In general, when extension A depends on extension B, and A runs the mediawiki-extensions-hhvm job but B doesn't, B could break something HHVM-specific without noticing, that then causes A's jobs to fail. This is what caused T91967: Flow doesn't run the HHVM tests, but Thanks and VE do.
In general I would argue that this situation shouldn't be allowed to exist: if A depends on B, either both or neither should run the HHVM tests.
| Subject | Author | Repo | Branch | Lines +/- | |
|---|---|---|---|---|---|
| Enable 'mediawiki-extensions-{hhvm,zend}' in all extensions it covers | Krinkle | integration/config | master | +5 -13 |
I don't think this is actually HHVM-specific. I think the issue is that the HHVM runs a broader suite of tests (more repos). However, the point still holds.
Change 195335 had a related patch set uploaded (by Krinkle):
Enable 'mediawiki-extensions-{hhvm,zend}' in all extensions it covers
Change 195335 merged by jenkins-bot:
Enable 'mediawiki-extensions-{hhvm,zend}' in all extensions it covers
Fixed.
Verified via recheck on https://gerrit.wikimedia.org/r/#/c/195330/.
It was basically already set up that way, but Flow got left out somehow when the list was last synchronised.