Previously the gradient walker was doing excessive resets, (such
as on every pixel in constant-colored regions or outside the
gradient with CAIRO_EXTEND_NONE). Don't do that.
The previous implementation fell apart quite badly when neither radius
value was equal to 0.0. I derived the math from scratch, (much thanks to
Vincent Torri <vtorri@univ-evry.fr> for guiding me to a simpler derivation
than I was doing originally), and it's working much better now without
being any slower, (in fact, cairo-perf shows speedup of 1.05x to 1.58x on
my laptop here).
This work also provides groundwork for defining the behavior of radial
gradients where neither circle is wholly contained within the other, (though
we haven't done that definition yet---it will require a new test case and a
very little bit of work on the implementation).
This is a fix for the following bug report:
Radial Gradients with nonzero inner radius misplace stops
https://bugs.freedesktop.org/show_bug.cgi?id=7685
Embarrassingly enough, the test suite previously never called
into cairo_pattern_create_radial at all. Unsurprisingly, this
has led to bugs creeping into the radial gradient implementation.
This reverts the following commits:
2715f2098167e3b3c53b
See this thread for an analysis of the problems it caused:
http://lists.freedesktop.org/archives/cairo/2007-February/009825.html
In short, a single cache for all backends doesn't work, as one thread
using any backend can cause an unused xlib pattern to be evicted from
the cache, and trigger an xlib call while the display is being used
from another thread. Xlib is not prepared for this.
This is needed since cairo-test does not issue a show_page now, and
the one coming from paginated layer is too late (around finish).
The way we force the show_page is kinda funny. That is, we create
a temporary cairo_t and call cairo_show_page() on it. But works.
That is, we always emit at least one page of output. Otherwise our
PS/PDF/SVG files would be corrupt, and output with zero pages does
not make much sense.
This means, PS/PDF/SVG do not need the final cairo_show_page() anymore.
If there is any drawing in the page, a cairo_show_page() is done
automatically.
with surface patterns. One test scaled the pattern up, another scales
down. We observe that both PS and PDF are broken when scaling down.
This is the reason that PDF is failing in the fallback-resolution test
too.
Previously we were generating an image object with the desired
transform and then a pattern with only a translation. This means,
the pattern was alwasys axis-aligned. Obviously this does not
work correctly with repeated rotated patterns.
We now use an identity matrix for the image and put all the
pattern transformation (well, it's inverse) into the pattern.
This fixes the issue nicely.
The surface-pattern test is still failing even with a reference
image, because the gs rendering of the normal case and the
device-offset=25 case are different and both have seams.
The surface-pattern test was very naive, painting a surface pattern
repeated at identity size. With the new test, the surface pattern
is scaled and rotated. This reveals a serious bug in the PS backend.
With the hack in _cairo_pattern_acquire_surface to return a 2x2
surface for reflected patterns, we can now accept REFLECT surface
patterns natively in all backends. SVG was already doing that.
The PDF case needed some changes to go through
_cairo_pattern_acquire_surface. A similar change to the recent
change in SVG.
This change is required for the upcoming hack to support
CAIRO_EXTEND_REFLECT. In short, the idea is that the surface returned
by _cairo_pattern_acquire_surface may look different from the one
returned by _cairo_surface_acquire_source_image, to support some pattern
modes not supported by the backends directly.
This was needed for SVG backend because it does not implement clone_similar.
However, I'm worried about possible infinite recursion here. Not sure what
to do.
We do this through a hack, that is, we make
_cairo_pattern_acquire_surface to return a surface that has four
copies of the original surface painted such that this image can
be simply repeated to get the effect of reflecting the original
surface.
This fixes the formerly XFAIL test extend-reflect.
The PS backend was accepting all surface patterns, just to ASSERT_NOT_REACHED
later if extend type of the pattern was PAD or REFLECT. That didn't make
sense and was causing crash for PAD (surprisingly not for REFLECT. Not sure
why). So now it rejects those types of surface patterns, just like the PDF
surface does.
