What this PR does / why we need it:

feast plan shows spurious changes to a feature view's batch_source / stream_source that differ only in meta.created_timestamp / meta.last_updated_timestamp. That's the noise reported in #5573.

Root cause. DataSource.__init__ sets created_timestamp and last_updated_timestamp to "now", and _set_timestamps_in_proto writes them to the source proto's meta. The registry copy of a feature view and the one just declared in the repo therefore always carry different source timestamps. diff_registry_objects only walks the spec fields once current != new, but then it compares each field as a whole proto. As soon as a feature view has any real change (tags, description, ttl, …), its embedded source is reported as changed too. The same thing happens for:

• stream_source (and the batch_source nested inside a Push/Kafka source)
• LabelView.source
• ODFV request sources
• the top-level meta field when a data source itself is diffed

Minimal reproduction on current master:

registered = FeatureView.from_proto(make_fv().to_proto())  # built earlier
declared = make_fv()                                        # built by feast plan
declared.description = "changed"
[d.property_name for d in diff_registry_objects(registered, declared, "feature view").feast_object_property_diffs]
# ['batch_source', 'description']   <- batch_source only differs in meta timestamps

It happens the same way with FileSource and BigQuerySource, so it's not BigQuery-specific.

Fix. Before comparing, diff_registry_objects now works on copies of both specs with meta cleared on every DataSource message they contain (walked recursively through message, repeated and map fields). Real source changes, like a different path, table or timestamp field, are still reported. The printed val_existing / val_declared also no longer include the timestamp noise. Whether an object counts as changed at all still comes from the objects' __eq__, so this PR doesn't change it.

Which issue(s) this PR fixes:

Fixes #5573

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

New test_diff_registry_objects_ignores_data_source_meta in tests/unit/diff/test_registry_diff.py builds the registered and declared objects from separately constructed sources, the way feast plan does. The existing tests share a single source instance, which hides the bug. It checks three things:

• a tag-only change reports only tags, with a push source wrapping a file source so the nested batch_source is covered
• a real path change still reports batch_source / stream_source
• diffing a data source directly reports timestamp_field and not meta

It fails on master (['tags', 'batch_source', 'stream_source'] == ['tags']) and passes with the fix. The rest of tests/unit/diff/test_registry_diff.py passes. ruff check and ruff format --check are clean on both files.

Local environment note: my venv doesn't have the GCP extra, and tests/utils/data_source_test_creator.py imports google.cloud.bigquery at module level. To run this test file locally I stubbed that import out. I also don't have types-protobuf, so mypy only reports the same missing-stubs error it gives for every other google.protobuf import.

🤖 Generated with Claude Code