mirror of
https://github.com/ajisaacs/OpenNest.git
synced 2026-10-11 08:34:13 -04:00
merge: integrate ci/parallel-tests
Runs the four cross-platform suites as a matrix and the six synthetic Irregular nests in their own job, with a fail-closed final "tests" check. Conflict in ci.yml: take the branch's Linux jobs and keep master's windows-desktop job unchanged after them. actionlint and the aggregate checker self-tests pass on the merged file.
This commit is contained in:
commit
bec4db21a8
4 files changed
+129
-12
No files matched your search
@@ -73,9 +73,21 @@ The unit checks use labeled synthetic CSV rows and a real watchdog process-tree
|
||||
probe. The end-to-end command uses actual benchmark results and the production
|
||||
independent layout validator. This gate supplements, never replaces, the existing
|
||||
geometry, contract, cancellation, sequencing, posting and cross-platform unit suites.
|
||||
GitHub's cross-platform workflow runs both the checker tests and the six-fixture gate.
|
||||
Windows runtime acceptance is separate; a Linux pass does not certify Windows tree
|
||||
termination or desktop/shop interactions.
|
||||
GitHub's cross-platform workflow runs both the checker tests and the six-fixture gate
|
||||
in a separate job, concurrently with four independent unit-suite jobs (main, Engine,
|
||||
IO and Server). Unit jobs retain the default Debug configuration and do not cancel
|
||||
siblings on failure. The final `tests` check requires the entire unit matrix and
|
||||
synthetic job to succeed; failed, cancelled or skipped dependencies cannot produce
|
||||
a green final check. This preserves the existing check name without changing branch
|
||||
protection or removing tests. `python3 scripts/test_ci_results.py -v` exercises the
|
||||
same fail-closed aggregate command used by CI. The six-minute step limit and
|
||||
`--parallel 2` synthetic-runner limit are unchanged.
|
||||
|
||||
This fan-out removes inter-suite serialization, not individual geometry-test work.
|
||||
It uses additional independent runners and repeats restore/build setup; actual wall
|
||||
time depends on the slowest job and runner queueing. It is not a CPU speed claim or
|
||||
a reason to skip correctness fixtures. Windows runtime acceptance is separate; a
|
||||
Linux pass does not certify Windows tree termination or desktop/shop interactions.
|
||||
|
||||
## Validity is not speed
|
||||
|
||||
|
||||
Reference in new issue
Block a user