This has irked me before too. An obvious compromise would that the backout proceeds, but it must include a test that would have failed on CI when the patch was landed. This puts the onus on the owners of the broken functionality to make sure that this supposedly-critical functionality has basic smoketest coverage, and gives the developer of the backed-out patch a way to verify that their code is correct on CI.
Backouts should require tests. On Thu, Oct 16, 2014 at 4:33 AM, Ehsan Akhgari <ehsan.akhg...@gmail.com> wrote: > On 2014-10-15, 10:19 PM, Karl Tomlinson wrote: > >> Jonas Sicking writes: >> >> But any type of regression is cause for backout. >>> >> >> While I agree regressions are bad, this isn't the usual process. >> >> If it were, then I wouldn't bother filing bugs, but merely back >> out the offending change. >> >> There is some kind test for whether the regression costs more than >> the improvements made, but it comes down to a judgement call from >> the module owner AIUI. >> >> Regressions that sit in the tree make it dramatically much harder to >>> write and test other patches. It's generally much better to back the >>> offending patch out to allow everyone else to go at full speed. >>> >> >> Perhaps it is, but this would be quite a change in process. >> Some kind of policy or guidelines would be helpful, or it could >> well get out of control. >> >> Backouts usually cause regressions too. >> > > In my experience, regressions that break something in a way that makes > dogfooding difficult are open to a backout without questions asked policy > (but often times in practice we'd try to reach out to the author and the > folks who know the area of the code). For other regressions we typically > file follow-up bugs. > > Note that the definition of what makes dogfooding difficult is has not > been entirely consistent all the time. > > Cheers, > Ehsan > > > _______________________________________________ > dev-platform mailing list > dev-platform@lists.mozilla.org > https://lists.mozilla.org/listinfo/dev-platform > _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
