(In reply to comment #0)
> Currently Pango is still not listed as threadsafe.  Thus to reliably load svgs
> we need to do the loading in the main thread.

er? did you mean something other than "pango" there? "librsvg" maybe?

also, the RsvgHandle would be created and destroyed in a single thread...

(In reply to comment #1)
> (In reply to comment #0)
> > Currently Pango is still not listed as threadsafe.  Thus to reliably load svgs
> > we need to do the loading in the main thread.
> 
> er? did you mean something other than "pango" there? "librsvg" maybe?
> 
> also, the RsvgHandle would be created and destroyed in a single thread...

librsvg uses Pango, Pango is also being used from the main thread: problem

is this still true and we're just living dangerously?

*** Bug 648638 has been marked as a duplicate of this bug. ***

(In reply to comment #3)
> is this still true and we're just living dangerously?

Good question.  Behdad?

Living dangerously.  I fixed a bug a couple years ago that made this a bit better.  Still, just being lucky.

It's actually not *that* hard to fix pango.  For the most part, we need:

  - Use g_once_init/leave around all static data,

  - Use g_object_lock/unlock (bug 608695) whenever accessing/modifying shared objects (fontmaps mostly, as fonts, families, etc are for the most part immutable after construction).  I can look into it if someone makes g_object_lock() happen.

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

https://bugzilla.redhat.com/show_bug.cgi?id=698589 shows a crash when the main thread and a SVG loading thread are both trying to use FreeType and we get a double free or memory corruption. Interestingly (the reason why I mention it) the crash really looks more like a cairo bug; it's a little hard to see how Pango non-thread-safety could trigger a crash given the backtraces.

(But if Pango is corrupting memory elsewhere, then of course, all bets are off.)

Wow, that trace is gold!

I easily tracked the bug down.  Pushing to cairo master.

diff --git a/src/cairo-ft-font.c b/src/cairo-ft-font.c
index acb7121..4313148 100644
--- a/src/cairo-ft-font.c
+++ b/src/cairo-ft-font.c
@@ -2142,7 +2142,7 @@ _cairo_ft_scaled_glyph_init (void                 *abstract_font,
        vertical_layout = TRUE;
     }
 
-    error = FT_Load_Glyph (scaled_font->unscaled->face,
+    error = FT_Load_Glyph (face,
                           _cairo_scaled_glyph_index(scaled_glyph),
                           load_flags);
     /* XXX ignoring all other errors for now.  They are not fatal, typically

Humm.  Perhaps not.  My bad.  Still investigating.

Nope.  I'm giving up.  The trace doesn't make sense.