On Thu, 27 Oct 2022 13:57:47 GMT, Erik Joelsson <er...@openjdk.org> wrote:
> The default set of tests included in GHA is currently defined to be exactly > tier1. That is the set of tests that we have collectively agreed on > represents a reasonably sized set that gives the best coverage for the time > spent (at least that's the intention of tier1). The time spent here is > probably the most important variable, as whatever we put here, everyone runs > all the time in their PRs and PFs. Adding 5 minutes may not seem like much, > but multiplied with platforms and all the users and PRs it adds up to a lot. > Also imagine if more components would want to add to the set, pointing to > this change as a precedent ,adding another 5 minutes, and then another 5 > minutes, eventually the reasonable size of the test set becomes completely > unreasonable. > > My point is that we can't just add more and more to GHA on a whim. The > current agreement is for it to be exactly tier1. If you can't motivate that > your tests should be part of tier1, then I find it pretty hard to motivate > that it should be part of GHA. I don't know enough about these tests to know > if they should fit in tier1/GHA or not, I'm just pointing out that we need to > keep a pretty high bar to avoid bloating the set. Okay. I got your point. So only tier1 tests can be tested in GHA. > > For this particular issue, would it be possible to define a minimal set of > vector api tests that would have discovered the problem in #10807 without > taking 5 minutes? Perhaps the better solution here would be to allow a single > or very few specific tests into tier1? I'm afraid it's hard to prevent bugs like https://github.com/openjdk/jdk/pull/10807 by just adding a single or very few specific tests into tier1. So can we add jdk_vector (which is still in incubation) into jdk/tier1 group? @vnkozlov ------------- PR: https://git.openjdk.org/jdk/pull/10879
