[Sugar-devel] Datastore is broken

[Daniel Narvaez]

On 9 November 2012 09:19, Simon Schampijer <simon at schampijer.de> wrote:
> Ok, I pushed Martin's fix. Sorry we did regress in the latest commit :/
>
> Daniel, the buildbot is very valuable, which are the places a developer
> should look at if he wants to get a quick overview.
>
> So, let's say, Fedora should at least build (given the current status of
> Ubuntu) [1] and then check the tests result here [2]?
>
> Thanks,
>    Simon
>
> [1] http://buildbot.sugarlabs.org/one_line_per_build
> [2]
> http://buildbot.sugarlabs.org/builders/fedora-17-32bit/builds/665/steps/shell_3/logs/stdio

Yup! I tend to use the waterfall rather than one_line_per_build but
that's just a personal preference.

Just keep in mind that there are a few sources of unreliability at the
moment, so a red build doesn''t necessarily mean there is a bug with
your code. Those are:

1 Unreliability of the UI tests. They are timing dependent so it's
hard to get them right.
2 Network issues. Sometimes bender network goes down, sometimes freedesktop etc
3 Random segfaults when buillding gst-ffmpeg
4 Ubuntu tests are still a work in progress (especially the
crash-on-exit issue on 12.10)

I think I'm finally getting there for 1, I haven't seen random test
failures on the fedora slaves in a while and anyway I think I
understand the issues involved now, so I'll hopefully have this solved
soon. 2 cannot really be avoided completely, I hope to be able to mark
those builds with something different than a normal red failure, so
that it's immediately clear they don't need developer attention. I
have no idea what to do about 3 other than waiting Fedora 18 is out,
hopefully it doesn't happen too often. I'm working on 4, it will take
some time but it's just work.

Something else I want to do is to get fix of all the tracebacks
triggered by the UI tests, then we can mark the tracebacks as
warnings, and use buildbot to avoid that kind of regressions too (we
have so many right now that it's difficult to find a new issue in the
logs).

-- 
Daniel Narvaez