[webkit-dev] Trees on fire, how commit-bot does and does not help

[Adam Barth]

On Thu, Jul 8, 2010 at 11:05 AM, Darin Adler <darin at apple.com> wrote:
> On Jul 8, 2010, at 10:58 AM, Xan Lopez wrote:
>
>> Oh, and might this serve as a ping for whoever set the trees on fire... at least some of it seems related to the refPtr work that's been going on (see https://bugs.webkit.org/show_bug.cgi?id=41823)
>
> Yes, I’ll be working on this more today. I could use some help diagnosing what’s going wrong on various platforms.
>
> The discussion of the commit-bot overlooks the fact that the bot does not prevent problems like this, ones that affect only certain platforms.
>
> I think we‘re conflating things here. The commit bot enforces one particular item; it makes sure all the layout tests pass on a single platform. But few of the real world problems I end up dealing with fall into that category.
>
> The reason the commit-bot doesn’t work for me is that it randomizes the time my patch lands, and makes it hard for me to be there at that time to follow up with what we learn from all the other bots.
>
> What could solve that would be a much fancier version of the early warning system that could do all the same testing that the build bots do, before checking in. The most useful form of that would be something not built into the commit bot, but something designed to be useful for iterative development.

Yeah, I definitely agree that try servers would be useful.  I think
it's a matter of setting up a second buildbot master configured for
try jobs and attaching slaves.

Adam