GitLab

Mask execution closely follows background execution, which results in the
size of the mask being reduced to the exact size requested, rather than
filling out the dest_rect in BackgroundImageGeography::SetNoRepeatX and
::SetNoRepeatY. This CL puts the mask in the correct location, then
when we are creating the shader in image.cc, if the mask doesn't fill
out the dest_rect, decal tilemode is used to fill the rest with
transparency, which allows the mask to fill the full area.

Regarding the change to the fast/backgrounds/mask-composite.html test:
This change results in the original test failing, displaying the same
thing that it displays when using source-out. Based on my understanding
of how the 'copy' Porter-Duff keyword works, I believe this is correct
and that the test previously had an incorrect baseline because of this
bug. My understanding is that, if mask-composite is 'copy', then regardless
of the number of masks, only the first one will be displayed. The PD
definition of copy is that only the source mask is displayed, and since
this would be applied starting from the last two masks listed, moving up
to the first two, the end result is that only the first mask is displayed.
And, since this matches what happens when mask-composite:source-out is
used (and matches Edge and Firefox using 'subtract'), I just changed it
to a likely more commonly used keyword.

Bug: 917067
Change-Id: Ic9174ea67d92abe6b78acf9b7f2640796d30d249
Reviewed-on: https://chromium-review.googlesource.com/c/1447261
Commit-Queue: Mario Bianucci <mabian@microsoft.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>
Reviewed-by: Fredrik Söderquist <fs@opera.com>
Cr-Commit-Position: refs/heads/master@{#630004}