) (#1249) * fix(gfql/row): resolve bare relationship alias in row expressions (#1072) Bare edge aliases in select / where_rows / order_by row expressions raised "unsupported token in row expression" because _gfql_resolve_token only had a node-id fallback for the bare-alias case. Add an edge fallback: prefer the edge-id column, otherwise render the relationship as a Cypher-style [:TYPE {props}] string, mirroring what cypher RETURN <relAlias> already produces via result_postprocess. Direct GFQL chains and cypher-lowered chains now see the same column shape. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(cypher): support multi-alias carry through WITH before MATCH re-entry (#1071) Pre-rewrite the prefix WITH stage in _compile_bounded_reentry_query so secondary whole-row aliases are demoted to scalar property carries (S.X AS __cypher_reentry_<S>_<X>__) and downstream S.X references compose with the existing single-whole-row machinery (#1047/#1068). RETURN of a secondary whole-row alias, and re-binding a secondary alias as a node variable in the trailing MATCH, remain unsupported with precise errors. Closes #1071. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(gfql): unify relationship alias rendering and harden row resolution (#1072) * fix(gfql/cypher): close wave-3 alias resolution and reentry where gaps (#1072) * docs(changelog): add #1072 entry for relationship alias in row expressions Records the user-facing fix and amplifications for #1072 under ## [Development] → ### Fixed: bare edge-alias resolution in row expressions, shared vectorized renderer in row/entity_props.py, edge-alias gating via _gfql_rows_edge_aliases, string-escape / float / empty-type rendering normalization, and re-entry WHERE secondary-alias predicate rewriting via synthesized text. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(gfql/row): add cuDF smoke test for bare relationship alias rendering Wave 4 finding T2 (IMPORTANT): TestRelationshipAliasInRowExpression had 14 new tests in PR #1249 but zero cuDF coverage on the new edge-alias rendering path, despite Wave 2 vectorization being specifically motivated by cuDF support (Wave 1 finding #4). Adds test_select_bare_relationship_alias_renders_on_cudf_when_available following the sibling pattern at test_row_pipeline_vectorized_cudf_when_available (pytest.importorskip("cudf")). Asserts: (a) result keeps cuDF backend, (b) bare-alias rendering produces "[:WORKS_AT {workFrom: 2010}]", (c) select/return cell-for-cell parity holds on cuDF. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * refactor(gfql/row): collapse duplicate ValueError in _gfql_resolve_token Wave 4 finding D1 (SUGGESTION): the conditional at pipeline.py:2310 and the raise at pipeline.py:2312 produce identical errors. After the node-id branch (lines 2306-2309) returns, the only remaining outcome is "unsupported token", regardless of edge_aliases membership — both branches raised the same f-string. Collapsed to a single raise. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Send libname and version to avoid API breakage with server #4

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions