[rahulsmahadev]

Rationale for this change

A table partitioned by bucket[N] on a time column can be created, but every append() to it fails:

schema = Schema(NestedField(1, "id", LongType()), NestedField(2, "event_time", TimeType()))
spec = PartitionSpec(PartitionField(source_id=2, field_id=1000, transform=BucketTransform(4), name="event_time_bucket"))
table = catalog.create_table("default.t", schema=schema, partition_spec=spec)
table.append(pa.table({"id": [1], "event_time": pa.array([time(22, 31, 8)], pa.time64("us"))}))
# AttributeError: 'int' object has no attribute 'hour'

The spec lists time as a source type for bucket[N], and Java supports it. The bucket transform returns an int, which _to_partition_representation already passes through for date, timestamp and uuid sources, but the TimeType handler always called time_to_micros on it. This change makes the TimeType handler accept an int the same way.

Are these changes tested?

Yes. Both new tests fail without the fix:

• test_to_partition_representation_time in tests/test_transforms.py covers None, a time value, and the int a bucket transform returns. test_to_partition_representation_unrecognized_type_raises now also covers TimeType.
• test_append_to_table_bucketed_by_time in tests/io/test_pyarrow.py appends to a table bucketed on a time column and reads the rows back with and without a filter.

Are there any user-facing changes?

Bug fix only: appends to tables partitioned by bucket on a time column now succeed. No API changes.

This pull request and its description were written by Isaac.

[rahulsmahadev]

@Fokko @mittalpk could you take a look when you have a chance? It is a small fix in _to_partition_representation (last changed in #3695) so appends work for bucket partitions on a time column.

[mittalpk]

Looks good to me. This matches how #3695 handles bucketed date/timestamp sources, and the tests cover the int passthrough and the error path.

[rambleraptor]

This seems reasonable. Thanks for doing this!