Gtk::PixbufError exception not caught in Windows

[Johan Engelen]

For some reason, the code in the testfile does NOT catch the exception on windows. It seems to work for people on Linux, although not on my machine ...

Run the batch file to compile it. Then:
"pixbuferror.exe right.jpg" gives: "Success!"
"pixbuferror.exe wrong.jpg" *should* give: "caught Gdk::PixbufError" but instead gives something like "terminate was called after uncaught exception". For some reason, the catchers do not work!

[Johan Engelen (johanengelen)]

This seems related:
http://sourceforge.net/mailarchive/message.php?msg_id=E1IUIrP-0002LO-KC%40sc8-sf-web21.sourceforge.net
(the comment by Danny Smith (dannysmith) date: 2007-09-09 13:54)

[Bryce Harrington (bryce)]

Johan, have you tested the workaround suggested upstream, to build the dll and exe with --shared-libgcc?

[Johan Engelen (johanengelen)]

I cannot build the dlls needed for Inkscape on windows. Assigning to Ishmal, who makes the lib binary package on win32.

[Bryce Harrington (bryce)]

Ishmal - same question to you?

[Bryce Harrington (bryce)]

It sounds like this is a *potential* bug, rather than one that can be reliably reproduced in Inkscape in the wild. In talking with Ishmal, we don't really use C++ exceptions that much for this to be a significant issue, and danny is no longer available as a mingw contact. I'm dropping the milestone and reducing priority to Low.

We should look at it again some day, but if we don't have issues in the wild, this can probably be closed as invalid.

[amphi (i-launchpad-kaioa-com)]

This bug is the reason behind this critical crash bug:
https://bugs.launchpad.net/inkscape/+bug/168471

I "reproduced" it several times by accident and it's really annoying. ;)

[Johan Engelen (johanengelen)]

This is not a potential bug at all. It is highly reproducable (i.e. try it 100 times, it will crash 100 times), see bug 168471.
This bugreport is just a minimal testcase, to help solving it.

I regret this bugreport was dropped this easily, as it took quite some time and the fix seems so easy. It's not *just* a bug, it is a hard crash without data recovery.

[Johan Engelen (johanengelen)]

http://www.sourceware.org/ml/pthreads-win32/2000/msg00136.html

[sas (sas-sas)]

This bug also causes crashes with Glib::SpawnError exceptions (although they aren't as common as the Gtk::PixbufError crashes).

[Ulferikson (ulferikson)]

Wasn't it to avoid the (very common) Glib::SpawnError crashes that BYPASS_GLIB_SPAWN was added to io/sys.cpp? Looks like revision 18041 and bug #204779.

Maybe some code is still using Glib::spawn instead of Inkscape::IO::spawn? This works as long libinskcape is static..

[sas (sas-sas)]

I still get Glib::SpawnError crashes in revision 19339. For example, if I open a file, then try to save it as AutoCAD DXF (*.dxf).

[Krzysztof Kosinski (tweenk)]

This bug requires rebuliding or changing the compiler and rebuilding the devlibs (at least those that use C++) from scratch. I recommend using TDM-GCC build as out compiler, it doesn't have the exception-across-DLLs problem at all regardless of whether -shared-libgcc is used.

[Krzysztof Kosinski (tweenk)]

I rebuilt the C++ DLLs using TDM-GCC, now checking whether this fixes the problem.

[Krzysztof Kosinski (tweenk)]

Forgot about poppler and ImageMagick. They're more finicky that the gtkmm stack, but I managed to compile them. Rebuilding again.

[Krzysztof Kosinski (tweenk)]

Success! I have a build that does not crash when importing or opening a corrupted PNG file. Final rebuild test before committing. The MinGW compiler I used will be added to Devlibs.

[Krzysztof Kosinski (tweenk)]

Committed as devlibs r11, Inkscape r9041.

[Johan Engelen (johanengelen)]

This is great great great news, thanks so much !

[jazzynico (jazzynico)]

It seems that the issue is back in 0.48.2 and the trunk (0.48.1 is ok, but it uses a different devlibs revision).

[jazzynico (jazzynico)]

Reopening. 0.48.3 and the trunk are affected.

[jazzynico (jazzynico)]

Fixed again in the devlibs, revision 36.

[jazzynico (jazzynico)]

Reopening due to regressions in devlibs r36 (uncommitted).

[jazzynico (jazzynico)]

It should be ok now, devlibs r36.