Touches api/api_connection.h with a comment so determine-jobs.py routes
the api component's integration tests into the bucketed split job
alongside the cover probe. Used to validate the matrix expansion in
#16152.
Do not merge.
Touches cover/cover.h with a comment so determine-jobs.py routes the
cover component's integration tests into the bucketed split job. Used
to validate the matrix expansion in #16152.
Do not merge.
Pull the run_all glob expansion + bucket computation out of main() into
_compute_integration_test_buckets() returning (run_integration, buckets).
The boundary tests now call this helper directly instead of driving
main() through ~14 patched dependencies, which both shrinks the test
helper and removes duplication with the existing test_main_* fixtures.
- Emit each bucket's `tests` as a JSON list of file paths instead of a
space-joined string. The workflow now uses jq to build a bash array,
removing word-splitting / glob hazards on test paths.
- Guard against an empty integration test list after `run_all` expansion:
if the glob returns nothing, suppress the run rather than invoking
pytest with no path argument (which would collect tests outside
tests/integration/).
- Add boundary-case unit tests for the bucketing decision: empty
selection, explicit small lists, exactly threshold (single bucket),
one over threshold (3 buckets), and run_all-with-empty-glob (no run).
When more than 10 integration tests are scheduled (or any change that
triggers run_all, e.g. core/infra changes that would run all 117 files),
fan out the pytest job into 3 parallel matrix entries. Below the
threshold, a single bucket runs as before, so small targeted PRs see no
extra job overhead.
determine-jobs.py now owns the bucketing end-to-end: it expands run_all
into the explicit glob of tests/integration/test_*.py and pre-splits the
sorted list using the same balanced contiguous-partition formula as
script/clang-tidy. The CI workflow consumes the precomputed buckets via
fromJson() in the matrix, mirroring how component-test-batches works,
so no shell-side splitting is needed.
The previous integration-tests-run-all and integration-test-files
workflow outputs are replaced by a single integration-test-buckets
list-of-objects ({name, tests}); the integration-tests gate boolean is
unchanged.
