[Slackbuilds-users] Updates - 20170729.1
Alexander Verbovetsky
alik at ejik.org
Sat Jul 29 18:17:03 UTC 2017
On Sat, Jul 29, 2017, at 19:28, Matteo Bernardini wrote:
> 2017-07-29 15:34 GMT+02:00 Alexander Verbovetsky <alik at ejik.org>:
> > On Sat, Jul 29, 2017, at 16:09, Willy Sudiarto Raharjo wrote:
> >> >> academic/reduce-algebra: Force a single make job.
> >> >
> >> > May I ask why this is needed?
> >>
> >> In some environment where it has multi cores, using parallel job
> >> may lead to race condition and build failure.
> >
> > Is there a way to catch such failure? I always build reduce-algebra
> > on 8 cores without problems.
>
> Hi Alexander,
>
> I added that because I tried on a full clean installation with a lot
> of different MAKEFLAGS=-jn options, with n ranging from 2 to 32 and
> the build crashed twice, in a way not even reproducible. so I opted
> for the single make job that should be safe for everyone. if you
> prefer, you can also remove it forcing the single job but I suggest
> you to put in the README something like: "be aware that the build may
> crash unexpectedly if using multiple make jobs"
On Sat, Jul 29, 2017, at 19:46, David Spencer wrote:
> > Is there a way to catch such failure? I always build reduce-algebra
> > on 8 cores without problems.
>
> I don't know a good way to make them crash, sometimes it is hardware
> dependent. I have a P4 (one core of course) with hyperthreading that
> is very good at crashing with -j3, but I don't use it much :)
>
> If you have a crash, one way of checking is to look at other distros
> to see if they have added '-j1'. repology can help with this. For
> example,
> https://aur.archlinux.org/cgit/aur.git/tree/PKGBUILD?h=reduce-algebra-svn
> has -j1 already, so Matteo is not a freak :p
Thank you for testing and suggestions!
I run the build several time, measuring how much time it takes and
observing the cores activity via htop. And I see absolutely no changes
from adding "-j1". It takes exactly the same time and use all cores
except for some short periods when only one core is 100% busy, but these
periods happens during the same parts of build with "-j1" and also
without.
So, I think that somehow -j1 doesn't work! Is this possible?
As for Arch, the above "reduce-algebra-svn" was stopped in 2013 and
superseded by
https://aur.archlinux.org/cgit/aur.git/tree/PKGBUILD?h=reduce-svn
which has no -j1 :-)
Best regards,
Alexander
More information about the SlackBuilds-users
mailing list