With this change we ensure that Skia surfaces are allowed to be created
or destroyed only by one thread at a time. Not enforcing this before
resulted in "Single owner failure." assertion crashes in debug mode on
pages with canvas elements.
Skia has a check in debug mode to verify that surface is only used
within one thread. Before this change we were violating this by
allocating surfaces on the main thread while using and destructing them
on the rendering thread.
The Skia Ganesh backend we currently use doesn't support painting from
multiple threads, which could happen before this change when the main
thread used Skia to paint on the HTML canvas while the rendering thread
was working on display list rasterization.
Fixes https://github.com/LadybirdBrowser/ladybird/issues/4172
The display list is an immutable data structure, so once it's created,
rasterization can be moved to a separate thread. This allows more room
for performing other tasks between processing HTML rendering tasks.
This change makes PaintingSurface, ImmutableBitmap, and GlyphRun atomic
ref-counted, as they are shared between the main and rendering threads
by being included in the display list.
When decoding data into bitmaps, we end up with different alpha types
(premultiplied vs. unpremultiplied color data). Unfortunately, Skia only
seems to handle premultiplied color data well when scaling bitmaps with
an alpha channel. This might be due to Skia historically only supporting
premultiplied color blending, with unpremultiplied support having been
added more recently.
When using Skia to blend bitmaps, we need the color data to be
premultiplied. ImmutableBitmap gains a new method to enforce the alpha
type to be used, which is now used by SharedResourceRequest and
CanvasRenderingContext2D to enforce the right alpha type.
Our LibWeb tests actually had a couple of screenshot tests that exposed
the graphical glitches caused by Skia; see the big smiley faces in the
CSS backgrounds tests for example. The failing tests are now updated to
accommodate the new behavior.
Chromium and Firefox both seem to apply the same behavior; e.g. they
actively decode PNGs (which are unpremultiplied in nature) to a
premultiplied bitmap.
Fixes#3691.
GIF loader was completely failing when encountering errors with
frame descriptors or individual frames, even when some frames were
successfully loaded. Now we attempt to decode at least some frames
and fail only when no frames can be decoded at all.
This adds all parameters supported by skia by default. Supporting the
other parameters would just be a matter of defining the constants, but
that's work for another time.
This introduces a new API in ImageDecoderPlugins that allow an image
decoder to return a CICP struct. Also, we use this API in
ImageDecoder::color_space() to create a color space corresponding to
these CICP.
We have to be careful to always destroy the jpeglib decompression struct
before returning from JPEGLoadingContext::decode. We were doing this in
jpeglib error handlers, but we have a couple of paths that bail from the
decoder via TRY. These paths were neither cleaning up memory nor setting
the image decoder to an error state.
So this patch sets up a scope guard to ensure we free the decompressor
upon exit from the function. And it delegates the responsibility of
setting the decoder state to the caller (of which there is only one),
to ensure all error paths result in an error state.
The existing `::unite_horizontally()` and `::unite_vertically()` tests
did not properly test the edge cases where left/top in the Rect were
updated, so they get re-arranged a bit.
This improves the quality of our font rendering, especially when
animations are involved. Relevant changes:
* Skia fonts have their subpixel flag set, which means that individual
glyphs are rendered at subpixel offsets causing glyph runs as a
whole to look better.
* Fragment offsets are no longer rounded to whole device pixels, and
instead the floating point offset is kept. This allows us to pass
through the floating point baseline position all the way to the Skia
calls, which already expected that to be a float position.
The `scrollable-contains-table.html` ref test needed different table
headings since they would slightly inflate the column size in the test
file, but not the reference.
By doing that we eliminate the need for the vertical flip flag.
As a side effect it fixes the bug when doing:
`canvasContext2d.drawImage(canvasWithWebGLContext, 0, 0);`
produced a flipped image because we didn't account for different origin
while serializing PaintingSurface into Gfx::Bitmap.
Visual progress on https://ciechanow.ski/curves-and-surfaces/
If we have a valid PNG header with geometry info etc, we should still
display it as *something*, even if the image data itself is missing or
corrupted.
This matches the behavior of other browsers, and is something that
Cloudflare Turnstile checks for.
To achieve this, we split the PNG decoder's initialization into two
steps: "everything except reading frame data" and "reading frame data".
If the latter step fails, we yield a transparent bitmap with the
geometry from the PNG's IHDR chunk.
This would fail with EINVAL earlier, due to an attempt to create a
zero-length Core::AnonymousBuffer.
We fix this by transferring the buffer length separately, and only
going down the AnonymousBuffer allocation path if the length is
non-zero.
CSS filters work similarly to canvas filters, so it makes sense to have
Gfx::Filter that can be used by both libraries in an analogous way
as Gfx::Color.
