Switch from using GdkPixbuf to cairo for painting on Drawables.

Switch from using GdkPixbuf to cairo for painting on Drawables.

Make everything use ARGB order in registers (B.G.R.A order in memory on
little-endian systems)

[Evan Martin, 2008-11-05 23:03:21]

Code looks great, so my comments are mostly on the comments.

http://codereview.chromium.org/8227/diff/1/3
File base/gfx/bitmap_platform_device_linux.h (right):

http://codereview.chromium.org/8227/diff/1/3#newcode11
Line 11: #include <cairo/cairo.h>
I wonder if you could get away with forward-declaring cairo_surface_t, so that
we don't have this header leaking into all code that uses this.  I wonder if it
matters, though.

http://codereview.chromium.org/8227/diff/1/3#newcode22
Line 22: // be well documented anywhere) default to R.G.B.A order. They'll do
the needed
Though it is true the docs are lacking, the parenthetical comes across as
needlessly snarky and not something I think we should leave in our code.

http://codereview.chromium.org/8227/diff/1/3#newcode39
Line 39: // Thus, with all that, everything should match!
It's not especially clear how this follows from the above -- you perhaps could
mention here or in the GdkPixbuf paragraph that we don't use GdkPixbufs since
they don't match the other drawing orders...

http://codereview.chromium.org/8227/diff/1/3#newcode65
Line 65: /// the Windows and Mac versions of this class do.
I guess this comment should comment on whether it takes ownership of |surface|? 
And why is it with three slashes, anyway?  (I know you didn't do that here.)

http://codereview.chromium.org/8227/diff/1/4
File skia/include/corecg/SkUserConfig.h (right):

http://codereview.chromium.org/8227/diff/1/4#newcode122
Line 122: // ARGB (in registers)
Should we make a note about endianness here?
Isn't it more important to match cairo (for GTK) and the image decoders than X?

http://codereview.chromium.org/8227/diff/1/5
File webkit/tools/test_shell/gtk/webwidget_host.cc (right):

http://codereview.chromium.org/8227/diff/1/5#newcode8
Line 8: #include <cairo/cairo.h>
alphabetize

[agl, 2008-11-05 23:19:52]

http://codereview.chromium.org/8227/diff/1/3
File base/gfx/bitmap_platform_device_linux.h (right):

http://codereview.chromium.org/8227/diff/1/3#newcode11
Line 11: #include <cairo/cairo.h>
On 2008/11/05 23:03:21, Evan Martin wrote:
> I wonder if you could get away with forward-declaring cairo_surface_t, so that
> we don't have this header leaking into all code that uses this.  I wonder if
it
> matters, though.

I tried, you can. Have fixed.

http://codereview.chromium.org/8227/diff/1/3#newcode22
Line 22: // be well documented anywhere) default to R.G.B.A order. They'll do
the needed
On 2008/11/05 23:03:21, Evan Martin wrote:
> Though it is true the docs are lacking, the parenthetical comes across as
> needlessly snarky and not something I think we should leave in our code.

removed.

http://codereview.chromium.org/8227/diff/1/3#newcode39
Line 39: // Thus, with all that, everything should match!
On 2008/11/05 23:03:21, Evan Martin wrote:
> It's not especially clear how this follows from the above -- you perhaps could
> mention here or in the GdkPixbuf paragraph that we don't use GdkPixbufs since
> they don't match the other drawing orders...

Have added more.

http://codereview.chromium.org/8227/diff/1/3#newcode65
Line 65: /// the Windows and Mac versions of this class do.
On 2008/11/05 23:03:21, Evan Martin wrote:
> I guess this comment should comment on whether it takes ownership of
|surface|? 
> And why is it with three slashes, anyway?  (I know you didn't do that here.)

Three slashes fixed. Comment to the effect that we take ownership added.

http://codereview.chromium.org/8227/diff/1/4
File skia/include/corecg/SkUserConfig.h (right):

http://codereview.chromium.org/8227/diff/1/4#newcode122
Line 122: // ARGB (in registers)
On 2008/11/05 23:03:21, Evan Martin wrote:
> Should we make a note about endianness here?
> Isn't it more important to match cairo (for GTK) and the image decoders than
X?

This is the register order which I believe will be consistent across big and
little endian machines. However, I'll add a comment and a preprocessor error.

http://codereview.chromium.org/8227/diff/1/5
File webkit/tools/test_shell/gtk/webwidget_host.cc (right):

http://codereview.chromium.org/8227/diff/1/5#newcode8
Line 8: #include <cairo/cairo.h>
On 2008/11/05 23:03:21, Evan Martin wrote:
> alphabetize

ack.

[Evan Martin, 2008-11-05 23:52:23]

LGTM

warning: the #error bit might not work on other compilers (I don't know either
way, to be honest).  might be worth a trybot run.

http://codereview.chromium.org/8227/diff/402/206
File base/gfx/bitmap_platform_device_linux.h (right):

http://codereview.chromium.org/8227/diff/402/206#newcode11
Line 11: struct _cairo_surface;
this isn't necessary when you have the following line