Re: Timeouts, Was: Bug#571745: nmu: highlighting-kate_0.2.5-4
Excerpts from Marc 'HE' Brockschmidt's message of Sáb Mar 06 06:26:43 -0300 2010:
> Joachim Breitner <email@example.com> writes:
> > Am Freitag, den 05.03.2010, 14:53 +0100 schrieb Marc 'HE' Brockschmidt:
> >> Joachim Breitner <firstname.lastname@example.org> writes:
> >> > nmu highlighting-kate_0.2.5-4 . i386 kfreebsd-i386 amd64 kfreebsd-amd64 . -m "rebuild against changed pcre-light ABI"
> >> Done. Got lost in the cracks somehow. Do you know why its hitting
> >> timeouts on other archs? Is this something we could fix?
> > Probably just because it needs a lot of memory and CPU power, and the
> > other arches have difficulties compiling the package in time.
> This seems a bit more problematic than I originally assumed.
> Consider for example agda on sparc:
> sparc is not a fast architecture, but also not the slowest one supported
> by Debian. In this case, it failed to build the package because a single
> file needed more than 500 minutes to compile. If this is indeed not due
> to a bug, but expected behaviour, it seems not supportable anymore.
It seems that the build time of agda for some arches grew very much with
ghc6-6.12.1, specially hppa and sparc:
agda 2.2.6-2 2.2.6-3
hppa 04:57:19 60:57:00+
kfreebsd-i386 00:17:10 00:11:39
mips 02:35:19 03:54:06
mipsel 02:34:59 04:04:43
powerpc 00:26:37 00:27:16
s390 00:47:49 02:00:55
sparc 03:23:58 08:33:05+
For ghc6 this didn't happened; when the time grew, it grew not very much:
ghc6 6.10.4-1 6.12.1-12
alpha 13:05:04 14:40:30
amd64 01:24:19 01:31:30
armel 100:47:35 127:04:23
hppa 12:25:45 16:08:58
kfreebsd-i386 01:21:37 02:14:17
mips 20:03:36 15:31:51
mipsel 20:03:22 16:25:52
powerpc 04:59:03 03:25:32
s390 07:20:55 03:24:32
sparc 50:23:08 07:22:11
Maybe this is related to something in the agda's code interaction with the new
version of ghc6 in hppa and sparc; possibly a bug somewhere.
> How should slower archs (like mips*, armel) handle such packages? Is
> splitting this up in smaller files an option?
Splitting in smaller files is not very different from what I proposed in
debian-haskell of using a verbose build. The good option in this case would be
adding --ghc-options=-v2 to Setup configure. I think that, if these
packages are not going to be dropped for these arches, this is a good enough
workaround. At least it seems to me to be better than what's being used now: