[cairo] fixing Cairo's FreeType multi-threaded usage

[Jan Slupski]

Next issue:

After brief looking at 1.4.2 it seems that David patches weren't merged. 
Are there any plans to do so?

Without this fix, heavy multithread testing still fails after few hours
(at least used to fail at Cairo 1.3.16).

Thanks,
Jan

Link to final version of the patch:
http://lists.freedesktop.org/archives/cairo/2007-March/009924.html


On Wed, 28 Feb 2007, David Turner wrote:

> Hello,
>
> here's the first version of my patches to fix Cairo's usage of FreeType with multiple
> threads. It's rather big, so I'll summarize things here a little:
>
> - it's more or less a complete rewrite of the functionality, but it is
>  API and ABI compatible with the original code.
>
> - there is no need to modify any client code, except the one using the broken
>  cairo_ft_scaled_font_lock_face() and cairo_ft_scaled_font_unlock_face(),
>  which probably means only Pango at the moment.
>
>  normal users of Cairo should now be lock-free in all FreeType-related operations.
>  no need for thread-specific list and other hacks like that by the way.
>
> - even if you use cairo_ft_scaled_font_lock_face() and cairo_ft_scaled_font_unlock_face(),
>  you should be free from deadlocks with the rest of Cairo. that's because the returned
>  FT_Face objects are allocated into a different library instance than the one used by
>  the rest of the library.
>
>  Of course, all hell breaks loose if two threads use FT_Face objects returned by
>  cairo_ft_scaled_font_lock_face() at the same time, even if they're not the same
>  FT_Face.
>
>  The upgrade path to sanity is to use the functions named:
>
>     cairo_freetype_state_lock()
>     cairo_freetype_state_get_face()
>     cairo_freetype_state_get_size()
>     cairo_freetype_state_unlock()
>
>  defined and documented in the new file cairo-freetype-state.c
>
>
> - there are now two backends: a FreeType-specific one, and a FontConfig-specific one.
>  the only difference is that the FontConfig one is capable of implementing the
>  toy interface, and is thus still the default on Unix.
>
>  the idea is to be able to compile the FreeType backend on platforms where FontConfig
>  is not available or desirable. Of course, the toy interface would then be implemented
>  by the default backend for the platform, or simply not at all.
>
> - there are thus two additionnal public header files: "cairo-freetype.h" and
>  "cairo-fontconfig.h", with corresponding functions using the surprisingly long
>  "cairo_freetype_" and "cairo_fontconfig_" prefixes
>
> - this was done to avoid any conflict with the existing code and public headers.
>  "cairo-ft.h" is still there and its functions are fully supported even if we
>  can't make thread-safety guarantees for "locked" FT_Face usage.
>
> - the source code "cairo-ft-font.c" is still there, but is not compiled anymore.
>  this is essentially for comparison purposes.
>
> - there are quite a few new source files as well:
>
>  cairo-freetype-state.c    (thread-safe FT_Face/FT_Size management)
>  cairo-freetype-font.c     (FreeType font backend)
>  cairo-freetype-glyph.c    (glyph loading + rendering + filtering)
>  cairo-freetype.h          (public FreeType backend APIs)
>  cairo-freetype-private.h  (guess what)
>  cairo-fontconfig.h        (public FontConfig backend APIs)
>  cairo-fontconfig.c        (tiny FontConfig font backend)
>  cairo-ft.c                (stubs to implement original cairo_ft_ functions)
>
> - I'd be happy to use shorter prefixes like "cairo_ft_" and "cairo_fc_" once
>  Carl is happy with the implementation provided here. I wanted to be able to
>  easily compare the new and the old code, and the different prefixes just
>  makes things a lot easier there.
>
> - the cairo_ft_ functions do implement the current Cairo behaviour regarding
>  font options (i.e. the FontConfig backend ignores user-provided options for
>  antialiased setting, extracting those from the original FcPattern instead)
>
>  since I still don't understand why this is a good thing, the cairo_freetype_
>  and cairo_fontconfig_ equivalent functions to do implement this "feature",
>  but this can easily be changed.
>
>  (I plan to make a different email to discuss this issue in more details)
>
> - there is also a RATIONALE file that explains how certain things are implemented,
>  that should make understanding the changes easier
>
> Finally, all test cases pass, except the ft-text-vertical-layout-* ones, but
> I believe it's because I don't have the relevant font files installed on my
> machine. Will check later.
>
> There has been a lot of shuffle during the intermediate commits that led to this
> patchset, so I recommend only studying the final changes, instead of the individual
> ones.
>
> Voila, you can find all this here:
>
>  http://david.freetype.org/cairo/fix-freetype-usage-1.patchset.bz2
>  http://david.freetype.org/cairo/RATIONALE
>
>
> Hope this helps,
>
> - David Turner
> - The FreeType Project (www.freetype.org)
> _______________________________________________
> cairo mailing list
> cairo at cairographics.org
> http://cairographics.org/cgi-bin/mailman/listinfo/cairo
>

    _  _  _  _  _____________________________________________
    | |_| |\ |  S L U P S K I             jslupski at juljas.net
  |_| | | | \|                             http://juljas.net/