Created attachment 99610 [details]
The test case, which test multi-threading in rsvg.

Compile instructions are contained in the C file.

Run it using:

rsvg_render_test input.png <output-base-name> <nthreads>

Renders input.png in <nthreads> threads.
The output files are named output-base-name-%d.png

I'm not sure about the thread support in librsvg itself, but at least thread support at the glib level should be initiliazed by a call to gthread_init.

Also, it seems pango is not initiliazed. I don't know how to do it properly, but a call to gkt_init seems to fix the issue.

Please add a call to g_thread_init(NULL) before rsvg_init() and try this program again.

librsvg itself doesn't have any code that I"m aware of that shouldn't be MT safe, provided that you don't try to use the same RsvgHandle in different threads.

Created attachment 99633 [details]
Modified testcase with g_thread_init and prerendering.

This modified testcase uses rsvg_endle_new_from_file, which is a small simplification.

Additionally, it call g_thread_init(NULL) at startup. This fixes the program mostyl, but with a let's say 10% propability my program crashes with 5-7 thread like in the following example:

**********
  ./rsvg-thread-test2 /usr/share/icons/Mist/scalable/places/folder-new.svg tmp 7
Starting thread 0.
Starting thread 1.
Starting thread 2.
Starting thread 3.
Starting thread 4.
Starting thread 5.
Starting thread 6.

(process:25071): GLib-GObject-WARNING **: IA__g_object_new_valist: object class `GdkPixbuf' has no property named `colorspace'
Thread 0: Error saving pixel buffer: Fatal error in PNG image file: Image width or height is zero in IHDR.
Finished thread 2.
Finished thread 1.
Finished thread 3.
Thread 0 failed.
Finished thread 4.
Finished thread 6.
Finished thread 5.
**********

Another scenarios ocurring at a comparable propability is:

**********
 ./rsvg-thread-test2 /usr/share/icons/Mist/scalable/places/folder-new.svg tmp 7
Starting thread 0.
Starting thread 3.
Starting thread 2.
Starting thread 1.
Starting thread 6.
Starting thread 5.
Starting thread 4.
Thread 2: Error fetching pixel buffer from svg handle.

(process:24902): GdkPixbuf-CRITICAL **: _gdk_pixbuf_load_module_unlocked: assertion `image_module->module == NULL' failed

(process:24902): GdkPixbuf-CRITICAL **: gdk_pixbuf_savev: assertion `error == NULL || *error != NULL' failed
Segmentation fault (core dumped)
*******

Both problem disappear, if I render a single image before I enter multithreading, which may be established by the newly introduced '-prerender' cmdline option as in the following example:

./rsvg-thread-test2 -prerender /usr/share/icons/Mist/scalable/places/folder-new.svg tmp 7

The whole testcase indicates, that there is some non-thread-safe initialization code inside gdk-pixbuf. However, I did not find a gdk_pixbuf_init() function or a similar functionality to undertake the gdk-pixbuf initialization beforehand.

Calling gtk_init() is not an option for me,because my target application shall be a rendering server, which should run w/o a GUI.

  Thanks for your help so far,

    Wolfgang

Thanks. This looks like a thread-safety issue inside of GdkPixbuf. Reassigning.

librsvg is using pango for text, which is not threadsafe

Moving to pango then. We're only as threadsafe as our dependencies.

*** This bug has been marked as a duplicate of bug 377539 ***