen regression hardening (#1001)**: Strengthened reentry varlen acceptance assertions from shape-only checks to exact expected rows, and added forward/reverse split-vs-connected query equivalence regressions to guard against wrong-row drift in the `match5-25/26` query family.
 - **GFQL / Cypher reentry ordered-top-k amplification (#1342, #880 partial)**: Added lowering regressions for MATCH-after-WITH re-entry with single-column and multi-column ordered top-k prefixes, carried-scalar top-k alignment, `LIMIT 0` empty-prefix behavior, `SKIP` failfast retention, plus cuDF parity coverage for the multi-row top-k lane.
+- **GFQL / Cypher tag-cooccurrence join+aggregation cardinality amplification (#1396, #880 residual lane)**: Added focused IC6-shape regression coverage for `collect(distinct friend) -> UNWIND -> connected comma MATCH -> WITH tag.name, count(post)` with non-trivial grouped counts (`Alpha=2`, `Beta=1`) plus cuDF parity guard, so the residual tag-cooccurrence join-aggregation lane is pinned without adapter-side workaround assumptions.
 - **DataFrame join helper branch coverage expansion + RAPIDS matrix validation (#1380)**: Expanded `graphistry/tests/compute/dataframe/test_join.py` to cover non-overwrite and empty-join schema branches (`joined_hidden_scalar_columns`, `joined_alias_columns`, `connected_inner_join_rows`), prefix/no-label projection behavior (`project_node_attrs`), inequality branch coverage (`ineq_eval_pairs`), and semijoin no-shared/delegation paths (`semijoin_eval_pairs`). Kept GFQL reentry/collision semantics in integration suites. Validated on DGX GPU for both RAPIDS `25.02` and `26.02` using `docker/test-rapids-official-local.sh` with GFQL profile + dataframe join tests.
 
 ### Internal

graphistry/tests/compute/gfql/cypher/test_lowering.py

@@ -619,6 +619,54 @@ def _mk_issue_1000_ic6_minimal_graph_cudf() -> _CypherTestGraph:
     return _mk_cudf_graph(graph._nodes, graph._edges)
 
 
+def _mk_issue_1396_tag_cooccurrence_join_aggregation_graph() -> _CypherTestGraph:
+    return _mk_graph(
+        pd.DataFrame(
+            {
+                "id": [501, 502, 503, 4398046511333, 2, 3, 9001, 9002, 9003],
+                "label__Tag": [True, True, True, False, False, False, False, False, False],
+                "label__Person": [False, False, False, True, True, True, False, False, False],
+                "label__Post": [False, False, False, False, False, False, True, True, True],
+                "name": [
+                    "Carl_Gustaf_Emil_Mannerheim",
+                    "Alpha",
+                    "Beta",
+                    None,
+                    None,
+                    None,
+                    None,
+                    None,
+                    None,
+                ],
+            }
+        ),
+        pd.DataFrame(
+            {
+                "s": [4398046511333, 4398046511333, 9001, 9002, 9003, 9001, 9002, 9003, 9001, 9002, 9003],
+                "d": [2, 3, 2, 2, 3, 501, 501, 501, 502, 502, 503],
+                "type": [
+                    "KNOWS",
+                    "KNOWS",
+                    "HAS_CREATOR",
+                    "HAS_CREATOR",
+                    "HAS_CREATOR",
+                    "HAS_TAG",
+                    "HAS_TAG",
+                    "HAS_TAG",
+                    "HAS_TAG",
+                    "HAS_TAG",
+                    "HAS_TAG",
+                ],
+            }
+        ),
+    )
+
+
+def _mk_issue_1396_tag_cooccurrence_join_aggregation_graph_cudf() -> _CypherTestGraph:
+    graph = _mk_issue_1396_tag_cooccurrence_join_aggregation_graph()
+    return _mk_cudf_graph(graph._nodes, graph._edges)
+
+
 def _prefix_scalar_reentry_query(
     *,
     tag_name: str = "topic",
@@ -10734,6 +10782,35 @@ def test_string_cypher_executes_issue_1000_ic6_exact_runtime_minimal_on_cudf() -
     ]
 
 
+def test_issue_1396_tag_cooccurrence_join_aggregation_counts() -> None:
+    """IC6 tag-cooccurrence join+aggregation shape keeps grouped post cardinality."""
+    result = _mk_issue_1396_tag_cooccurrence_join_aggregation_graph().gfql(
+        _issue_1000_ic6_query(),
+        params=_issue_1000_ic6_params(),
+    )
+
+    assert result._nodes.to_dict(orient="records") == [
+        {"tagName": "Alpha", "postCount": 2},
+        {"tagName": "Beta", "postCount": 1},
+    ]
+
+
+def test_issue_1396_tag_cooccurrence_join_aggregation_counts_on_cudf() -> None:
+    pytest.importorskip("cudf")
+
+    result = _mk_issue_1396_tag_cooccurrence_join_aggregation_graph_cudf().gfql(
+        _issue_1000_ic6_query(),
+        params=_issue_1000_ic6_params(),
+        engine="cudf",
+    )
+
+    assert type(result._nodes).__module__.startswith("cudf")
+    assert result._nodes.to_pandas().to_dict(orient="records") == [
+        {"tagName": "Alpha", "postCount": 2},
+        {"tagName": "Beta", "postCount": 1},
+    ]
+
+
 def test_string_cypher_executes_scalar_only_prefix_with_match_reentry() -> None:
     query = _prefix_scalar_reentry_query(order_by="id")