Cypher: support bounded varlen WHERE pattern predicates (#973)

Author: lmeyerov

CHANGELOG.md

@@ -10,6 +10,10 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ### Documentation
 - **GFQL component-labeling examples + README clarity (#1324)**: Added concise WCC/SCC labeling examples for `compute_cugraph`, `compute_igraph('clusters')`, and local Cypher `CALL graphistry.cugraph.*` write/row modes in GFQL docs, clarified that component IDs are partition labels (not stable semantic IDs), and tightened the main README GFQL intro sentence for readability.
+- **GFQL / Cypher docs — variable-length boundary refresh (#973)**: Updated direct-Cypher capability docs (`docs/source/gfql/cypher.rst`, `docs/source/gfql/spec/cypher_mapping.md`) to reflect current support for connected variable-length patterns and bounded/exact variable-length `WHERE` pattern predicates, while preserving explicit fail-fast notes for remaining path/list-carrier and advanced row-shaping gaps.
+
+### Changed
+- **GFQL / Cypher lowering — bounded/exact variable-length `WHERE` pattern predicates (#973)**: Removed the pre-normalization compiler gate that rejected bounded/exact variable-length `WHERE` pattern predicates and now lower these shapes through the existing WHERE-pattern rewrite and row-filter paths. Converted the old fail-fast test into positive execution coverage and added boolean-wrapper amplification (`OR`/`XOR`/`NOT`) for bounded variable-length `WHERE` predicates in `graphistry/tests/compute/gfql/cypher/test_lowering.py`.
 
 ## [0.55.1 - 2026-05-05]
 

docs/source/gfql/cypher.rst

@@ -206,11 +206,11 @@ Support Matrix
      - Execute directly through ``g.gfql("...")``. Helper translation to a single ``Chain`` is stricter.
    * - Variable-length relationship patterns
      - Partial
-     - Direct Cypher supports endpoint-only traversals such as ``[*2]``,
-       ``[*1..3]``, ``[*]``, and typed forms like ``[:R*2..4]``, plus bounded
-       connected multi-relationship patterns where the row shape stays in the
-       current supported subset. Path/list-carrier uses, bounded/exact
-       ``WHERE`` pattern predicates, and broader branching/path-shaping cases
+     - Direct Cypher supports endpoint traversals such as ``[*2]``,
+       ``[*1..3]``, ``[*]``, and typed forms like ``[:R*2..4]``; connected
+       multi-relationship variable-length patterns; and bounded/exact/fixed-point
+       variable-length ``WHERE`` pattern predicates in the current row-shaped
+       subset. Path/list-carrier uses and unsupported path/row-shaping cases
        still fail fast.
    * - ``CREATE`` / ``DELETE`` / ``SET``
      - Not supported
@@ -236,9 +236,10 @@ Pattern Matching Forms
 - Node labels and multi-label node patterns such as ``(p:Person:Admin)``.
 - Relationship direction forms ``->``, ``<-``, and undirected ``-[]-``.
 - Relationship type alternation such as ``[r:KNOWS|HATES]``.
-- Single variable-length relationship patterns when they are the only
-  relationship in the connected pattern, including ``[*n]``, ``[*m..n]``,
-  ``[*]``, and typed forms such as ``[:R*2..4]``.
+- Single variable-length relationship patterns, including ``[*n]``,
+  ``[*m..n]``, ``[*]``, and typed forms such as ``[:R*2..4]``.
+- Connected patterns that mix variable-length and fixed-length relationships,
+  such as ``MATCH (a)-[:R*2]->()-[:S]->(c) RETURN c``.
 - Connected comma-separated patterns such as
   ``MATCH (a)-[:A]->(b), (b)-[:B]->(c)``.
 - Repeated ``MATCH`` clauses when they stay connected through shared aliases.
@@ -255,40 +256,35 @@ WHERE Forms
 - Label predicates such as ``WHERE b:Foo:Bar``.
 - Relationship-type predicates such as ``WHERE type(r) = 'KNOWS'``.
 - Positive relationship-existence pattern predicates such as
-  ``WHERE (n)-[:R]->()`` and bare fixed-point variable-length existence checks
-  such as ``WHERE (n)-[*]-()``.
-- One positive relationship-existence pattern predicate may be combined with
-  ordinary row filters through top-level ``AND``, for example
-  ``WHERE n.kind = 'x' AND (n)-[:R*]->() AND n.id <> 'a'``.
+  ``WHERE (n)-[:R]->()`` and variable-length existence checks such as
+  ``WHERE (n)-[*]-()`` and ``WHERE (n)-[:R*2]->()``.
+- Pattern predicates can be combined with row predicates in the current
+  boolean subset, including ``AND`` / ``OR`` / ``XOR`` and ``NOT`` forms.
 
 Variable-Length Relationship Boundary
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-Direct Cypher multihop support is intentionally narrow in the current landing
-slice. The supported direct forms include endpoint traversals and bounded
-connected multi-relationship patterns where the result stays in the current
-row-shaping subset, for example:
+Direct Cypher multihop support remains intentionally bounded. The supported
+direct forms include endpoint traversals, connected multi-relationship
+patterns, and variable-length ``WHERE`` pattern predicates where the result
+stays in the current row-shaping subset, for example:
 
 - ``MATCH (a)-[*2]->(b) RETURN b``
 - ``MATCH (a)-[:R*1..3]->(b) RETURN b``
 - ``MATCH (a)<-[*2]-(b) RETURN b``
 - ``MATCH (a)-[:R*1..2]-(b) RETURN b``
 - ``MATCH (a)-[:R*2]->(b)-[:S]->(c) RETURN c``
 - ``MATCH (a)-[:R]->(b), (b)-[:S*1..2]->(c) RETURN a.id AS a_id, c.id AS c_id``
+- ``MATCH (n) WHERE (n)-[:R*2]->() RETURN n``
+- ``MATCH (n) WHERE NOT (n)-[:R*2]->() RETURN n.id AS id``
 
 The current compiler explicitly rejects these remaining subfamilies with
 ``GFQLValidationError`` instead of attempting unsound execution:
 
 - path/list-carrier use of a variable-length relationship alias, such as
   ``RETURN r`` or ``count(r)``
-- exact or bounded variable-length ``WHERE`` pattern predicates such as
-  ``WHERE (n)-[:R*2]-()``
-- top-level ``OR`` / ``NOT`` around variable-length ``WHERE`` pattern
-  predicates, or more than one positive pattern predicate in the same
-  ``WHERE`` clause
-- branching connected multihop patterns, or shapes that would require
-  unsupported path/relationship-carrier row shaping around a variable-length
-  segment
+- shapes that still require unsupported path/relationship-carrier row shaping
+  around a variable-length segment
 - connected multi-pattern relationship-alias projection such as
   ``RETURN r`` / ``r.prop`` when it would require unsupported row shaping
 - multi-alias ``RETURN *`` projections that would require unsupported
@@ -431,10 +427,8 @@ Not Supported Today
 
 - Variable-length relationship aliases used as path/list carriers, such as
   ``RETURN r`` or ``count(r)``.
-- Exact or bounded variable-length ``WHERE`` pattern predicates such as
-  ``WHERE (n)-[:R*2]-()``.
-- Branching connected multihop patterns, or connected multihop shapes that
-  still require unsupported path/relationship-carrier row shaping.
+- Connected multihop shapes that still require unsupported
+  path/relationship-carrier row shaping.
 - Multiple disconnected ``MATCH`` patterns used as arbitrary joins.
 - Multi-pattern re-entry shapes beyond the bounded single
   ``MATCH ... WITH ... MATCH ... RETURN`` form.

docs/source/gfql/spec/cypher_mapping.md

@@ -36,11 +36,11 @@ When translating from Cypher, you'll encounter three scenarios:
 ### Direct Translations
 - Graph patterns: `(a)-[r]->(b)` → chain operations
 - Property filters: WHERE clauses embed into operations
-- Path traversals: direct `g.gfql("MATCH ...")` supports endpoint-only single
-  variable-length relationship forms such as `[*2]`, `[*1..3]`, and `[*]`.
-  Native GFQL still gives you the full explicit hop surface, including output
-  slicing, intermediate-hop aliasing, and rewrites for currently unsupported
-  direct-Cypher multihop shapes.
+- Path traversals: direct `g.gfql("MATCH ...")` supports single and connected
+  variable-length relationship forms such as `[*2]`, `[*1..3]`, and `[*]`,
+  including bounded/exact variable-length `WHERE` pattern predicates in the
+  current row-shaped subset. Native GFQL still gives you the full explicit hop
+  surface (output slicing, intermediate-hop aliasing, and custom rewrites).
 - Pattern composition: Multiple patterns become sequential operations
 - Same-path constraints: `WHERE` across steps → `g.gfql([...], where=[...])`
 
@@ -255,10 +255,10 @@ g.gfql([
 ### Edge Patterns
 
 Rows using `[*...]` below show the native GFQL rewrite for the same traversal
-intent. Direct `g.gfql("MATCH ...")` now accepts these endpoint-only
-single-variable-length relationship forms, while native GFQL remains the more
-explicit option when you need intermediate-hop control or unsupported mixed
-pattern shapes.
+intent. Direct `g.gfql("MATCH ...")` accepts these variable-length forms in
+the supported direct-Cypher subset, while native GFQL remains the more explicit
+option when you need intermediate-hop control or advanced path/list-carrier
+semantics.
 
 | Cypher / intent | Python | Wire Protocol (compact) |
 |-----------------|--------|-------------------------|
@@ -274,7 +274,7 @@ pattern shapes.
 | `-[r:BOUGHT {amount: gt(100)}]->` | `e_forward({"type": "BOUGHT", "amount": gt(100)}, name="r")` | `{"type": "Edge", "direction": "forward", "edge_match": {"type": "BOUGHT", "amount": {"type": "GT", "val": 100}}, "name": "r"}` |
 
 When you need constraints on intermediate hops, path/list-carrier semantics, or
-mixed connected patterns beyond the current direct-Cypher subset, use repeated
+advanced row-shaping beyond the current direct-Cypher subset, use repeated
 single-hop GFQL steps with aliases instead of collapsing the traversal into one
 multihop edge operator.
 

graphistry/compute/gfql/cypher/lowering.py

@@ -5590,34 +5590,6 @@ def _is_variable_length_relationship_pattern(relationship: RelationshipPattern)
     )
 
 
-def _reject_unsupported_variable_length_where_pattern_predicates(query: CypherQuery) -> None:
-    if query.where is None:
-        return
-    predicates: List[WherePatternPredicate] = [
-        predicate for predicate in query.where.predicates if isinstance(predicate, WherePatternPredicate)
-    ]
-    if query.where.expr_tree is not None:
-        predicates.extend(_where_expr_tree_pattern_predicates(query.where.expr_tree))
-    for predicate in predicates:
-        relationships = [
-            element
-            for element in predicate.pattern
-            if isinstance(element, RelationshipPattern)
-        ]
-        for relationship in relationships:
-            if not _is_variable_length_relationship_pattern(relationship):
-                continue
-            if relationship.min_hops is None and relationship.max_hops is None and relationship.to_fixed_point:
-                continue
-            raise _unsupported(
-                "Cypher WHERE pattern predicates currently support only bare variable-length fixed-point relationships, not exact or bounded hop counts",
-                field="where",
-                value=boolean_expr_to_text(query.where.expr_tree) if query.where.expr_tree is not None else None,
-                line=predicate.span.line,
-                column=predicate.span.column,
-            )
-
-
 def _reject_nonterminal_variable_length_relationship_patterns(query: CypherQuery) -> None:  # noqa: ARG001
     """No-op: variable-length rels in connected patterns are now supported.
 
@@ -8331,7 +8303,6 @@ def _attach_graph_context(result: CompiledCypherQuery) -> CompiledCypherQuery:
 
     normalizer = ASTNormalizer()
     query = normalizer.rewrite_shortest_path(query)
-    _reject_unsupported_variable_length_where_pattern_predicates(query)
     _reject_variable_length_path_alias_references(query, params=params)
     query = normalizer.rewrite_where_pattern_predicates(query)
 

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

@@ -5258,22 +5258,79 @@ def test_connected_variable_length_typed_mixed() -> None:
 
 
 @pytest.mark.parametrize(
-    "query",
+    "query,expected_rows",
     [
-        "MATCH (n) WHERE (n)-[:REL1*2]-() RETURN n",
-        "MATCH (n) WHERE (n)-[*2]-() RETURN n",
-        "MATCH (n) WHERE (n)<-[:REL1*1..2]-() RETURN n",
-        "MATCH (n) WHERE (n)-[:REL1*2]-() AND n.id <> 'a' RETURN n",
+        (
+            "MATCH (n) WHERE (n)-[:REL1*2]->() RETURN n.id AS id ORDER BY id",
+            [{"id": "a"}, {"id": "b"}, {"id": "c"}],
+        ),
+        (
+            "MATCH (n) WHERE (n)-[*2]->() RETURN n.id AS id ORDER BY id",
+            [{"id": "a"}, {"id": "b"}, {"id": "c"}],
+        ),
+        (
+            "MATCH (n) WHERE (n)<-[:REL1*1..2]-() RETURN n.id AS id ORDER BY id",
+            [{"id": "b"}, {"id": "c"}, {"id": "d"}],
+        ),
+        (
+            "MATCH (n) WHERE (n)-[:REL1*2]->() AND n.id <> 'a' RETURN n.id AS id ORDER BY id",
+            [{"id": "b"}, {"id": "c"}],
+        ),
     ],
 )
-def test_string_cypher_failfast_rejects_bounded_variable_length_where_pattern_predicates(query: str) -> None:
-    graph = _mk_empty_graph()
+def test_string_cypher_executes_bounded_variable_length_where_pattern_predicates(
+    query: str,
+    expected_rows: list[dict[str, object]],
+) -> None:
+    graph = _mk_graph(
+        pd.DataFrame({"id": ["a", "b", "c", "d"]}),
+        pd.DataFrame(
+            {
+                "s": ["a", "b", "c"],
+                "d": ["b", "c", "d"],
+                "type": ["REL1", "REL1", "REL1"],
+            }
+        ),
+    )
 
-    with pytest.raises(GFQLValidationError) as exc_info:
-        graph.gfql(query)
+    result = graph.gfql(query)
+    assert result._nodes.to_dict(orient="records") == expected_rows
 
-    assert exc_info.value.code == ErrorCode.E108
-    assert "WHERE pattern predicates" in exc_info.value.message
+
+@pytest.mark.parametrize(
+    "query,expected_rows",
+    [
+        (
+            "MATCH (n) WHERE (n)-[:REL1*2]->() OR n.id = 'd' RETURN n.id AS id ORDER BY id",
+            [{"id": "a"}, {"id": "b"}, {"id": "d"}],
+        ),
+        (
+            "MATCH (n) WHERE (n)-[:REL1*2]->() XOR n.id = 'd' RETURN n.id AS id ORDER BY id",
+            [{"id": "a"}, {"id": "b"}, {"id": "d"}],
+        ),
+        (
+            "MATCH (n) WHERE NOT (n)-[:REL1*2]->() RETURN n.id AS id ORDER BY id",
+            [{"id": "c"}, {"id": "d"}],
+        ),
+    ],
+)
+def test_string_cypher_executes_bounded_variable_length_where_pattern_boolean_wrappers(
+    query: str,
+    expected_rows: list[dict[str, object]],
+) -> None:
+    graph = _mk_graph(
+        pd.DataFrame({"id": ["a", "b", "c", "d"]}),
+        pd.DataFrame(
+            {
+                "s": ["a", "b", "c"],
+                "d": ["b", "c", "d"],
+                "type": ["REL1", "REL1", "REL1"],
+            }
+        ),
+    )
+
+    result = graph.gfql(query)
+    assert result._nodes.to_dict(orient="records") == expected_rows