Unreliability of the build system

[Jan Stolarek]

T149 has been failing for a very long time (a few weeks at least). I also had process007 fail.

Janek

Dnia wtorek, 25 czerwca 2013, Mateusz Kowalczyk napisał:
> On 25/06/13 15:43, Jan Stolarek wrote:
> > I just ran validation (once without any parameters and once with
> > --normal) and got no failures, except for some tests that always
> > fail. Looks like the problem may be solved.
> >
> > Janek
> >
> > Dnia wtorek, 25 czerwca 2013, Geoffrey Mainland napisał:
> >> On 06/24/2013 09:21 PM, Jan Stolarek wrote:
> >>>> My only claim was that I have a set of steps that can
> >>>> reliably reproduce an error on my system, not that these
> >>>> steps will reliably reproduce it on an arbitrary system,
> >>>> e.g., yours. Still, being able to reliably reproduce an
> >>>> error, even if only on one system, is *very* useful.
> >>>
> >>> just misunderstood your use of the word "reliable". Certainly
> >>> having
> >>
> >> this level of
> >>
> >>> reproducibility is very helpful in debugging, even if it's only
> >>> on one
> >>
> >> machine.
> >>
> >>>> So, to be clear, you see recache errors sporadically and for
> >>>> different sets of tests across different runs, and you are
> >>>> running validate with no extra arguments, single-threaded, on
> >>>> a Linux x86_64 system with an ext4 file system?
> >>>
> >>> Yes, except I would say "very often" instead of "sporadically".
> >>> I'm
> >>
> >> not sure about
> >>
> >>> single-threaded. I run validate without any parameters -
> >>> doesn't it
> >>
> >> use more then one thread in
> >>
> >>> this case?
> >>
> >> It looks like running validate without any parameters will use 2
> >> threads.
> >>
> >> I've pushed a fix (I hope) for this issue. Could you check and
> >> see if it's fixed for you now?
> >>
> >> Geoff
> >
> > _______________________________________________ ghc-devs mailing
> > list ghc-devs at haskell.org
> > http://www.haskell.org/mailman/listinfo/ghc-devs
>
> Which tests are these? I ran `./validate' yesterday on a stock (that
> is, I haven't made any changes to the source) and had about 5-7 tests
> fail and wasn't sure if this was a known issue or something I should
> be reporting.