On Wed, Jul 22, 2009 at 5:59 PM, Paweł Hajdan Jr.
<[email protected]>wrote:
> One thing that would help us keep the tree more green is avoiding compile
> failures. A compile failure is very bad, because without binaries the tests
> can't run, and then we have to wait for all of them to run, which may reveal
> additional failures etc.
> I'm actually surprised by some failures on buildbot, but at least one thing
> was not surprising for me: Windows Release compile failures when the Debug
> compiles fine (because we don't have Release trybot).
>
How often does something run in Windows when compiled with the release
configuration but not the debug? I've definitely seen it, but I'm not sure
it's terribly common. My guess is that there are other causes of the build
breaking that should be addressed first. Are there any stats on this?
My gut feeling is that many of the build breaks are for things that never
passed on a try bot. For example, WebKit gardening patches almost never
work on the try bots so we just ignore them. I think working on stuff like
this will bear more fruit.
Not to mention that each bot costs a lot in terms of the machine,
power, maintenance time, etc.
> What do you think? Do you have any ideas how we could avoid more compile
> failures, even if they are not possible to apply now due to lack of
> resources? (for example adding trybots, which seems to not happen soon).
>
> I was also thinking about allowing simple check-ins when the tree is
> "waiting for cycle" state (when the sheriff wants to verify that bots cycle
> green after a lot of redness). The status would say ("Tree closed, waiting
> for cycle; ask sheriff to commit a simple change"), or maybe some
> abbreviation for that. It would help people getting code in, and the sheriff
> could require really a lot from that change (like full green trybot pass
> etc). What do you think about that (especially sheriffs)?
>
I think you can always ask the sheriffs if you can put something small in.
I don't see the point of making any such message policy or a convention.
That said, unless it doesn't compile or is REALLY obviously OK, I don't
think it's a good idea.
--~--~---------~--~----~------------~-------~--~----~
Chromium Developers mailing list: [email protected]
View archives, change email options, or unsubscribe:
http://groups.google.com/group/chromium-dev
-~----------~----~----~----~------~----~------~--~---
