Summary
On the polars engine, a native op-list chain whose edge alias has the same name as an edge column that the same edge step filters on raises GFQLSchemaError [incompatible-column-type]. The pandas engine executes the same chain (alias marker authoritative over the colliding column, as combine_steps documents).
Same on 0.59.0 (3fb216dd1) with and without resident indexes, so it is independent of the fast paths and of the #2037/#2038 stack.
Repro
import pandas as pd, polars as pl, graphistry
from graphistry.compute.ast import n, e_forward
nodes = pd.DataFrame({"key": [1, 2, 3], "id": [10, 20, 30], "type": ["p", "p", "m"]})
edges = pd.DataFrame({"s": [3, 3], "d": [1, 2], "type": ["HAS_CREATOR", "OTHER"]})
ops = [n({"id": 30}, name="m"), e_forward({"type": "HAS_CREATOR"}, name="type"), n(name="p")]
g_pd = graphistry.nodes(nodes, "key").edges(edges, "s", "d")
g_pl = graphistry.nodes(pl.from_pandas(nodes), "key").edges(pl.from_pandas(edges), "s", "d")
print(g_pd.gfql(ops, engine="pandas")._edges.columns) # ['s', 'd', 'type'] (marker replaced the column)
g_pl.gfql(ops, engine="polars") # GFQLSchemaError: column "type" is numeric but filter value is stringA node alias colliding with a node property (n({"id": 30}, name="id")) is served by both engines.
Expected
Both engines agree: either both serve with the documented marker-authoritative collision contract, or both reject the collision with a structured error before execution. The polars full path appears to tag the alias marker onto the edge frame before applying the step's own edge_match, so the filter then sees the boolean marker.
Context
Found while adding native op-list shapes to the pyg-bench SNB point-latency sentinel (the SNB frames carry id/type columns, so alias/column collisions are realistic). The native polars seeded fast path (#2038) declines alias collisions, so this is the full path. Related: #2034 (rows() duplicate/null ids).
Summary
On the polars engine, a native op-list chain whose edge alias has the same name as an edge column that the same edge step filters on raises
GFQLSchemaError [incompatible-column-type]. The pandas engine executes the same chain (alias marker authoritative over the colliding column, ascombine_stepsdocuments).Same on 0.59.0 (
3fb216dd1) with and without resident indexes, so it is independent of the fast paths and of the #2037/#2038 stack.Repro
A node alias colliding with a node property (
n({"id": 30}, name="id")) is served by both engines.Expected
Both engines agree: either both serve with the documented marker-authoritative collision contract, or both reject the collision with a structured error before execution. The polars full path appears to tag the alias marker onto the edge frame before applying the step's own
edge_match, so the filter then sees the boolean marker.Context
Found while adding native op-list shapes to the pyg-bench SNB point-latency sentinel (the SNB frames carry
id/typecolumns, so alias/column collisions are realistic). The native polars seeded fast path (#2038) declines alias collisions, so this is the full path. Related: #2034 (rows() duplicate/null ids).