[Octave-bug-tracker] [bug #65866] make check: Fatal segmentation fault when running image/getframe.m test

[Markus Mützel]

Update of bug #65866 (group octave):

                  Status:                    None => Need Info              

    _______________________________________________________

Follow-up Comment #1:

Thank you for the report.

There is a known compatibility issue with Qt6 6.7.x in Octave 9.1.0. But that
should have been addressed by the changes for Octave 9.2.0.

ISTR that we have seen on other platforms (Windows iirc) that some headless
runners have issues with that particular test.
ISTR, that running the test suite with `make check
RUN_OCTAVE_OPTIONS="--no-gui-libs"` worked around that issue. 
Using that flag means that `octave-cli` (instead of `octave-gui`) will be used
for the test suite. That executable doesn't link to libgui (that is used for
the OpenGL graphics toolkits among other things). That also means that the
tests for getframe.m (and some other files) will be skipped.

I'm seeing the following prior output in the test log which could be related:

[ 1056s]   libinterp/corefcn/gh-manager.cc-tst
............................MESA: error: ZINK: vkCreateInstance failed
(VK_ERROR_INCOMPATIBLE_DRIVER)
[ 1056s] libEGL warning: egl: failed to create dri2 screen
[ 1056s] libEGL warning: DRI2: failed to create any config
[ 1056s] MESA: error: ZINK: vkCreateInstance failed
(VK_ERROR_INCOMPATIBLE_DRIVER)
[ 1056s] libEGL warning: egl: failed to create dri2 screen
[ 1056s] libEGL warning: DRI2: failed to create any config
[ 1056s]  pass    1/1    [ 0.087s /  0.086s]


Is xvfb correctly configured? Does Mesa work correctly on the runners that are
used for that test?

If you opt in to the workaround with `--no-gui-libs`, please test manually
that `test getframe` doesn't crash Octave on a local installation (with head).
After all, that could be an issue in the GUI libraries that is not only
triggered on headless systems.

Does it make a difference if you run the test suite with that flag?

If it still crashes, would it be possible to get a backtrace from that crash?
If you can build Octave locally and reproduce the crash there, it should be
possible to attach a debugger to Octave by starting it with `./run-octave -g
--gui` in the build tree.



    _______________________________________________________

Reply to this item at:

  <https://savannah.gnu.org/bugs/?65866>

_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/