Thu 03 Dec 2015 04:35:09 PM UTC, comment #20:
I was the original reporter and I can no longer reproduce this on a Windows XP virtual machine. I am using a recent development tip 20787:62564952e161 which I have cross-compiled with MXE. Closing report.
|
Wed 01 Apr 2015 08:02:18 PM UTC, comment #19:
@Rik: I agree. Time to do the release. Just wanted to
make sure that those 2 problems were not "forgotten."
The current dev system is vastly better than the
(long ago) previous release.
|
Wed 01 Apr 2015 06:06:17 PM UTC, comment #18:
@Michael: Not the memory leaks. They are a performance issue, rather than a correctness issue, and the leak itself is small and slow.
The pause bug (bug #44661) is a problem I would like to see fixed, but I'm also willing to let it go. We have an enormous backlog of features that really needs to get out and get tested. That latter fact is weighing on my mind more and more.
|
Wed 01 Apr 2015 05:57:20 PM UTC, comment #17:
Rik: Do your think that the pause() problem
or the memory leaks that you reported need attention befor 4.0.0?
|
Wed 01 Apr 2015 05:53:25 PM UTC, comment #16:
@jwe: Avinoam also reports in comment #2 that he has success with Win 7, but failures with Windows XP. We could just declare Windows XP an unsupported OS, although Matlab is still supporting it (bug fixes only, not for new development).
I think Windows XP is really "the canary in the coal mine" for whatever the root problem is with Qt graphics and Octave, but I would be happy to move on. These segfaults are the last thing I see preventing us from releasing 4.0.0.
|
Wed 01 Apr 2015 04:36:29 PM UTC, comment #15:
I'm able to run test graphics.cc-tst numerous times on a Windows 7 system running in a virtual box VM. I ran it in a while loop, I interrupted it after about 25 iterations. I'm using a build of 4.0.0-rc2.
|
Tue 31 Mar 2015 12:06:18 AM UTC, comment #14:
Also, check bug #44459 which has segfaults with graphics while run on Windows XP. This may be something with that old operating system, but it sure feels like there is still a race condition or other bug in the graphics code.
|
Mon 30 Mar 2015 09:33:47 PM UTC, comment #13:
I can still get the originally reported problem to segfault which is running 'test graphics.cc-tst' on Windows XP. It fails nearly every time, but certainly running it 4 times in a row is enough. When I run under gdb the segfault disappears.
|
Mon 30 Mar 2015 02:52:21 PM UTC, comment #12:
Still no confirmation that this was a QtHandles-only problem? I would like to fix the root problem if at all possible. Otherwise, we might run into it again.
|
Mon 30 Mar 2015 11:56:14 AM UTC, comment #11:
Is there any progress on the hydra build (at least it has not failed for a few days), do the examples in comment #7 still segfault consistently? What about the segfaults on XP?
The patch seems to have "solved" (avoided, hidden ...) bug #44621: reseting the toolkit to the same value was useless (that's why I pushed the patch) but should not have been armful. If the crashes here also have disappeared it is at least a clue if not a satisfactory ending.
|
Fri 27 Mar 2015 08:01:22 AM UTC, comment #10:
Sorry the patch I sent was wrong (forgot to "hg qref") but I pushed the fixed version anyway: it doesn't hurt even if it doesn't fix the bug here.
http://hg.savannah.gnu.org/hgweb/octave/rev/3f461c2b990e
|
Fri 27 Mar 2015 12:23:45 AM UTC, comment #9:
For what it's worth, the examples which crash on Hydra do not crash on Windows XP. The failing example there was 'test graphics.cc-tst' and in that file each of the figures is invisible so that part looks common to both failures.
|
Fri 27 Mar 2015 12:14:21 AM UTC, comment #8:
Hi,
@Mike: are you able to test patches on the hydra build? If so can you try the attached changeset that avoids an update of the toolkit (finalize/initialize new window) when resetting it to the same value ("qt" here).
(file #33465)
|
Thu 26 Mar 2015 02:01:54 AM UTC, comment #7:
We do have debug symbols available for all Octave code, I can poke at whatever you think might be useful.
Anyway, I worked out a couple of minimal examples that demonstrate the segfault every time on the hydra built Octave:
or this:
The following does not segfault:
IOW, the reset has to be done on a figure created with "visible" = "off", and it has to be on the same line as a close or after a close has been performed at least once on any figure.
And of course none of these segfault for me on Debian.
|
Wed 25 Mar 2015 05:39:58 PM UTC, comment #6:
Hm, tantalising! Maybe we should modify the Hydra build to compile debug symbols in.
At any rate, I think this is pretty damning evidence towards QtHandles and away from being a Windows-specific problem. I'm retitling the bug accordingly.
|
Tue 24 Mar 2015 11:28:34 PM UTC, comment #5:
I managed to get a backtrace from the segfault on the hydra autobuilder:
Certainly looks very similar. I can reproduce it every time with the hydra built Octave.
|
Tue 24 Mar 2015 08:25:19 PM UTC, comment #4:
We're seeing a segfault at the same location in the Hydra build (warning, unstable url):
http://hydra.nixos.org/build/20635689
The problem may well be unrelated to Windows. The stack trace points to QtHandles.
|
Fri 20 Mar 2015 06:53:59 PM UTC, comment #3:
I have a backtrace now from a recent development version (19969:3f3418ccd6d4)
|
Fri 20 Feb 2015 11:07:57 PM UTC, comment #2:
I have seen this problem too on a Windows XP non virtual machine,
but not on Win7 machine.
|
Fri 20 Feb 2015 10:34:06 PM UTC, comment #1:
Just for ease, I'm uploading the file graphics.cc-tst. Otherwise, you have to hunt for it in the install directory under share/octave/...
(file #33142)
|
Fri 20 Feb 2015 09:43:50 PM UTC, original submission:
This is on a Windows XP virtual machine with a very new build of MXE-Octave. It happens in the CLI or in the GUI. Sometimes I have to run the command twice, but it usually does it on the first time.
|