12:02 p.m.
Hi,
One of my koji buildjobs seems to be hanging for almost 2 days:
http://koji.fedoraproject.org/koji/buildinfo?buildID=35211
Ralf
9:35 a.m.
On Mon, 11 Feb 2008 19:02:03 +0100
Ralf Corsepius <rc040203(a)freenet.de> wrote:
One of my koji buildjobs seems to be hanging for almost 2 days:
http://koji.fedoraproject.org/koji/buildinfo?buildID=35211
I freed the hung ppc64 task to see if it'll make it through on a
(third?) attempt. The build may have issues on ppc64 that is causing
it to hang.
--
Jesse Keating
Fedora -- All my bits are free, are yours?
12:37 a.m.
New subject: ppc64 gcc4.3 broken? [Was: koji build job hanging for 2 days?]
On Mon, 2008-02-11 at 10:35 -0500, Jesse Keating wrote:
On Mon, 11 Feb 2008 19:02:03 +0100
Ralf Corsepius <rc040203(a)freenet.de> wrote:
> One of my koji buildjobs seems to be hanging for almost 2 days:
> http://koji.fedoraproject.org/koji/buildinfo?buildID=35211
I freed the hung ppc64 task to see if it'll make it through on a
(third?) attempt. The build may have issues on ppc64 that is causing
it to hang.
Hangs again - Looks like a GCC bug to me.
At least,
http://koji.fedoraproject.org/koji/getfile?taskID=410994&name=build.log
ends (or hang) midst of a compilation statement:
...
if g++ -DHAVE_CONFIG_H -I. -I. -I../../.. -I. -I../../../tests/cxts
-I../../../otherlibs/highgui -I../../../cxcore/include -I../../../cv/include
-I../../../cvaux/include -DNDEBUG -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2
-fexceptions -fstack-protector --param=ssp-buffer-size=4 -m64 -mminimal-toc -MT
aoptflowpyrlk.o -MD -MP -MF ".deps/aoptflowpyrlk.Tpo" -c -o aoptflowpyrlk.o
aoptflowpyrlk.cpp; \
then mv -f ".deps/aoptflowpyrlk.Tpo" ".deps/aoptflowpyrlk.Po"; else
rm -f ".deps/aoptflowpyrlk.Tpo"; exit 1; fi
In file included from ../../../cvaux/include/cvaux.h:1458,
from cvtest.h:51,
from aoptflowpyrlk.cpp:42:
[EOF]
=> I am inclined to think ppc64-g++-gcc4.3 has entered a compiler race.
Ralf
11:55 p.m.
New subject: ppc64 gcc4.3 broken? [Was: koji build job hanging for 2 days?]
Ralf Corsepius <rc040203 <at> freenet.de> writes:
At least,
http://koji.fedoraproject.org/koji/getfile?taskID=410994&name=build.log
ends (or hang) midst of a compilation statement:
Be careful there before jumping to conclusions, file output is cached, so the
log doesn't always indicate the exact place the crash takes place (because
there's more output stuck in the cache).
The log now says this:
contour-approx: [32mOk
/bin/sh: line 4: 23972 Terminated ${dir}$tst
FAIL: cvtest
==========================================================
1 of 1 tests failed
so it looks like the testsuite is locking up.
Of course, this may still be a compiler bug (miscompilation), but it could also
be a bug in the code exposed by the new compiler (e.g. an aliasing violation).
Kevin Kofler
1:17 a.m.
New subject: ppc64 gcc4.3 broken? [Was: koji build job hanging for 2 days?]
On Wed, 2008-02-13 at 05:55 +0000, Kevin Kofler wrote:
Ralf Corsepius <rc040203 <at> freenet.de> writes:
> At least,
> http://koji.fedoraproject.org/koji/getfile?taskID=410994&name=build.log
> ends (or hang) midst of a compilation statement:
Be careful there before jumping to conclusions, file output is cached, so the
log doesn't always indicate the exact place the crash takes place (because
there's more output stuck in the cache).
You call looking at build.logs for
several hours and watching them not
receiving any updates/progress for several hours a premature conclusion?
Sorry, but then koji has a usability problem ...
1:38 a.m.
New subject: ppc64 gcc4.3 broken? [Was: koji build job hanging for 2 days?]
Ralf Corsepius <rc040203 <at> freenet.de> writes:
You call looking at build.logs for several hours and watching them
not
receiving any updates/progress for several hours a premature conclusion?
Sorry, but then koji has a usability problem ...
How is that Koji's fault? The operating system and/or C library caches stdio
writes, they aren't output immediately unless fflush is explicitly called, and
Koji has no control on the processes which are outputting to the redirected
stdout (make and its subprocesses). There's no way to fix that in Koji.
Kevin Kofler
1:44 a.m.
New subject: ppc64 gcc4.3 broken? [Was: koji build job hanging for 2 days?]
On Wed, 2008-02-13 at 07:38 +0000, Kevin Kofler wrote:
Ralf Corsepius <rc040203 <at> freenet.de> writes:
> You call looking at build.logs for several hours and watching them not
> receiving any updates/progress for several hours a premature conclusion?
>
> Sorry, but then koji has a usability problem ...
How is that Koji's fault? The operating system and/or C library caches stdio
writes, they aren't output immediately unless fflush is explicitly called, and
Koji has no control on the processes which are outputting to the redirected
stdout (make and its subprocesses).
stdout/stderr are being flushed upon all "\n"'s.
The compiler-call the build job hung on ppc64, on all other
architectures takes in the "order of minutes"
=> the logs are being flushed at the same rate
There's no way to fix that in Koji.
no comment.
1:40 p.m.
New subject: ppc64 gcc4.3 broken? [Was: koji build job hanging for 2 days?]
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Ralf Corsepius wrote:
stdout/stderr are being flushed upon all "\n"'s.
No. There are three different buffering schemes:
- - no buffering
- - line buffering
- - block buffering
The last is always used when the data is written to a file or pipe or
socket or ... In that case you'll only see output when the buffer
overflows.
- --
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (GNU/Linux)
iD8DBQFHuI202ijCOnn/RHQRAjoWAJ4xGJTizrVXzooOojCbd19Eox08hgCeO4ip
YK6Fj2NxBPUH2B1AMUKAFQM=
=4vVn
-----END PGP SIGNATURE-----
5936
days inactive
5942
days old
8 comments
4 participants
participants (4)
-
Jesse Keating -
Kevin Kofler -
Ralf Corsepius -
Ulrich Drepper