This should help us avoid accidentally breaking the build for AArch64.
Currently, some tests are expected to fail, so CI runs will be
considered successful even if the kernel panics or if there are test
failures.
For now, we have to build Qemu with a custom patch from source in order
for SystemServer to detect self-test mode.
When building for AArch64 with UBSan enabled, GCC 13.1 reports a false
"array out of bounds" error on access to offset `1 * sizeof(u64)`.
Changing the order of the stores seems to silence it.
These are not useful for us, and unnecessarily waste time when building
the toolchain. Even if we used libstdc++ more heavily, our use of
uncommon, PCH-affecting flags and our frequent LibC header changes would
make these ineligible for use anyways.
This is the only kernel issue blocking us from running the test suite.
Having userspace backtraces printed to the debug console during crashes
isn't vital to the system's function, so let's just return an empty
trace and print a FIXME instead of crashing.
Instead of manually compressing/decompressing a toolchain tarball if
`TRY_USE_LOCAL_TOOLCHAIN` is set, let's use the cache action's automatic
built-in compression (which is zstd, I believe).
This version now natively supports read-only caches (`cache/restore@v3`)
so we no longer need to pin the version to a commit in actions/cache#489
which is an unmerged PR.
The update is mostly mechanical:
- Steps with `CACHE_SKIP_SAVE` not set can use the plain `cache@v3`
action.
- Steps with `CACHE_SKIP_SAVE` set to a constant `true` are changed to
`cache/restore@v3`.
- Steps with saving disabled when running on a pull request are changed
to a pair of `cache/restore@v3` and `cache/save@v3`. This setup is
used for the large (100s of MB) ccache and Toolchain caches. As caches
saved in pull requests can only be utilized from within the same PR,
uploading these would only waste time and our storage quote.
Therefore, we skip the `save` steps if running on a PR.
Co-authored-by: Cameron Youell <cameronyouell@gmail.com>
Instead of eagerly populating the stack trace with a textual
representation of every call frame, just store the raw source code range
(code, start offset, end offset). From that, we can generate the full
rich backtrace when requested, and save ourselves the trouble otherwise.
This makes test-wasm take ~7 seconds on my machine instead of ~60. :^)
This PR corrects the signs of the start/end points that are passed
to the elliptical_arc_to(), it also removes the bogus path.close()
after creating the arc. Now .arc() seems to work correctly.
Instead of starting a new headless-browser for every layout & text test,
headless-browser now gets a mode where it runs all the tests in a single
process.
This is massively faster on my machine, taking a full LibWeb test run
from 14 seconds to less than 1 second. Hopefully it will be a similarly
awesome improvement on CI where it has been soaking up more and more
time lately. :^)
...and use it for `margin` and `padding`.
So now, if all four margins are 12px, we output "12px" instead of "12px
12px 12px 12px". This matches the spec's requirement to use the smallest
non-ambiguous representation.
Serenity handles this in the SystemServer already, but the simplest
place to put this logic is the ViewImplementation base class.
This is trivial to see when running Ladybird without SERENTIY_SOURCE_DIR
set, or set improperly.
This closely mirrors Socket::set_blocking. Note that it does not make
sense to make this a virtual method on a base class, since
SeekableStream and FixedMemoryStream cannot possible be anything except
than blocking.
Previously, this would blindly use parse_number() without checking if
there was a number to parse. This meant we would crash on transforms
like:
rotate(BAD 32 0)
Now the attribute just gracefully fails to parse.
BrowsingContext::set_active_document() may end up asking for the Page's
top level browsing context before Page has one. Do a workaround for now
where we have an API to ask if it's initialized.
Long-term we should find a cleaner solution.
This also straightens out the logic to determine the project_path.
Instead of calling realpath on potentially-null strings and sometimes
not even reading the result, we now only make these calls when required,
and properly handle any error.
This fixes an issue where loading an iframe would cause the current
browser tab title to get overwritten with an empty string.
The problem is that nested browsing contexts can be considered "top
level" during their initialization, but only one browsing context is
ever the Page::top_level_browsing_context(), so that's what we check.
This looks a lot more "at home" than usual pixel art logo on
non-SerenityOS systems. :^)
Also, stop using site favicons as the app icon as that made it
annoyingly hard to find Ladybird in task switchers sometimes.
Although we translate e.g `block` to `block flow` for internal use in
the engine, CSS-DISPLAY-3 tells us to use the short form in
serializations for compatibility reasons.
This adds 9 points to our score on https://html5test.com/ :^)
This will contain several of the fixed data tables from the VP8 spec.
For starters, it contains the tables needed to read the frame header
in the first partition. These tables are needed to read the
probabilities of the metablock predicition modes, which in turn will
be needed to read the metablock predicition modes themselves.
