What this PR does / why we need it:

StreamFeatureView.__eq__ returns early when neither view has a udf:

if not self.udf:
    return not other.udf

so for UDF-less stream feature views (the common case for aggregation-only views) mode, timestamp_field, udf_string and aggregations are never compared. enable_tiling and tiling_hop_size are not compared at all, although both are serialized to the proto. As a result, two stream feature views with different aggregations compare equal:

a = StreamFeatureView(..., aggregations=[Aggregation(column="f", function="sum", time_window=timedelta(hours=1))])
b = StreamFeatureView(..., aggregations=[Aggregation(column="f", function="max", time_window=timedelta(days=7))])
a == b                         # True on master
a.to_proto() == b.to_proto()   # False

The SQL registry relies on this equality: _apply_object skips the write when existing_obj == obj, and apply_feature_view uses != to raise FeatureViewPinConflict when a pinned version is combined with a modified definition. On current master those paths happen to still work, because the registry copy of a stream feature view lists the join key in features (fixed in #7033) and that difference makes FeatureView.__eq__ return False. With #7033 applied and without this change, I reproduced end to end with a SQLite-backed SQL registry that:

• re-applying the view with aggregations changed from sum/1h to max/7d is silently skipped (the registry keeps sum/1h), and
• pinning version="v0" while also changing the aggregations does not raise FeatureViewPinConflict.

So this should land before (or together with) #7033.

The fix compares these fields regardless of whether a UDF is set, and adds enable_tiling and tiling_hop_size. mode and udf_string are compared in normalized form (mode_to_string(...), and None treated as ""), matching how to_proto() writes them. Otherwise a view and its from_proto(to_proto(...)) copy would compare unequal, which would make every apply look like a change and create extra version snapshots.

Which issue(s) this PR fixes:

No existing issue; found while working on #7033. I searched open and closed issues and PRs for a duplicate.

Checks

• I've made sure the tests are passing.
• My commits are signed off (git commit -s)
• My PR title follows conventional commits format

Testing Strategy

• Unit tests
• Integration tests
• Manual tests
• Testing is not required for this change

test_stream_feature_view_equality_with_udf covers the UDF branches (only one view has a udf, or both have different udfs). New parametrized test test_stream_feature_view_equality_without_udf (aggregations, mode, timestamp_field, enable_tiling, tiling_hop_size): all 5 cases fail on master and pass with this change; the existing round-trip test test_stream_feature_view_initialization_with_optional_fields_omitted still passes thanks to the normalization above. Manual end-to-end check with a SQLite-backed SQL registry, both with and without #7033 applied: changed aggregations are written, the pin conflict is raised, and re-applying an unchanged view twice creates no new version snapshot. ruff check, ruff format --check and mypy feast pass.

Misc

Developed with Claude Code (AI) on my behalf.

🤖 Generated with Claude Code