I suggest adding an Auto branch between Try and Central. Advantages:
* Pulling from Central is safe, because it only gets csets that passed both Try (as individual developer pushes) and Auto (as a group).
* Infrastructure load will be slightly lower, because everyone's pushes to Try will have a working Central base.
* We can use jcranmer's "sparse test matrix" trick for some Try pushes, knowing that Auto will catch platform-specific failures before they reach Central.
* In Scenario D (when subtle patch interactions cause build or test failures), automation can move on to another set of Try landings, giving sheriffs time react without the pressure of tree closure. (Sheriffs can then fill in the test matrix, redo the merge in a way that keeps mozilla-central free of non-working changesets, or just let the group of patches bounce.)
* We can add expensive tests (ASan, MSan, TSan, UBSan) without requiring every Try push to run them.
To clarify my proposal, the ideal scenario looks like: 1. Five patches land separately on Try. 2. (They pass quick(?) tests and don't conflict in obvious ways.) 3. Automation lands the combination (merged or rebased) on Auto. 4. (The combination passes full tests.) 5. Automation pushes the resulting changeset to Central. 6. (No additional tests run, because they'd be redundant.) _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
