feat(gfql/cypher): standard =~ regex + numeric fns + toLower/toUpper (#1673) by lmeyerov · Pull Request #1675 · graphistry/pygraphistry

lmeyerov · 2026-07-02T05:25:30Z

Implements the streamgl-viz filter-language conformance gaps (#1673) by adopting standard openCypher/neo4j Cypher syntax — not inventing dialect. Based directly on master (the polars contains regex fix it originally stacked on merged via #1660).

Research first

Verified each construct against the neo4j Cypher manual, openCypher 9, and ISO GQL (sources in the issue). Finding: most "gaps" are standard Cypher, so we adopt the standard forms. LIKE/ILIKE/BETWEEN are SQL-only (not in Cypher or GQL) → intentionally not implemented; chained comparison (1 < n.age < 65) already worked.

What's added (all standard openCypher/neo4j)

=~ regex-match operator — Java-regex, full/anchored match (n.name =~ 'AB' matches only 'AB'), inline (?i)/(?m)/(?s) flags; lowers to fullmatch. Simple WHERE …=~… on all engines (filter_by_dict); composes through AND/OR/NOT/RETURN on pandas/cuDF (polars declines the complex OR/NOT row-filter form with an honest NotImplementedError — pre-existing where_rows gap). Adds native polars Match/Fullmatch lowering.
Numeric functions floor, ceil (≡ceiling), round(x)/round(x, precision) (complementing existing abs/sqrt/sign).
toLower/toUpper — the idiomatic case-insensitive compare.

Wired end-to-end: cypher grammar + shared expr_parser (REGEX_MATCH terminal) + allow-lists + evaluators (pandas/cuDF row/pipeline.py+row/dispatch.py; polars predicates.py+row_pipeline.py).

Tests / quality

Differential-parity tested pandas↔polars for =~, Match/Fullmatch, numeric fns, sign, toLower/toUpper. Full polars lane passes; ruff+mypy clean. A focused adversarial review found no BLOCKERs; its two findings are addressed (polars parity now guarded by parametrized tests; polars sign() cast to Int64 to match pandas/neo4j Integer spec).

GPU parity verified on dgx (GB10, --gpus all, 2026-07-05): full polars lane + conformance matrix + test_engine_polars_gpu.py = 1803 passed, 0 skipped/failed — cuDF + polars-gpu (cudf_polars) matrix cases (incl. Match/Fullmatch + the scalar fns) all parity-green vs the pandas oracle.

Review hardening (4-wave converged adversarial review, 2026-07-05)

A multi-wave findings-mode review (artifacts: plans/viz-filter-pipeline/waves/pr1675/) found and fixed, each dgx-repro'd before and re-verified after: composed =~ broken on engine='cudf' (no .str.fullmatch; now delegates to the hardened Fullmatch predicate); cuDF fullmatch alternation mis-anchoring ('ab|cd' silently matched 'abXXX'); the cuDF (?i) lowercase-fold silently inverting escape classes (now a precise decline gate: uppercase escapes / case-crossing ranges / hex escapes / non-ASCII decline, lowercase escapes keep folding); polars Match/Fullmatch missing the Rust-regex guard (lookarounds crashed non-NIE); toLower(numeric col) broadcasting the Series repr on the pandas oracle (now declines like neo4j's type error); polars floor/ceil/round returning Int64 on int columns (now Float64 like neo4j/pandas); NotImplementedError being re-labeled by broad excepts (honest NIEs now survive to the caller); plus matrix coverage for the =~ family, null cells, dtype pins, and discriminating regression fixtures. Waves 3–4 converged at zero BLOCKER/IMPORTANT.

Docs

docs/source/gfql/cypher.rst: =~ in WHERE Forms + a new "Scalar Functions and Operators" section (fns/toLower + why LIKE/BETWEEN aren't provided). CHANGELOG entries for both increments.

Deferred (documented)

round(x, precision, mode) 3-arg form; at this layer round uses numpy half-to-even — the stacked fix(gfql/cypher): neo4j round() tie-breaking + lower/upper GQL aliases (#1673) #1677 replaces it with neo4j's documented tie-breaking.
^ power operator: initially implemented right-associative, then removed during standards vetting — the openCypher TCK pins ^ LEFT-associative and the current conformance corpus marks it reject-expected, so re-adding is a coordinated corpus change (spec record in fix(gfql/cypher): neo4j round() tie-breaking + lower/upper GQL aliases (#1673) #1677/GFQL Cypher dialect: host the streamgl-viz filter language (regex/case, floor/ceil/round/pow, prune-isolated, two-mask via GRAPH{}) #1673).

🤖 Generated with Claude Code

…s (no crash) GPU-parity pass (viz-filter #1673 item 2) on dgx found `MATCH (n) WHERE n.name =~ '(?i)…'` CRASHES on engine='cudf' with libcudf "invalid regex pattern: nothing to repeat". Root cause: libcudf's regex engine rejects inline flag groups ((?i)/(?m)/ (?s)) at ANY position (verified: '(?i)abc', '^(?i)abc$', '(?i)^abc$' all raise; only flag-free '^abc$' works) — not merely a position issue. Fix: the Match/Fullmatch cuDF branches now translate a leading (?i) to the existing lowercase case-folding workaround (parity with pandas' (?i)), and honestly decline any other inline flag with NotImplementedError instead of crashing. Shared helper _cudf_regex_prep. pandas/polars paths untouched. Validated on dgx (RAPIDS 26.02): cudf =~ '(?i)a.c' == pandas [2,3] (was RuntimeError); 446 regex/match/fullmatch/contains/numeric tests pass across pandas/polars/cudf; +1 cudf-gated regression test (test_regex_cudf_inline_flag_parity). ruff+mypy clean. Also confirms viz-filter #1673 item 2: cuDF numeric (floor/ceil/round) + toLower/ toUpper parity OK; polars-gpu is N/A on this branch (#1675 is off #1660, below #1655 which introduces polars-gpu). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…in _cudf_regex_prep The cuDF =~ flags commit annotated _cudf_regex_prep with "tuple[object, bool]", which mypy rejects on py3.8/3.9 (builtin generics need 3.9+/3.10+); broke the python-lint-types (3.8, 3.9) lanes on #1675/#1677. Use typing.Tuple. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… transform) _cudf_regex_prep is a pure string transform (no cuDF required), but its only tests were cuDF-gated, so CPU CI left the new lines uncovered and #1675's changed-line-coverage gate fell to 65.9%. Add direct unit tests for every branch: non-str/no-flag passthrough, leading (?i)/(?ii) -> case-fold, other inline flags -> honest NotImplementedError. Remaining uncovered lines are the is_cudf execution branches (GPU-only; covered by the dgx parity pass). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>