fix: remove XFAIL from GroupByQuery test (now supported in zvec v0.4.0)

[s2x]

Summary

tests/bug_0002.phpt is marked --XFAIL-- because zvec v0.2 had GroupByQuery as "Coming Soon" (returned all docs in a single group with empty group_by_value). In zvec v0.4.0, the implementation exists — CollectionImpl::GroupByQuery() calls sql_engine_->execute_group_by(). The test should now pass.

What to do

1. Remove --XFAIL-- lines (3-4) from tests/bug_0002.phpt
2. Run the test: php run-tests.php tests/bug_0002.phpt
3. If it passes — done
4. If it still fails — investigate and fix

Files affected

• tests/bug_0002.phpt — remove --XFAIL-- section

Existing test content

The test creates a collection with category and name string fields and an FP32 vector field, inserts 5 documents across 3 categories, then calls:

$groups = $c->groupByQuery(
    'v', [0.5, 0.5, 0.5, 0.5],
    groupByField: 'category',
    groupCount: 2,
    groupTopk: 3,
);

Expects multiple groups with non-empty group_value.

Expected result

Test passes — GroupByQuery returns properly grouped results with correct group_by_value strings.

Dependencies

None — no code changes, just test annotation change and verification.

Rebuild

Only if the FFI library is stale:

./build_zvec.sh

[s2x]

Investigation Results

After investigating, GroupByQuery still does not work correctly in zvec v0.4.0. The issue is deeper than expected.

Root Cause

The _zvec_group_id column in the search results is always empty strings. This happens because:

1. The SQL engine correctly extracts group_by params (field name, group_count, group_topk)
2. The VectorRecallNode correctly sets up the group_by function callback on query params
3. But the underlying Proxima vector search engine never calls set_group_params() or set_group_by() on the search context
4. The search context's group_num_ stays at 0, so group_by_search() returns false
5. The regular (non-grouped) search path is used instead of the grouped search path
6. All results get an empty string _zvec_group_id

Debug Evidence

_zvec_group_id: ["", "", "", "", ""]

Even though the category field has correct values ("books", "electronics", "food"), all group IDs are empty.

Verification

• Regular query() works correctly and returns proper results with correct field values
• GroupByQuery() returns 1 group with empty group_value containing all 5 docs (because all are grouped under "")
• XFAIL test bug_0002.phpt still fails as expected (XFAIL is accurate)

Options

1. Keep XFAIL with updated reason - the feature is still not functional in v0.4.0
2. Implement workaround in FFI bridge - do client-side grouping after a regular vector query (fetch group_by field values and group in C++/PHP)
3. Wait for upstream fix in a future zvec version

Updated XFAIL Reason

GroupByQuery API exists in zvec v0.4.0 but grouping still doesn't work - _zvec_group_id column is always empty because Proxima engine's group_by context params are not propagated to the search context

[s2x]

After deeper investigation: the GroupByQuery bug is actually upstream in Alibaba's zvec v0.4.0, not just a "Coming Soon" feature.

Evidence from upstream

1. Empty test: tests/db/collection_test.cc has TEST_F(CollectionTest, Feature_GroupByQuery) {} — completely empty stub
2. Python stub: __init__.pyi has def GroupByQuery(self, arg0: ...) -> list[...]: ... — placeholder args
3. No GitHub issues/PRs: zero issues or PRs about GroupByQuery in github.com/alibaba/zvec
4. Not on roadmap: Roadmap issue #309 doesn't mention GroupByQuery
5. Root cause confirmed: _zvec_group_id column is always empty because the Proxima engine's set_group_params() / set_group_by() context methods are never called — the group_by params are set up correctly in the SQL engine layer but never propagated to the vector search engine

Current status

The GroupByQuery API and pipeline exist (collection method, SQL engine, vector recall node, result parsing), but the actual grouping step in the Proxima vector search is not implemented. This is an upstream limitation, not something we can fix in php-zvec without a workaround.

Recommendation: Keep this issue open and XFAIL until upstream implements GroupByQuery properly, or implement a client-side grouping workaround in php-zvec.