On Sat, Sep 29, 2012 at 08:58:19AM -0700, Steve Fink wrote: > On Sat 29 Sep 2012 07:53:36 AM PDT, Chris AtLee wrote: > >One proposal that's been made elsewhere > >(https://bugzilla.mozilla.org/show_bug.cgi?id=791385) is to have a > >soft limit of one active push per developer on try. If you try and > >push a 2nd time before your previous jobs are all finished, you will > >be asked to cancel your previous jobs. There would be some kind of > >manual override that would allow you to push additional patches. > > Alternatively, you could go way back to my WONTFIXed bug > https://bugzilla.mozilla.org/show_bug.cgi?id=593096 with a similar > idea -- by default, pushing to try cancels any previous jobs for the > same user. It allows a 'project' token for people who have multiple > unrelated things going on. (We'd probably want a general override > token when you're looking into an intermittent failure.) You could > even default the token to the bug mentioned in the tipmost patch. > > From a developer's point of view, I would prefer this because the > cancel is done for me, and it matches my 90% case -- usually, if I'm > pushing before a previous push is fully done, I don't care about the > remaining results of the previous push. And you don't have to worry > about people finding it easier to manufacture an override token than > going to a web cancellation UI.
I'm not sure this is still the case, but something i noticed is that cancelling windows builds actually doesn't cancel them until the current step they are doing is finished. So, if you cancel a windows build just after it has actually started to build (that is, after hg clone and other preparation steps, which, btw, take an embarassingly enormous amount of time), you've wasted 40 minutes of windows builder time, because that's the time it will take before it actually is cancelled. It's a chance that we now are using pymake, which makes it ~40 minutes (much better than the 1 hour+ times we had before). I'll also note that a lot of tester time is probably wasted by the fact that trychooser doesn't allow to disable extra testing architectures. For instance, if you push -p win32 -u all, you get win7 tests *and* winxp tests, even when you don't actually care about the winxp tests. And cancelling winxp tests only with self-serve is an annoying process. Mike _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
