gfql: an alias named like the source, destination or edge-id binding column is not rejected (silent clobber on pandas/cuDF, raw polars SchemaError)

Summary

A node alias equal to the node-id binding column is rejected up front with GFQLValidationError on every engine. The three other binding columns are not guarded:

alias	pandas / cuDF	polars
edge alias = source column (`name="s"`)	served, but the marker overwrites the source column: result has no edges	served with an extra `s_right` column
edge alias = destination column (`name="d"`)	served, destination column replaced by `True` (`(3, True)` edge pairs)	served, extra `d_right`
edge alias = edge-id binding (`name="eid"`)	served, edge-id column replaced by `True`	raw `polars.exceptions.SchemaError: datatypes of join keys don't match` (not a GFQL error)

Repro

import pandas as pd, graphistry
from graphistry.compute.ast import n, e_forward
nodes = pd.DataFrame({"key": [1, 2, 3], "id": [10, 20, 30]})
edges = pd.DataFrame({"s": [3, 3], "d": [1, 2], "type": ["HAS_CREATOR", "OTHER"], "eid": [100, 101]})
g = graphistry.nodes(nodes, "key").edges(edges, "s", "d", "eid")
ops = [n({"id": 30}, name="m"), e_forward({"type": "HAS_CREATOR"}, name="s"), n(name="p")]
g.gfql(ops)._edges          # empty on pandas; the seed's HAS_CREATOR edge is gone

Found by the alias/column collision probe run for the #2046 review.

Expected

The chain validator rejects an alias that equals any binding column (source, destination, edge id) the same way it rejects the node-id binding, on every engine, before execution.

Pins

Strict-xfail pins live in graphistry/tests/compute/test_chain_alias_column_collision.py (every engine) and flip when the validator covers these.

fix(gfql): polars node aliases that collide with a filtered column ar…

fix(gfql): an edge alias named like a binding column is a typed decli…

Merge pull request #2056 from graphistry/fix/gfql-2050-binding-alias-…

Verified resolved at master 599c405a234bff2402e4f76041cac6e169f6f7c8 (released runtime fixes are present in 0.60.0).

All three edge binding aliases (s, d, eid) now raise structured E108 / field chain.name before execution on both gfql() and chain(). The common validator runs before engine resolution/coercion. A 24-case probe across requested pandas/Polars/cuDF/Polars-GPU dispatches confirms the rejection; this is pre-engine validation, not a claim of native GPU execution. Existing binding/node-alias tests also pass on native pandas and Polars: 16 passed; 8 unavailable cuDF cases were explicitly skipped.

Reproduce the native CPU pins with python -m pytest -q graphistry/tests/compute/test_chain_alias_column_collision.py -k 'alias_named_like_a_binding_column or alias_named_like_the_node_binding'. The original silent endpoint/edge-ID overwrite is prevented, so this issue's requested validation is complete.

Navigation Menu

Repository navigation

gfql: an alias named like the source, destination or edge-id binding column is not rejected (silent clobber on pandas/cuDF, raw polars SchemaError) #2050

Summary

Repro

Expected

Pins

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

gfql: an alias named like the source, destination or edge-id binding column is not rejected (silent clobber on pandas/cuDF, raw polars SchemaError) #2050

Description

Summary

Repro

Expected

Pins

Activity

lmeyerov commented on Oct 9, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions