webkit-patch optimize-baselines should add/delete files in batches from the vcs
Created attachment 144153 [details] Patch
Comment on attachment 144153 [details] Patch Wow. That was easy. I suppose this is covered by existing tests? If so, the changelog description should say so.
Do we pass all the tests from garden-o-matic to a single "webkit-patch optimize-baselines"?
(In reply to comment #2) > (From update of attachment 144153 [details]) > Wow. That was easy. I suppose this is covered by existing tests? If so, the changelog description should say so. I will double-check the test coverage and update the change log (and add tests) accordingly. We optimize one test at a time. If we start to rebaseline multiple tests at once using garden-o-matic (or whatever) we will probably want to either extend optimize-baselines to take a list of tests or defer the scm modifications and do it all at once at the end like we're gonna do with rebaseline-test.
(In reply to comment #4) > (In reply to comment #2) > > (From update of attachment 144153 [details] [details]) > > Wow. That was easy. I suppose this is covered by existing tests? If so, the changelog description should say so. > > I will double-check the test coverage and update the change log (and add tests) accordingly. > > We optimize one test at a time. > > If we start to rebaseline multiple tests at once using garden-o-matic (or whatever) we will probably want to either extend optimize-baselines to take a list of tests or defer the scm modifications and do it all at once at the end like we're gonna do with rebaseline-test. Makes sense. I suppose it's rare that garden-o-matic does multiple tests at once. Although, I've had times when I've queued up tons of rebaselines that in theory could be batched. But maybe with all these performance improvements it won't matter.
Comment on attachment 144153 [details] Patch It's important to do all the deletes before the adds, which you've done. Thanks.
(In reply to comment #5) > (In reply to comment #4) > > (In reply to comment #2) > > > (From update of attachment 144153 [details] [details] [details]) > > > Wow. That was easy. I suppose this is covered by existing tests? If so, the changelog description should say so. > > > > I will double-check the test coverage and update the change log (and add tests) accordingly. > > > > We optimize one test at a time. > > > > If we start to rebaseline multiple tests at once using garden-o-matic (or whatever) we will probably want to either extend optimize-baselines to take a list of tests or defer the scm modifications and do it all at once at the end like we're gonna do with rebaseline-test. > > Makes sense. I suppose it's rare that garden-o-matic does multiple tests at once. Although, I've had times when I've queued up tons of rebaselines that in theory could be batched. But maybe with all these performance improvements it won't matter. I think there still could be wins from pipelining multiple tests through and batching more things. Since the 'rebaselineall' http entry point does take multiple lists, it won't be hard to extend the backend if we need it to. Even with all of these improvements, Rebaselining still is multiple seconds per test over a fast network. We should have somebody test this from asia or europe to see how it is from there.
I nominate Noel once this and https://bugs.webkit.org/show_bug.cgi?id=87451 are committed.
Happy to do it. Let me know what you need when you're ready.
Committed r118838: <http://trac.webkit.org/changeset/118838>
Created attachment 146485 [details] webkit-patch-rebaseline-expectations.txt Timing: rebaseline one test from AEST time zone.
(In reply to comment #11) > Created an attachment (id=146485) [details] > webkit-patch-rebaseline-expectations.txt > > Timing: rebaseline one test from AEST time zone. Unfortunately, webkit-patch rebaseline-expectations doesn't yet do the rebaselines in parallel. We only do them in parallel from garden-o-matic. So, this 21 seconds is the time to do it serially. Looking at the output here, doing the rebaselines in parallel should cut the time down to ~5 seconds.
5 seconds sounds good. "Let me know what you need when you're ready." ... and so with no news, I just punted that webkit-patch rebaseline-expectations had the new fancy :)