This is because a mutation step precedes the projection+filtering:

[kszucs]

This is because a mutation step precedes the projection+filtering:

functional_alltypes = functional_alltypes.mutate(
    bool_col=t.bool_col == 1,
    timestamp_col=t.timestamp_col.cast('timestamp'),
)

Translated to datafusion level:

import datafusion as df

# bool_col is an integer here, convert it to be actually boolean
bool_col = (df.column("bool_col") == df.literal(1)).alias("bool_col")
result = table.select(bool_col).filter(df.column("bool_col"))
result.collect()

raises

Exception: Error during planning: Ambiguous reference to field named 'bool_col'

We could use LogicalPlanBuilder::project_with_alias() so ibis can assign and track unique qualifiers for ambiguous field names coming from subselects. Though this must be wired through all the way to PyDataFrame::select()

cc @houqp

Originally posted by @kszucs in #2918 (comment)

[kszucs]

Wanted to create in the datafusion repository, closing.