wptrunner internally derives the path to the MANIFEST.json file from the
`metadata_path` passed via `test_paths`. The current logic overrides
only the `metadata_path`, leaving the manifest_path pointing at the
wrong (non-legacy) layout's MANIFEST.json.
In #34436 we observed that the recent WPT imports create a transient
`.cache` diretory that wptrunner logic uses for optimization. This has
not been observed until #34436 because a [recent bump in MANIFEST.json's
schema version][1] triggered the creation of the cache. Because of the
above issue with incorrect path and the fact that we *first* trigger
`mach update-wpt` for legacy layout during WPT import, the MANIFEST.json
of non-legacy layout gets incorrectly migrated during the invocation of
`update-wpt` for legacy layout but the cache is still created under
legacy-layout's path as it is not based on `manifest_path`. The
subsequent invocation of `mach update-wpt` for non-legacy finds the
MANIFEST.json already migrated so the `.cache` directory is not
constucted.
This change simply replaces the whole object using the wptrunner's
`TestRoot` class constructor so that all derived paths are calculated
correctly. We also add the `.cache` folders to gitignore as it seems
like they are expected to be created during such version migrations.
[1]: https://github.com/web-platform-tests/wpt/pull/49406
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
- The standard `dev` mode is often too slow for daily work on servo
- Increasing the optimization level to 2 improves the performance noticably.
- Build time for opt-level = 1 is slightly higher than 2 on my machine
- Reducing debug info to `line-tables-only` improves link and build times,
while still keeping relevant info for backtraces.
Also extend .servobuild to allow setting custom cargo profiles instead
of just release and dev.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
Windows and Mac have unit tests enabled when triggered via 'full'
alias but they are disabled on Linux which is confusing.
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
Co-authored-by: sagudev <16504129+sagudev@users.noreply.github.com>
Bumps notify-py to the latest version v0.3.43.
Release notes: https://github.com/ms7m/notify-py/releases/tag/v0.3.43
This fixes notifications on windows 11.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
`hvigorw --version` might fail if some required environment variables
are not set. In this case we should print the output of the command,
so that the user can report what caused `hvigorw --version` to fail.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* ohos: Add basic IME and keyboard support
- Add extremely basic support for keyboard events
- Add basic IME support
- Showing and hiding the IME
- inserting text
- deleting characters
- very basic configuration of the IME
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* Apply suggestions from code review
Improve the log message
Co-authored-by: Josh Matthews <josh@joshmatthews.net>
Signed-off-by: Jonathan Schwender <55576758+jschwe@users.noreply.github.com>
* Update ports/servoshell/egl/ohos.rs
Co-authored-by: Mukilan Thiyagarajan <mukilanthiagarajan@gmail.com>
Signed-off-by: Jonathan Schwender <55576758+jschwe@users.noreply.github.com>
* ohos: Bump the minimum required SDK version to 5.0
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* ohos: Remove pub from callbacks
The callbacks don't need to be public, as we will be registering them.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* Rename composition event
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* ohos: clippy in log
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* ohos: address some clippy warnings
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* ohos: Raise Error in mach if unsupported SDK version is used.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* Add keyboard-types dependency for android
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
---------
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
Signed-off-by: Jonathan Schwender <55576758+jschwe@users.noreply.github.com>
Co-authored-by: Josh Matthews <josh@joshmatthews.net>
Co-authored-by: Mukilan Thiyagarajan <mukilanthiagarajan@gmail.com>
As mentioned in https://github.com/servo/servo/pull/34257
we should also lock the cargo-deny version in mach bootstrap.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
With the latest release (5.0) of the OpenHarmony SDK
libclang.dll is now available and we can support
building for OpenHarmony from windows hosts.
Other changes required for building OH on windows:
- We can't use the `<target_triple>-clang` wrappers, since
those are bash scripts, which fails on windows when cc-rs
tries to directly call them. However, we already pass all
the required flags the wrapper script would set, so this
is not an issue.
- We need to use posix paths, otherwise the sysroot parameter
will not be applied correctly (by bindgen). It seems to only
cause issues with bindgen in practice, possibly because
bindgen interprets the path with
[`shlex::split`](8a6d851318/bindgen/lib.rs (L312C27-L312C40))
which presumably causes the issues with windows paths.
To be consistent I decided to use posix paths for all paths.
- Fix checks for copying dlls. We need to check the target OS,
not the host OS when determining what libraries to copy.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
- We should use the CMake version from the NDK
- ignore errors from unused commandline parameters,
which can easily happen due to --gcc-toolchain, which is added by the
default CMake code when compiling with Clang.
- Previously servo didn't have any CMake dependency projects anymore,
so these issues didn't surface.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* fixed some clippy warnings
Signed-off-by: L Ashwin B <lashwinib@gmail.com>
* Delete extra file
Signed-off-by: chickenleaf <lashwinib@gmail.com>
* preserved newline in compositionevent.rs
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
Signed-off-by: chickenleaf <lashwinib@gmail.com>
* removed the newline in PrototypeList
Signed-off-by: L Ashwin B <lashwinib@gmail.com>
* removed the trailing whitespace
Signed-off-by: L Ashwin B <lashwinib@gmail.com>
---------
Signed-off-by: L Ashwin B <lashwinib@gmail.com>
Signed-off-by: chickenleaf <lashwinib@gmail.com>
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
Move `--flavor` to a new "Packaging options" group, so that
we can only apply the flavor option to the commands that support it.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* ohos: Support product flavors
Support different product flavors (e.g. with different signing
configurations) by supporting --flavor on ohos.
The flavor influences the package path for OH packages, so
allow the flavor parameter on build + package + install
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* Fix smoketest
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
---------
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
Bundle resource files into the .hap, so they are available
as files in the application sandbox, instead of included
into the shared library.
This should slightly reduce the binary size in debug and
release mode.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* android: publish nightly builds for aarch64
Publish the aarch64 apk and aar packages to both Github Releases
and S3 so that it can be linked from servo.org.
The focus is on getting a working version of the APK on the homepage,
so few issues are resolved with temporary solutions:
1) We publish the "release" profile instead of "production" since the
latter will need changes in the gradle configuration (the changes
required was previously blocked on #32720 which is now closed).
2) The scheme for the version code is simple and doesn't consider
other factors such as API level and product variants discussed in the
Android docs (https://developer.android.com/google/play/publishing/multiple-apks#VersionCodes)
This should be fine for now as we don't publish to any store yet.
The change also makes it so that the 'Release nightly' workflow will
endup building all 4 variants for the Android target, but only aarch64
is uploaded. This is because GH Actions doesn't have a good way to skip
a specific job in a matrix and the additionally code complexity needed
to acheive it (either generating a JSON dynamically in a new job and using
`fromJSON` in the matrix definition or skipping each individual step
based on matrix.target and `inputs.upload`) didn't seem worth the cost
saved (this is executed only once a day).
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
* android: add attestation for nightly build artifacts
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
---------
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
* util.py: Remove six
We don't need to support python2 anymore, so we can just use
`str(x)` instead of `six.ensure_str(x)`
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* config.py: Remove six
We don't need to support python2 anymore, so we can just use
`str(x)` instead of `six.ensure_str(x)`
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* main.py: Remove six
We don't need to support python2 anymore, so we can just use
`str(x)` instead of `six.ensure_str(x)`
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* main.py: Fix `--settings` being ignored
Previously `paths` was unused in this function,
and the usage for the settings_file would have no effect.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* terminal.py: Remove `six`
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* registrar.py: Remove `six`
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* mach/util.py: Remove `six`
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
* mach: Remove python2 from the list of programming languages
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
---------
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
This change fixes all rustdoc errors and also adds a tidy check for a
very common rustdoc URL issue. Eventually rustdoc warnings should likely
cause the build to fail, but this catches those issues sooner in order
to not waste so much developer time.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
- Previously on fedora `./mach bootstrap` would always detect it needs to
reinstall packages and require root permissions.
- use custom queryformat for `rpm -qa` to to just get the package name
(e.g. `openssl-libs` instead of `openssl-libs-3.2.2-3.fc40.i686`
- Use a list to store the output result instead of one string
- Fedora (40) installs `zlib-ng` instead of `zlib` and `libjpeg-turbo` instead
of `libjpeg`, meaning that `rpm` / dnf commands report `zlib` as not installed.
Specifying the actually installed package avoids this problem.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
self.config["android"]["target"] is unset, causing an exception.
We can just use self.target.triple() instead.
Signed-off-by: Jonathan Schwender <schwenderjonathan@gmail.com>
The GitHub search API is a bit sensitive. There isn't a great way to
search for the repository organization and the branch name when looking
for open PRs. Instead use the bot username as the author name, which
should likely have been having before. This fixes the WPT export job.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
This switches the WPT export script to use Servo's fork the `wpt`
upstream repository. The reasoning is that Servo developers's can fix
issues with branches there. This isn't possible because no one has
permissions to the servo-wpt-sync fork of the repository.
I have given `servo-wpt-sync` write persmissions to `servo/wpt`.
This also updates the tests a bit to reflect the new default repository.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
This was due to uname reporting the architecture as arm64
instead of aarch64 on macos
Signed-off-by: Benjamin Vincent Schulenburg <bennyschulenburg@gmx.de>
* mach: Do not manually install toolchain for non-cross builds
This fixes the Apple Silicon build where the installed toolchain does
not match the host triple. We only install toolchains manually for cross
builds, because cargo and friends will do this automatically when
executing.
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
* fix lints
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
---------
Signed-off-by: Martin Robinson <mrobinson@igalia.com>
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
Co-authored-by: Mukilan Thiyagarajan <mukilan@igalia.com>
Introduce a new `BuildTarget` abstraction to centralize the code for
supporting different ways of choosing the build target (e.g --android,
--target x86_64-linux-android , --target aarch64-linux-ohos). This
is currently handled in an adhoc fashion in different commands (
mach package, install, run) leading to a proliferation of keyword
parameters for the commands and duplicated logic.
The patch introduces a new `allow_target_configuration` decorator to
do the validation and parsing of these parameters into the appropriate
`BuildTarget` subclass, which is now stored as an instance attribute
of the CommandBase class. All the code that previously relied on
`self.cross_compile_target` has been switched to use the BuildTarget.
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
* Sync `GPUObjectDescriptorBase` (label is not option anymore)
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Sync `GPUFeatureName`
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* shader_f16 feature is not usable in wgpu so disable it
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* sync `GPUTextureFormat`
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* `validate_texture_format_required_features`
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Sync `GPUTexture` attributes
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Make `entryPoint` in `GPUProgrammableStage` optional
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Set good expectations
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Bad expectations because naga does not support cons declarations
Also fail on firefox, where skipped before due to missing device features
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Bad expectation, also fails on firefox
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Bad expectations, because naga does not support `let pos = positions[vertex_index];`
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Set expectation
external texture does not work in firefox too (again naga)
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* set bad expectations, because naga does not support `enable`
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Set bad expectations for, `Texture with '' label has been destroyed`
also fails in firefox with same reason
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* one bad expectation
also on firefox
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* expect that also matches firefox
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* more expect
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* Use only 1 proc for _webgpu
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
* better doc comment
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
---------
Signed-off-by: sagudev <16504129+sagudev@users.noreply.github.com>
This code is unused and unmaintained. We can bring it back from version
control if needed in the future.
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
* ohos: Fix compilation for x86_64
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
* ohos: Use the SDK compiler-wrapper
When compiling for x86_64-unknown-linux-ohos without the compiler
wrapper, for some reason mozjs_sys will be refercing a wrong mangled symbol
resulting in the following error when loading the .so at runtime:
```
_ZNSt3__111this_thread9sleep_forERKNS_6chrono8durationIxNS_5ratioILl1ELl1000000000EEEEE: symbol not found
```
This problem does not occur when compiling for aarch64 or when using the compiler wrapper.
In this case the correct symbol
`_ZNSt4__n111this_thread9sleep_forERKNS_6chrono8durationIxNS_5ratioILl1ELl1000000000EEEEE`
is referenced.
It's unclear why manually passing the flags via CFLAGS / CXXFLAGS does not work.
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
---------
Signed-off-by: Jonathan Schwender <jonathan.schwender@huawei.com>
In Japanese Windows, `vswhere` puts cp932 string in default and fails to
load json in Pyhton.
This fixes it by adding `-utf8` option
Signed-off-by: hatoo <hato2000@gmail.com>
The default user name in Windows installations is of the form "FirstName
LastName", so it seems likely that there will be spaces in the user's
path. Based on my testing on Windows 11, the only Servo's bootstrap
script has trouble dealing with spaces in paths. This patch fixes that
by quoting such paths correctly. Our direct and indirect dependencies
seem to handle these without issue and Servo does build and run
correctly with this patch.
In this patch, the logic for gstreamer bootstrap now uses powershell
instead of directly invoking msiexec.exe via cmd.exe as I was unable to
get the installer to run correctly, even with quoting. Some extra hacks
were necessary to propagate the exit code correctly to mach.
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
