RFC: Enabling -DECM_ENABLE_SANITIZERS='address' in jenkins

[Albert Astals Cid]

El Dilluns, 28 de setembre de 2015, a les 16:41:58, Milian Wolff va escriure:
> On Friday, September 25, 2015 11:35:35 PM CEST Albert Astals Cid wrote:
> > El Divendres, 25 de setembre de 2015, a les 05:15:35, Milian Wolff va
> > 
> > escriure:
> > > On Montag, 21. September 2015 20:29:32 CEST Albert Astals Cid wrote:
> > > > El Dijous, 10 de setembre de 2015, a les 22:36:10, Albert Astals Cid
> > > > va
> > > 
> > > escriure:
> > > > > We have this nice ECM module that gives us the option to compile
> > > > > with
> > > > > ASAN.
> > > > > 
> > > > > I'd like to propose that we enable it by default in jenkins.
> > > > > 
> > > > > This way we get all the autotests run with ASAN and potentially
> > > > > catch
> > > > > more
> > > > > bugs/regressions.
> > > > > 
> > > > > Comments?
> > > > 
> > > > This has now been commited.
> > > > 
> > > > https://quickgit.kde.org/?p=sysadmin%2Fci-builder-tools.git&a=commit&h
> > > > =5
> > > > e4
> > > > f2 e673c902e9d3b09aeffe56c20838918c681
> > > > 
> > > > Watch out for tests failing, let's hope it doesn't find anything bad
> > > > :D
> > > 
> > > \o/
> > > 
> > > Thanks for taking the initiative here, Albert! Much appreciated!
> > > 
> > > Also, did you think of additionally enabling UBSan where available? It
> > > can
> > > be combined with ASan and should thus find even more issues in our
> > > codebase!
> > 
> > I've never tried the undefined sanitizer so can't comment on how "correct"
> > or "pedantic" its checks are.
> > 
> > Have you? What's your opinion on it?
> 
> Yes, it works nicely for the simple examples I've tried, see also:
> 
> http://developerblog.redhat.com/2014/10/16/gcc-undefined-behavior-sanitizer-> ubsan/
> > Also it seems it doesn't really crash(stop the app like asan does, so it
> > would not be of much use unless someone spends lots of time parsing the
> > tests logs, no?
> 
> It could either be scripted, or we can try to pass -fno-sanitize-recover,
> which should make it crash instantly on the first error (indeed, it does in
> my test).

Have you compiled some frameworks with those flags on? Are we confident it 
won't blow up in a red-festival?

Cheers,
  Albert

> 
> Bye