[META] Cypher binder coverage gaps blocking #1357 strict_name_resolution rollout #1371

Meta-issue tracking the binder coverage gaps surfaced when PR #1372 (work on #1357) ran a discovery flip of strict_name_resolution=True at graphistry/compute/gfql/cypher/lowering.py:8400 and observed 75 test failures in graphistry/tests/compute/gfql/cypher/. Most failures are false positives in strict mode — pre-existing binder limitations, not real-bugs-strict-catches. Each gap below is a separate binder fix that, once landed, unlocks one slice of the full strict flip.

PR #1372 ships the safe portion of #1357 (cross-kind alias rebind guard at _bind_node_pattern / _bind_relationship_pattern / _bind_path_alias — node↔edge, edge↔node, path↔node, path↔edge). Validator strict mode (gfql_validate(..., strict=True)) already exercises all of the below; runtime parity follows the items below.

Discovery analysis: plans/1357-binder-strict-name-resolution/research/discovery-flip-blast-radius.md (in PR #1372).
Regression-pin baselines: graphistry/tests/compute/gfql/cypher/test_binder_strict_compile_baseline.py (each parametrized case will flip from "admits" → "raises" as gaps close).
Validator/runtime parity ledger: graphistry/tests/compute/gfql/cypher/test_validator_runtime_strict_parity.py.

Gaps in priority order

P1 — Post-WITH UNWIND traversal order in _bind_graph_sequence

Symptom: MATCH ... WITH collect(b1) AS bees UNWIND bees AS b2 MATCH ... — strict binder rejects bees as unresolved.

Root cause: _bind_graph_sequence (binder.py:289) iterates ast.unwinds before the WITH stages. The parser puts post-WITH unwinds into ast.unwinds, so the binder evaluates the UNWIND expression against pre-WITH scope.

Fix: interleave UNWIND/WITH/MATCH by AST text position in _bind_graph_sequence, or fix the parser to route post-WITH UNWINDs to ast.reentry_unwinds.

Why P1: every LDBC SNB IC1/IC4/IC6 shape that uses collect(...) AS x UNWIND x AS y MATCH (...) is blocked by this — the highest-traffic shape on the benchmark lane. ~12 of the 75 failures are direct hits, plus most of the post-with-match-collect-unwind-match category (~5 more).

P2 — Quantifier predicates and list comprehensions

Symptom: MATCH (n) WHERE all(x IN n.labels WHERE x = 'A') RETURN n — strict rejects x as unresolved.

Root cause: _infer_expression_binding (binder.py:942) does not bind comprehension-scoped locals (x in all(x IN list WHERE ...), [x IN list WHERE ... | ...]).

Fix: in the strict-name-resolution unresolved-identifier check (_unresolved_identifiers in binder.py), recognize comprehension-binder syntax (all|any|none|single|[ ... IN ... ]) and exclude the local from the unresolved set.

Why P2: ~7 of 75 failures, common in WHERE clauses across many SNB shapes (IC*, BI*).

P3 — CALL/YIELD scope through prepass→normalize→bind

Symptom: CALL graphistry.degree() YIELD nodeId RETURN nodeId — strict rejects nodeId in RETURN.

Root cause: _bind_call_clause (binder.py:561) populates the scope from clause.yield_items, but the post-normalize bind path may receive a normalized AST shape that drops the YIELD aliases before re-binding. Need to verify and fix the cycle.

Fix: ensure YIELD aliases survive prepass→ASTNormalizer→bind. Likely a normalizer-side fix (don't strip YIELD context on rewrite).

Why P3: ~4 of 75 failures, narrow to graphistry.* / cugraph.* CALL paths.

P4 — Namespaced builtin function calls

Symptom: RETURN duration.inSeconds(localtime(), localtime()) AS duration — strict rejects duration as unresolved alias.

Root cause: _PROPERTY_RE.fullmatch(\"duration.inSeconds\") matches and routes through the alias-scope check; the binder does not recognize duration.* as a builtin namespace.

Fix: in _validate_property_ref_schema and _infer_expression_binding, short-circuit the alias-scope check when the prefix is a known builtin namespace (duration, time, date, datetime, localtime, localdatetime, point, …).

Why P4: ~5 of 75 failures, isolated to temporal/duration test fixtures. Niche.

Smaller items

• type(x) builtin not modeled by _infer_expression_binding (~3 failures).
• Aggregate precedence after collect() + range comparison (~3 failures).
• [x IN list | x.foo] map-comprehension form — same root cause as P2.

Closure criteria

Once all four P1-P4 are landed, the post-normalize strict flip at graphistry/compute/gfql/cypher/lowering.py:8400 (FrontendBinder().bind(query, PlanContext(), strict_name_resolution=True)) becomes safe. The regression-pin baselines in test_binder_strict_compile_baseline.py will then flip from "admits" → "raises" as the corresponding fix lands. When the file is empty (all cases moved to parity in test_validator_runtime_strict_parity.py), this meta closes alongside #1357.

Related

• Parent meta: [FEA] Cypher binder: enable strict-name-resolution to centralize alias-scope enforcement #1357
• Surfaced by: PR GFQL/Cypher: admit carried-endpoint rebind via single-MATCH flatten (#1341) #1351 round-001 agent-03 binder-parity audit, then quantified by PR Binder cross-kind alias rebind guard (#1357 partial) #1372.

Status (2026-05-09)

• PR Binder cross-kind alias rebind guard (#1357 partial) #1372 is merged and shipped the safe cross-kind guard slice.
• Remaining scope in this meta (P1-P4) is still open and continues to block enabling strict-name-resolution by default in runtime compile.

Status (2026-05-09)

• PR Fix #1371 P2 strict binder comprehension-local scope #1386 merged (4e6a18638) and closed P2 (quantifier/list comprehension strict scope).
• PR fix(gfql): preserve CALL/YIELD scope through strict row bind (#1371 P3) #1387 merged (cb5eaef6a) and closed P3 (CALL/YIELD strict scope through row-sequence binding).
• PR fix(cypher): close #1371 P4 namespaced builtin binder gap #1388 merged at 2026-05-09T11:14:44Z (ef6e7bf2e) and closed P4 (namespaced builtin function handling under strict binder).
• Together with earlier landed slices (P1 + cross-kind guard from Binder cross-kind alias rebind guard (#1357 partial) #1372), the P1-P4 coverage gaps tracked in this meta are now landed.