Re: Bug#758671: hurd-i386 libpthread can't create thousands of threads (fixing pulseaudio test suite on hurd)
On Wed, 2014-08-20 at 11:36 +0200, Samuel Thibault wrote:
> Svante Signell, le Wed 20 Aug 2014 10:24:01 +0200, a écrit :
> > On Tue, 2014-08-19 at 20:37 +0200, Samuel Thibault wrote:
> > > Samuel Thibault, le Tue 19 Aug 2014 20:36:35 +0200, a écrit :
> > > > It creates thousands of threads, and it seems the stack addresses keep
> > > > going up, so we apparently have a problem of stack reuse here, so
> > > > something on the libpthread side, not the pulseaudio side.
> > >
> > > That's possibly related with the EAGAIN error encountered in the perl
> > > testsuite.
> > Maybe also causing most of the test errors in the swi-prolog testsuite.
> > Do you have the updated glibc patch cvs-libpthread_guardsize.diff as a
> > debian patch, or can I directly use the one at anonscm.d.o?
> Well, just running svn log, svn diff & such would have given you the
> answer immediately, instead of having to wait for my mail...
> cvs-libpthread_guardsize.diff was just refreshed with the additional
> Anyway, to save you the build, I've put the patched libpthread I built
>From perl 5.20 testsuite:
re/regexp_qr_embed_thr.t .......................................... ok
I'll file bugs with patches for tests
to disable them for Hurd.
Remains to discuss with upstream about:
> not ok 11 - peer from recv() should be empty or the remote name
>From swi-prolog 6.6.6 testsuite:
Running scripts from thread ....................................... done
1,545 garbage collections gained 124,260,304 bytes in 0.000 seconds.
6,534 atom garbage collections gained 288,358 atoms in 0.000 seconds.
Stack shifts: 738 local, 919 global, 450 trail in 0.000 seconds.
2 threads, 3,329 finished threads used 0.000 seconds.
All tests passed
Only one failing test left: src/Tests/GC/test_ch_shift.pl (a segfault)