Instead of rendering a reference HTML page that wraps an <img> tag
pointing to a PNG, Screenshot tests now load the expected PNG directly
from disk and compare it against the rendered screenshot. This
eliminates the indirection of loading and rendering a second page just
to display a static image.
This also means --rebaseline now works for Screenshot tests, generating
the expected PNG automatically instead of requiring manual screenshot
capture and placement.
Changes:
- Add TestMode::Screenshot with its own collector and runner
- Move PNGs from Screenshot/images/ to Screenshot/expected/ with
normalized names matching input filenames
- Remove all 92 reference HTML wrapper files and the images/
directory
- Remove <link rel="match"> from all 94 Screenshot input HTML
files
- Update add_libweb_test.py Screenshot boilerplate accordingly
- Add Screenshot mode to results viewer image comparison tabs
Instead of rendering text glyphs into a separate mask surface and using
clipShader, paint the backgrounds first and then composite the text
glyphs via saveLayer with SkBlendMode::kDstIn. Skia's saveLayer
automatically sizes its backing at device resolution including CSS
transforms, so no manual scale computation is needed.
Fixes pixelation when zooming in on clipped backgrounds on e.g. the
title of https://modern-css.com/.
Skia deprecated some non-span versions of their API, but they provided
SK_SUPPORT_UNSPANNED_APIS to expose the legacy versions.
SkFontMgr_New_FontConfig now requires a font scanner to be passed in.
There were a few screenshot tests that visibily looked the same but skia
must've changed some rendering infrastructure as the PNGs were not
matching anymore so I rebaselined those and adjusted the fuzzy matching
config to allow them to pass on both macOS and Linux.
The empty-radial-gradient-crash Ref test started to fail as we were
setting the horizontal scale factor to inf in when the height = 0. It
looks like something changed to make doing that not valid anymore.
The overlay port is removed as the issues, mainly skcms symbol import
and export were resolved upstream in skia and utilized in the new port
version.
Skia deprecated some non-span versions of their API, but they provided
SK_SUPPORT_UNSPANNED_APIS to expose the legacy versions.
SkFontMgr_New_FontConfig now requires a font scanner to be passed in.
There were a few screenshot tests that visibily looked the same but skia
must've changed some rendering infrastructure as the PNGs were not
matching anymore so I rebaselined those and adjusted the fuzzy matching
config to allow them to pass on both macOS and Linux.
The empty-radial-gradient-crash Ref test started to fail as we were
setting the horizontal scale factor to inf in when the height = 0. It
looks like something changed to make doing that not valid anymore.
The overlay port is removed as the issues, mainly skcms symbol import
and export were resolved upstream in skia and utilized in the new port
version.
We implemented repeating linear gradients by expanding a vector of color
stops until the entire range was covered. This is both a bit wasteful
and caused Skia to draw corrupted gradients to screen whenever the total
amount of color stops and positions exceeded 127.
Instead of doing that, use the original color stops for the shader and
repeat it instead of clamping it. We need to do a bit of math to project
positions correctly, but after that the shader repeats itself nicely.
While we're here, calculate the gradient's length and the center point
as floats instead of ints, yielding a slight but noticeable improvement
in gradient rendering (see the diff on the zig zag pattern in
css-gradients.html for an example of this).
Call paint_border() recursively, once for the outer line, and once for
the inner one. This is done in a lambda so that we can reuse it for a
couple of other line styles.
Border-radius behaviour doesn't match other browsers, and goes a bit
haywire in some cases. I've left some FIXMEs for someone who
understands the maths here better than I do. 😅
The LineStyle handling is moved to the start of the function, to avoid
unnecessary work.
With the newly supported fuzzy matching in our test-web runner, we can
now define the expected maximum color channel and pixel count errors per
failing test and set a baseline they should not exceed.
The figures I added to these tests all come from my macOS M4 machine.
Most discrepancies seem to come from color calculations being slightly
off.
