[vlc-devel] breaking build

[Rémi Denis-Courmont]

Hi,

If you use asan, how come you missed the packetizer test leaking for several weeks? It's got nothing to do with my hardware.

Le 21 octobre 2019 16:06:07 GMT+03:00, Thomas Guillem <thomas at gllm.fr> a écrit :
>
>On Mon, Oct 21, 2019, at 14:04, Rémi Denis-Courmont wrote:
>> Hi,
>> 
>> With all due respect, this fixation on Gitlab CI seems delusional to
>me. It took several weeks for anybody to report an obvious and
>universal memory leak in the unit tests. Because nobody uses the leak
>sanitizer on regular basis except me.
>> 
>> As for the video problem, it only occurs in smoke testing not in unit
>testing, so any CI is entirely helpless there.
>> 
>> It's rather ironic that I was the one to object against adding the
>sanitizers to the configure script, and I'm semingly the only one to
>use them systematically in the end.
>
>A lot of devs are using asan. Personally, I always use it when I work.
>
>The bug you caught was not reproducible on my main workstation because
>I don't have a nvidia GPU and it happens with a specific user
>interaction. When I tested VDPAU an a secondary desktop, I forgot to
>enable ASAN on this specific case (because I didn't use my usual helper
>scripts).
>
>Human errors like that will continue to happen. I think we can live
>with it. They are generally quickly fixed by the author.
>
>I already had to fix your errors, like you already had to fix mine. In
>that case, either we fix it, assign a ticket or signal it on the
>mailing list. This process seems OK to me.
>
>> 
>> Besides premerge CI will fail in so many other ways that it's not
>even funny. It's great for an admin who has the power to fix the CI
>system or to overrule it. It sucks big time for everyone else who gets
>stuck whenever:
>> - one platform is down,
>> - contrib or sync fails due to network issue,
>> - build fails due to transient out-of-ressource situation,
>> - nondeterministic test case breaks pseudorandomly,
>> - system update introduces unrelated regressions (especially likely
>with leaksan).
>> And if it's gating, then it'll be miserable for people working in
>different continent or hobbyists, who'll have nobody to get live
>support from.
>> 
>> 
>> No, the only way to fix this is that developers use sanitizers, which
>was the whole point of adding them to configure.
>> 
>> Le 20 octobre 2019 21:01:25 GMT+03:00, Jean-Baptiste Kempf
><jb at videolan.org> a écrit :
>>> Yo,
>>> 
>>> On Sun, Oct 20, 2019, at 15:33, Rémi Denis-Courmont wrote:
>>>> The points are that some people ostensibly don't run the test
>suite, and/or 
>>>> smoke tests and/or enable the ADDR and UB sanitisers, causing
>easily avoidable 
>>>> regressions in master.
>>> 
>>> Agreed.
>>> 
>>> Breaking the build or breaking tests should not happen anymore. This
>is not normal.
>>> 
>>> We will then move to gitlab to get the CI to run pre-merge, as soon
>as possible.
>>> CI should catch all of those.
>> 
>> -- 
>> Envoyé de mon appareil Android avec Courriel K-9 Mail. Veuillez
>excuser ma brièveté. 
>> _______________________________________________
>> vlc-devel mailing list
>> To unsubscribe or modify your subscription options:
>> https://mailman.videolan.org/listinfo/vlc-devel

-- 
Envoyé de mon appareil Android avec Courriel K-9 Mail. Veuillez excuser ma brièveté.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/vlc-devel/attachments/20191021/a0f9b29c/attachment.html>