gluesql / gluesql / 26680078255

99%

push

github

Keep primary key access path in index planner (#1919)

The planner pipeline runs `plan_primary_key` before `plan_index`. When a
query combines a primary key equality predicate with a secondary index
equality predicate via AND (e.g. `WHERE id = 1 AND name = 'x'`), the
primary key planner first moves `id = 1` into a `PrimaryKey` access path,
leaving `name = 'x'` as the selection.

The index planner's selection branch lacked the `index.is_none()` guard
that its order-by branch already has, so it overwrote the `PrimaryKey`
access path with the `idx_name` secondary index and dropped the remaining
selection. The `id = 1` predicate was silently lost, returning extra rows.

Add the `index: None` guard to the selection branch so an access path
chosen by an earlier pass is preserved, mirroring the order-by branch.

19 of 20 new or added lines in 1 file covered. (95.0%)

44388 of 45233 relevant lines covered (98.13%)

75504.89 hits per line