infer_rel_schema raises AttributeError on an unbound column: the nullability helpers assume every Type.kind has a nullability field #276

_field_nullability (src/substrait/type_inference.py:624) and _with_field_nullability (type_inference.py:630) both do getattr(t, kind).nullability after only checking that t has a kind. Type.Unbound is the one Type.kind oneof variant with no nullability field, so both raise on it.

Per spec v0.99.0 site/docs/types/type_classes.md, the unbound type "may appear wherever a concrete type would normally appear, for example as a field type within ReadRel.base_schema", and it "carries no nullability and accepts no parameters; those properties are determined when it is bound to a concrete type". So an unbound field should pass through nullability derivation untouched rather than crash.

Reached today through _set_output_struct, i.e. any SetRel over a partially bound input.

Reproducer

import substrait.algebra_pb2 as stalg, substrait.type_pb2 as stt
from substrait.type_inference import infer_rel_schema


def read(name, type_):
    return stalg.Rel(read=stalg.ReadRel(
        named_table=stalg.ReadRel.NamedTable(names=[name]),
        base_schema=stt.NamedStruct(names=[name], struct=stt.Type.Struct(
            types=[type_], nullability=stt.Type.NULLABILITY_REQUIRED))))


unbound = stt.Type(unbound=stt.Type.Unbound())
rel = stalg.Rel(set=stalg.SetRel(
    inputs=[read("a", unbound), read("b", unbound)],
    op=stalg.SetRel.SET_OP_UNION_ALL))

print(infer_rel_schema(rel))

  File "src/substrait/type_inference.py", line 627, in _field_nullability
    return getattr(t, kind).nullability if kind else stt.Type.NULLABILITY_UNSPECIFIED
AttributeError: nullability

Measured on main at afa3c52. Pre-existing.

Guarding both helpers on kind != "unbound" covers it. Note #272 introduces a second caller of _with_field_nullability (join null-padding), which reaches the same gap for LEFT / RIGHT / OUTER / *_SINGLE joins; that half is raised on the PR. The set-operation nullability rules themselves came from #219.

🤖 Filed with AI assistance as a follow-up to review of #272.