lsm / HyperNeo / 31912746341

79%

push

github

refactor(ci): daemon online shards via dir-globs + hash-split [#912] (#2497)

* refactor(ci): daemon online shards via dir-globs + hash-split [#912]

Replace every hand-listed online test shard in main.yml with directory
globs consumed by a new scripts/test-online.sh resolver (built on the
hash-split runner from #911). A module is either a whole directory or a
stable-hash bucket of one, so a new online test file auto-routes to a
shard with no YAML edit; rebalancing is a one-number split-count edit.

- scripts/test-online.sh: module config + resolver + --verify (glob
  coverage, bucket partition, matrix↔config wiring both directions,
  full-disk ownership walk). Wired into the CI check job.
- main.yml test-daemon-online: drop all test_path rows; runner resolves
  ${{ matrix.module }} via test-online.sh. New splits sized from CI
  test-step medians: space x4 (was space-2, the 212s pole -> ~121s),
  rpc x6, rewind x2, features x3. mock_sdk/timeout preserved per module.
- validate-test-matrix.sh: section 3 now sources test-online.sh and
  derives file ownership from resolution (file->owners map) instead of
  hand-listed RPC/SPACE/FEATURES/REWIND arrays; runner checks pin the
  test-online.sh resolution; real-api-tests.yml per-file rows unchanged.
- validate-test-matrix.test.ts: +3 mutation tests (reintroduced
  test_path row, resolver bypass, unresolvable matrix module).

Projected balance (test-step medians + ~50s job overhead): slowest job
drops 212s -> ~145s (convo; space pole -> ~121s), 19 -> 24 jobs.

* refactor(ci): address review on #2497 — guard hardening + runner fail-loud

P2s:
- validate-test-matrix.sh: restore the marker-position check the rewrite
  dropped — the resolution must reach vitest as the post-`vitest run`
  positional ($paths), not merely appear somewhere in the command
  (resolving to /dev/null ran the whole suite while every textual
  detector passed).
- main.yml runner: set -o pipefail + `[ -n "$paths" ] || exit 1` so an
  empty/... (continued)

42623 of 57973 branches covered (73.52%)

Branch coverage included in aggregate %.

58104 of 69939 relevant lines covered (83.08%)

3860.07 hits per line