RT crashes when trying to open processing queue

heckflosse · February 8, 2018, 11:42am

@HIRAM can you try this patch please?

diff --git a/rtgui/multilangmgr.cc b/rtgui/multilangmgr.cc
index b4896d69..23bf4c45 100644
--- a/rtgui/multilangmgr.cc
+++ b/rtgui/multilangmgr.cc
@@ -117,9 +117,13 @@ void setGtkLanguage(const Glib::ustring &language)
     auto l = localeToLang.getLocale(language);
 #ifdef WIN32
     putenv(("LANG=" + l).c_str());
+#else
+#ifdef __APPLE__
+    setenv("LANG", (l + ".UTF-8").c_str(), true);
 #else
     setenv("LANG", l.c_str(), true);
 #endif
+#endif
 }
 
 }

HIRAM · February 8, 2018, 5:54pm

@heckflosse, still have the problems and crash when switching to queue.

HIRAM · February 8, 2018, 6:21pm

I even tried
#elifdef __APPLE__
{;} //do nothing
but always the same crash…

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   libcairomm-1.0.1.dylib        	0x000000010a47efe0 Cairo::Context::set_source(Cairo::RefPtr<Cairo::Surface> const&, double, double) + 16 (surface.h:364)
1   rawtherapee-bin               	0x00000001087262ab ThumbBrowserEntryBase::draw(Cairo::RefPtr<Cairo::Context>) + 699
2   rawtherapee-bin               	0x0000000108723081 ThumbBrowserBase::Internal::on_draw(Cairo::RefPtr<Cairo::Context> const&) + 369
3   libgtkmm-3.0.1.dylib          	0x0000000109726f2f Gtk::Widget_Class::draw_callback(_GtkWidget*, _cairo*) + 175
4   libgtk-3.0.dylib              	0x0000000109e1971c gtk_widget_draw_internal + 364 (gtkwidget.c:7027)
5   libgtk-3.0.dylib              	0x0000000109c35e9a gtk_container_propagate_draw + 362 (gtkcontainer.c:3840)
6   libgtk-3.0.dylib              	0x0000000109c36443 gtk_container_draw + 147 (gtkcontainer.c:3655)

hening · February 8, 2018, 11:43pm

I’m not a developer, so my idea is probably naïve, but let go: Is it not possible just to take the corresponding parts of the code from RT 4.2 which had overcome this problem, unlike earlier versions?

heckflosse · February 8, 2018, 11:56pm

That sounds good, but

rt 4.2 used gtk2 while rt 5.3 uses gtk3
we didn’t yet indentify the culprit

hening · February 9, 2018, 12:13am

I have no idea what gtk means, but I have a version of RT 4.2 that has gtk3 in the name. That does not solve problem 2. of course. But should it not have something to do with non-ASCII characters in the Path, since also 5.3 works if the image file is on the Desktop?

houz · February 9, 2018, 10:12am

HIRAM:

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   libcairomm-1.0.1.dylib        	0x000000010a47efe0 Cairo::Context::set_source(Cairo::RefPtr&lt;Cairo::Surface&gt; const&, double, double) + 16 (surface.h:364)
1   rawtherapee-bin               	0x00000001087262ab ThumbBrowserEntryBase::draw(Cairo::RefPtr&lt;Cairo::Context&gt;) + 699
2   rawtherapee-bin               	0x0000000108723081 ThumbBrowserBase::Internal::on_draw(Cairo::RefPtr&lt;Cairo::Context&gt; const&) + 369
3   libgtkmm-3.0.1.dylib          	0x0000000109726f2f Gtk::Widget_Class::draw_callback(_GtkWidget*, _cairo*) + 175
4   libgtk-3.0.dylib              	0x0000000109e1971c gtk_widget_draw_internal + 364 (gtkwidget.c:7027)
5   libgtk-3.0.dylib              	0x0000000109c35e9a gtk_container_propagate_draw + 362 (gtkcontainer.c:3840)
6   libgtk-3.0.dylib              	0x0000000109c36443 gtk_container_draw + 147 (gtkcontainer.c:3655)

That seems to correspond to this code:

github.com

Beep6581/RawTherapee/blob/dev/rtgui/thumbbrowserentrybase.cc#L537


      
                  || (parent->getLocation() == ThumbBrowserBase::THLOC_EDITOR && options.filmStripShowFileNames && !options.filmStripOverlayedFileNames)) {
              // dimensions of the info text
              getTextSizes (infow, infoh);
              infoh += textGap;
              //preh -= infoh;
              height += infoh;
          }
          
          // Minimum size for thumbs
          if (preh < 24) {
              preh = 24;
              height = preh + (upperMargin + 2 * borderWidth + lowerMargin) + bsh + infoh;
          }
          
          calcThumbnailSize ();  // recalculates prew
          
          width = prew + 2 * sideMargin + 2 * borderWidth;
          
          if (    (parent->getLocation() != ThumbBrowserBase::THLOC_EDITOR && options.showFileNames && !options.overlayedFileNames)
                  || (parent->getLocation() == ThumbBrowserBase::THLOC_EDITOR && options.filmStripShowFileNames && !options.filmStripOverlayedFileNames)) {
              width = prew + 2 * sideMargin + 2 * borderWidth;

Now, when backBuffer is NULL or garbage that will crash. Can that happen? After all, that case is checked a few lines earlier? Yes, it can, as updateBackBuffer returns early when there is no parent:

github.com

Beep6581/RawTherapee/blob/dev/rtgui/thumbbrowserentrybase.cc#L93-L95


      
          
          state = State::OTHER;
          break;

If any of that is relevant I don’t know. But it might be a start for some further debugging.

Edit: I might have been wrong there. It could be that backbuffer is fine but that it has no surface set.

floessie · February 9, 2018, 10:36am

Thanks, @houz, for looking into it. Our current problem is, why locales are so difficult (under macOS). I would be really glad, if you could share some thoughts on the issue.

Thanks,
Flössie

HIRAM · February 9, 2018, 6:59pm

houz:

Now, when backBuffer is NULL or garbage that will crash. Can that happen? After all, that case is checked a few lines earlier? Yes, it can, as updateBackBuffer returns early when there is no parent:
  github.com

Beep6581/RawTherapee/blob/dev/rtgui/thumbbrowserentrybase.cc#L93-L95
if (!parent) {
return;
}

If any of that is relevant I don’t know. But it might be a start for some further debugging.

Edit: I might have been wrong there. It could be that backbuffer is fine but that it has no surface set

Interesting @houz
Perhaps having a weird combo like en_DE causes problems with backBuffer or surface?