There is a bug in the recording surface where if one recording surface
is used as the same source on different paginated targets that support
fine grained fallbacks, the output may be incorrect because the
analysis results of a previous target are re-used on another target.
The base image has 16 bits per channel, with colors that do not
have an exact corresponding 8 bit representation. The reference
image uses 8bpc, like the rgb24/argb32 formatted images which
the test outputs.
Running test/check-refs.sh reports:
redundant: arc-direction.pdf.ref.png and arc-direction.ref.png are byte-by-byte identical files
redundant: big-little-triangle.traps.argb32.ref.png and big-little-triangle.argb32.ref.png are byte-by-byte identical files
redundant: big-little-triangle.traps.rgb24.ref.png and big-little-triangle.rgb24.ref.png are byte-by-byte identical files
redundant: clip-fill-rule.pdf.rgb24.ref.png and clip-fill-rule.rgb24.ref.png are byte-by-byte identical files
redundant: dash-offset-negative.pdf.ref.png and dash-offset-negative.ref.png are byte-by-byte identical files
redundant: font-matrix-translation.traps.ref.png and font-matrix-translation.ref.png are byte-by-byte identical files
redundant: ft-show-glyphs-positioning.traps.ref.png and ft-show-glyphs-positioning.ref.png are byte-by-byte identical files
redundant: ft-show-glyphs-table.traps.ref.png and ft-show-glyphs-table.ref.png are byte-by-byte identical files
redundant: glyph-cache-pressure.traps.ref.png and glyph-cache-pressure.ref.png are byte-by-byte identical files
redundant: inverse-text.traps.ref.png and inverse-text.ref.png are byte-by-byte identical files
redundant: line-width-large-overlap-offset.ps.ref.png and line-width-large-overlap-offset.ref.png are byte-by-byte identical files
redundant: partial-clip-text-right.traps.ref.png and partial-clip-text-right.ref.png are byte-by-byte identical files
redundant: partial-clip-text-top.traps.ref.png and partial-clip-text-top.ref.png are byte-by-byte identical files
redundant: record90-fill-alpha.pdf.ref.png and record90-fill-alpha.ref.png are byte-by-byte identical files
redundant: record90-paint-alpha-clip.quartz.ref.png and record90-paint-alpha-clip.ref.png are byte-by-byte identical files
redundant: record-fill-alpha.pdf.ref.png and record-fill-alpha.ref.png are byte-by-byte identical files
redundant: recordflip-whole-fill-alpha.quartz.ref.png and recordflip-whole-fill-alpha.ref.png are byte-by-byte identical files
redundant: recordflip-whole-paint-alpha-clip-mask.quartz.ref.png and recordflip-whole-paint-alpha-clip-mask.ref.png are byte-by-byte identical files
redundant: record-mesh.ps.ref.png and record-mesh.ref.png are byte-by-byte identical files
redundant: select-font-face.traps.ref.png and select-font-face.ref.png are byte-by-byte identical files
redundant: show-glyphs-advance.traps.ref.png and show-glyphs-advance.ref.png are byte-by-byte identical files
redundant: show-text-current-point.traps.ref.png and show-text-current-point.ref.png are byte-by-byte identical files
redundant: text-antialias-gray.traps.ref.png and text-antialias-gray.ref.png are byte-by-byte identical files
This commit removes these redundant files.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This commit changes the height of the coverage-column-triangles test by
using a smaller height for the test. Previously the image was 40 pixels
in height and the test took 517 seconds for me. Now, the image has a
height of 4 pixels and the test takes 55 seconds.
This is possible since column_triangles() does the same thing for each y
coordinate. Thus, previously this test just did the same thing 40 times.
Times are measured with the backends I happened to have enabled locally:
script, xcb, xlib, image, recording.
Thanks a lot to Christian Rohlfs for coming up with this idea in
https://gitlab.freedesktop.org/cairo/cairo/-/issues/498#note_1700197
Possibly-fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/498
Signed-off-by: Uli Schlachter <psychon@znc.in>
CoreText uses different advances depending on the font size, with very
small point sizes sometimes getting advances that are smaller than the
glyph's width. This is manifested in the Apple Color Emoji font with the
Emoji glyphs having a width of 1.25 and an advance width of 1.0. That
results in overlapping emoji when they're in a string.
The small spacing difference also affects 3 tests so updated reference
images are included in this commit.
# Please enter the commit message for your changes. Lines starting
This commit copies the current alpha-similar.image.rgb24.out.png to
alpha-similar.rgb24.ref.png to make the test pass.
This failure goes back to commits 6647511 / 049060 which changed the
image comparison in the test suite. Previously, transparent "nothing"
and black compared equal, but said commits fixed that. Thus, having a
special reference image with black instead of transparency is the right
fix.
Signed-off-by: Uli Schlachter <psychon@znc.in>
CGContextDrawTiledImage turned up as a significant time-user while
profiling a benchmark created to evaluate
https://gitlab.gnome.org/GNOME/gtk/-/issues/3714. Without this commit
the benchmark is able to perform a mean frame rate of 2.19 frames per
second with a standard deviation of 0.09; with the commit the mean frame
rate is 2.37 fps, s.d. 0.30, both over 15 10-second samples. Student's
t-test reports a 9.8% likelyhood that the two represent the same
distribution.
cairo-script has a special case for drawing a set of rectangles. The
rectangle detection can trip over a "degenerate rectangle": Just a
horizontal line. It detects the line as the beginning of a rectangle and
then claims CAIRO_STATUS_INVALID_PATH_DATA when the other sides of the
rectangle are missing.
This commit simply changes the return value to
CAIRO_INT_STATUS_UNSUPPORTED to trigger a fallback to the generate case.
Test case is a straight-forward adaption from the original bug report.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/535
This commit copies clip-unbounded.base.rgb24.ref.png to
clip-unbounded.rgb24.ref.png. This makes the test pass for rgb24
targets.
This failure goes back to commits 6647511 / 049060 which changed the
image comparison in the test suite. Previously, transparent "nothing"
and black compared equal, but said commits fixed that. Thus, having a
special reference image with black instead of transparency is the right
fix.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The proper pthread check activated some tests that weren't active
before, which resulted in a test failure apparently caused by
inexact test rendering.
Update the quartz reference image accordingly.
