Fix empty post-aggregate list projection

[lmeyerov]

Fixes #1367
Refs #1353

Summary

• Fix empty MATCH aggregate synthesis for post-aggregate projections by evaluating the public projection before seeding empty_result_row.
• Consolidate synthetic empty-row graph shims so the Cypher surface guard remains under baseline.
• Add amplified Tranche: Comparison/list/null wrong-row semantics (14 cases) under #1353 #1367 regressions for empty/global aggregate rows, grouped-empty negative behavior, optional-null extension, real relationship counting, and cuDF paths.
• Update the changelog.

TCK sidecar

• Sidecar PR: Promote empty aggregate direct Cypher rows tck-gfql#109
• Rebased sidecar directly onto main after tck-gfql feature(client path control): specify scheme/authority/path #110 absorbed the prior stacked string-literal base.
• Promotes expr-list12-3 plus current-main compatibility promotion return2-10.

Validation

• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k "issue_1367 and not cudf"
  • 5 passed
• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k "(issue_1367 or post_aggregate_size_collect_projection) and not cudf"
  • 7 passed
• python3 bin/ci_cypher_surface_guard.py
  • pass, lowering.py lines 8966 <= 8977
• ./bin/ruff.sh graphistry/compute/gfql/cypher/lowering.py graphistry/tests/compute/gfql/cypher/test_lowering.py
  • pass
• ./bin/typecheck.sh graphistry/compute/gfql/cypher/lowering.py
  • pass
• tck-gfql sidecar direct-Cypher contract
  • 764 passed, 3705 deselected
• tck-gfql sidecar post-fix full suite
  • 3935 passed, 763 xfailed
• GitHub CI
  • pygraphistry Fix empty post-aggregate list projection #1445 full checks pass after rerunning tck-gfql against sidecar commit b9d79b6
  • tck-gfql Dev/neo4j tuning #109 tests pass

DGX / cuDF

• Ran on dgx-spark via docker/test-rapids-official-local.sh, narrow selector:
  • test_issue_1367_empty_optional_match_post_aggregate_list_comprehension_size_on_cudf
  • test_issue_1367_post_aggregate_list_projection_counts_existing_relationships_on_cudf
• RAPIDS 25.02 official image: 2 passed
• RAPIDS 26.02 official image: 2 passed
• Used narrow selector because dgx-spark is shared; GPU was visible as NVIDIA GB10, utilization 0 %, memory fields N/A.

Review

• Used agents/skills/review/SKILL.md in findings mode through convergence.
• Original review converged after 4 waves.
• Follow-up review after rebase/test amplification/DGX evidence converged after 2 clean waves.
• Follow-up final report: plans/issue-1367-comparison-list-null/review/followup-2026-05-15/final-report.md.

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.