GTK+ assert when opening a PO file

[png2378]

If I run Poedit and then open some po-file, everything is fine. But if I try to open po-file directly (for example from desktop), Poedit crashes.

Terminal log:

[maxim@ProBook]$ poedit gnome-software.master.ru.po
Fontconfig error: "local.conf", line 2: no element found

(poedit:23250): Gtk-CRITICAL **: gtk_box_gadget_distribute: assertion 'size >= 0' failed in GtkScrollbar

(poedit:23250): Gtk-CRITICAL **: gtk_box_gadget_distribute: assertion 'size >= 0' failed in GtkScrollbar

(poedit:23250): Gtk-WARNING **: Negative content width -33 (allocation 1, extents 17x17) while allocating gadget (node button, owner GtkToggleButton)

(poedit:23250): Gtk-WARNING **: Negative content height -9 (allocation 1, extents 5x5) while allocating gadget (node button, owner GtkToggleButton)

(poedit:23250): Gtk-CRITICAL **: gtk_box_gadget_distribute: assertion 'size >= 0' failed in GtkScrollbar
poedit: cairo-arc.c:189: _cairo_arc_in_direction: Assertion `angle_max >= angle_min' failed.
Aborted (стек памяти сброшен на диск)

OS: Fedora 26
DE: Gnome 3.24
Version: Poedit 2.0.2

[vslavik]

From a quick glance, looks like you got an assert in a library Poedit uses (GTK+ or wxGTK, versions of neither of which you provide), not in Poedit itself. You probably should file a bug with Fedora.

If you think otherwise, please follow this advice and also provide backtrace from the debugger (with debug symbols compiled in!).

[png2378]

So, I left feedback in Fedora's update gating system. I hope that my info will help (I don't know how I can provide backtrace).

[c72578]

Attached you will find a backtrace
Some version info:
Poedit: commit d2084cf
wxGTK3.x86_64 3.0.3-1.fc26
gtk+.x86_64 1:1.2.10-83.fc26
gtk3.x86_64 3.22.15-1.fc26

Remark: The problem does not appear in previous Poedit 2.0.1 release.

gdb_poedit_backtrace.zip

[c72578]

In the meantime I could do some further tests.

• The problem appeared with this commit: f98feb2
  git checkout f98feb2
  commit f98feb2 (HEAD)
  Author: pepc pcondal@apsic.com
  Date: Sat Mar 25 17:27:53 2017 +0100

  Add --line command-line switch

Problem appeared

• Previous commit: 0f9dac7
  git checkout 0f9dac7
  commit 0f9dac7 (HEAD)
  Author: Václav Slavík vaclav@slavik.io
  Date: Wed May 10 18:36:32 2017 +0200
  Highlight 2+ spaces in the middle of string too

OK

[vslavik]

Thanks! That's a useful clue that I wouldn't have suspected.

[vslavik]

This seems to be largely coincidental with f98feb2, not directly caused by it. Not reproducible with wx trunk either, only 3.0.

[Martchus]

Maybe it is coincidental, but the issue also occurs under Arch Linux and the patch/revert provided by @c72578 fixes it here, too.

[vslavik]

@Martchus You will observe that I don't dispute that; in fact I explicitly confirm it (and thus your "me too" comment adds little value: yes, it is already known that wx-3.0 causes it). That a commit changes circumstances in the code in such a way that it triggers a bug in some unrelated code is not unheard of — that doesn't mean that said change causes it.

The far more interesting question that I would appreciate help with is what causes the actual bug and if there's anything that can be done about it in client code, i.e. outside of wxGTK. Reverting f98feb2 is out of the question, but it must be some small bit of it triggering this and that may be possible to work around.

(And sorry, I apparently closed this by accidentally clicking the other comment button. Didn't mean to. If it can be worked around in Poedit code, it should.)

[Martchus]

I've just responded because you've been closing the issue and I wasn't sure how to interpret your last comment. So thanks for making it clear. And I agree that the commit should not be reverted.

[vslavik]

Yeah, I see the ambiguity now :(

[vslavik]

(And sorry, I apparently closed this by accidentally clicking the other comment button. Didn't mean to. If it can be worked around in Poedit code, it should.)

@Martchus Actually, I was confused: as seen in the log above, commit 86e0677 fixed this 4 days before your comment. Closing again; please reopen/comment only if you tested with the latest master, not 2.0.2.