* build: Clean up post-build copy of Windows DLLs
- No longer use vcvarsall.bat at all. Instead find the Windows SDK
directory by looking in the registry.
- Split logic for copying Windows dependencies into its own function and
do some minor clean up, such as collecting all MSVC functionality into
visual_studio.py.
- Remove support for Visual Studio 2015 and Visual Studio 2017.
This is a preparatory change in order to support Visual Studio 2022.
* More cleanup of the code
This makes it more foolproof to install crown from inside the Servo
directory, because the root Servo config.toml overrides the rustc to use
crown (an obvious circular dependency).
This was enabled to allow using the simd / std::simd / packed-simd crate
in the glyph cache [1][2]. Support for simd in the gfx crate was removed
though [3], so this flag is not really doing anything -- and the Android
build is currently broken. Plus, it's unclear what target features we
can enable using stable Rust. We can explore adding neon support when
Android is working again.
This is part of a long-term effort to remove build complication and make
it so that `cargo build` is equivalent to `./mach build`.
1. https://github.com/servo/servo/pull/10916
2. https://github.com/servo/servo/pull/10900
3. https://github.com/servo/servo/pull/24304
Add documentation and simplify the way that GStreamer shared objects
lists are stored. In addition, move the list of extra GStreamer DLL
dependencies to to the `gstreamer.py` file.
The conditional plugin logic is no longer required as we are having
already increased our GStreamer version requirements.
Ubuntu 20.04 doesn't have a new enough version of GStreamer, so
automatically disable media when running on that platform.
This also cleans up the media detection a bit, putting the result in a
`enable-media` variable and moving some of the logic into the build
scripts themselves rather than the platform module.
It's often the case (especially with the taplo installation and on
Windows) that bootstrap is doing lots of stuff in the background for a
long amount of time. Without output it's hard to tell what exactly is
going on. This change adds more output to this process as well as
removing some Pythong 2.x era code.
- Upgrade the version of GStreamer for Windows
This upgrades the Windows build to use the most recent version of
GStreamer. This is necessary to upgrade our GStreamer dependency.
- Stop shipping GStreamer binaries on Linux
The binary bundle of GStreamer that we package is not used to compile --
only to run layout tests. It's too old for the APIs that we are using
(as evidenced by needed 1.18 for WebRTC) and nowadays Linux
distributions carry a new version so it's unecessary for our build
machines. No longer using this binary bundle will allow us to upgrade
our GStreamer dependency -- which now has stricter checks that we
are using at least version 1.18.
- Upgrade media to use newer versions of GStreamer / GLib dependencies
1. Make the tidy output easier to follow
2. Integrate the WPT manifest cleanliness step into tidy
itself and don't run it if nothing has changed in the WPT
directory.
3. Fix an issue where Python test requirements were not installed,
which could cause issues with some modules not being found.
Fixes#30002.
hyper_serde changes very infrequently and typically only in order to
upgrade Servo dependencies. crates.io lists hyper_serde as having no
dependents, and its integration will not prevent releasing new versions.
Integrating it will both reduce the number of repositories we have to
maintain as well as making it easier to upgrade Servo dependencies.
* Replace virtualenv with Python's built-in venv.
* Apply Delan's suggestions and make a couple small fixes
- Fix a tidy warning about directories that don't exist
- Use shutil instead of the redundant get_exec_path
- Miscellaneous cleanups
* Fix typo in environment variable
* fix bug where pip still tries to the wrong site-packages
---------
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
Co-authored-by: Delan Azabani <dazabani@igalia.com>
This completes the transition to compiling Servo with rust stable. Some
nightly-only features are still used when compiling the `script` and
`crown` crates, as well as for some style unit tests. These will likely
break with newer compiler versions, but `crown` can be disabled for them
conditionally. This is just the first step.
This has some caveats:
1. We need to disable setting up the special linker on Linux. The -Z
option isn't supported with stable rust so using this is out --
meanwhile we can't be sure that lld is installed on most systems.
2. `cargo fmt` still uses some unstable options, so we need to rely on
the unstable toolchain just for running `fmt`. The idea is to fix this
gradually.
Rust uses a version of ThinLTO by default now [1]. This can be tweaked
by adjusting rust compiler flags, which is probably a better way of
controlling this than a custom servo configuration considering:
1. We want to remove the custom servo configuration eventually.
2. The -Z option that this configuration currently uses is unsupported
by stable rust.
1. https://blog.rust-lang.org/inside-rust/2020/06/29/lto-improvements.html
* Remove script_plugins
* Use crown instead of script_plugins
* crown_is_not_used
* Use crown in command base
* bootstrap crown
* tidy happy
* disable sccache
* Bring crown in tree
* Install crown from tree
* fix windows ci
* fix warning
* fix mac
libscript_plugins.dylib is not available anymore
* Update components/script/lib.rs
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
* Update for nightly-2023-03-18
Mostly just based off https://github.com/servo/servo/pull/30630
* Always install crown
it's slow only when there is new version
* Run crown test with `mach test-unit`
* Small fixups; better trace_in_no_trace tests
* Better doc
* crown in config.toml
* Fix tidy for real
* no sccache on rustc_wrapper
* document rustc overrides
* fixup of compiletest
* Make a few minor comment adjustments
* Fix a typo in python/servo/platform/base.py
Co-authored-by: Samson <16504129+sagudev@users.noreply.github.com>
* Proper test types
* Ignore tidy on crown/tests
---------
Co-authored-by: Martin Robinson <mrobinson@igalia.com>
* Run main and try jobs with debug assertions
* use single quotes in workflow expressions
* set force-debug-assertions in main.yml
* set force-debug-assertions as part of decision job
* fix typo in MachCommands.build
* fix more hardcoded profile names
* fix tidy
* split cargo_profile_option on windows
* Fix running servoshell and unit tests through a symlink
* rename steps to make them less confusing
* fix more hardcoded cargo profile options
* fix missing inputs in linux-wpt and mac-wpt
* make filename an inherent method of Resource
* rework release-with-debug-assertions profile to production profile
* rework resource logic to eliminate std_test_override
* set production flag in nightly release builds
* clean up servobuild.example and windows.yml
* oops forgot to check in embedder_traits/build.rs
* fix mach test-unit behaviour through symlink
* unit tests only need current_dir and ancestors
* fix macOS package smoketest breakage
* expect css/css-color/currentcolor-003 to crash under layout 2013
* fix more references to {force,release-with}-debug-assertions
* fix local build failures under --profile production
These need to be installed in order to build so we can install them via
Homebrew. Do this by simply restoring the Homebrew bootstrapping logic
we had in place previously.
Fixes#27171.
* Update mozjs
* moztools4 in bootstrap
* no autoconf
* tidy
* switch to servo-build-deps
* update mozjs for real
* glue mozjs
* fmt
* move to servo/mozjs
* cleanup and move user input logix into servoshell
* fix fmt
* moves test from servoshell file
* move command-line args into servoshell
* remove feature media-gstreamer
* fix fmt
* move user input logic code into lib to make it more testable
* remove opts_matches in fn instead get it from main2
* remove pub and fix import
* add licence in new file
* revert passing Matches, instead pass Option String
* review update, also move sanitize fn to parser file
* fmt fix
* review fix: remove extra line
LLVM is the largest package that we get from servo-build-deps, so
installing it via chocolatey should reduce the amount of data that we
transfer from that source. In addition, it's one less dependency that we
have to manage.
It also seems that installing LLVM to the default location with choco
means that we no longer have to set the LIBCLANG_PATH environment
variable for bindgen.
Co-authored-by: Mukilan Thiyagarajan <mukilan@igalia.com>
* Allow noidl files in script/dom/webidls
* Upgrade wgpu to 0.16 and refresh whole webgpu implementation
* Update WebGPU test expectations
* misc
* MutNullableDom -> DomRefCell<Option<Dom for GPUTexture
* Direct use of GPUTextureDescriptor
* Remove config from GPUCanvasContext
* misc
* finally blue color
* gpubuffer "handle" error
* GPU object have non-null label
* gpu limits and info
* use buffer_size
* fix warnings
* Cleanup
* device destroy
* fallback adapter
* mach update-webgpu write webgpu commit hash in file
* Mising deps in CI for webgpu tests
* Updated expectations
* Fixups
* early reject
* DomRefCell<Option<Dom -> MutNullableDom for GPUTexture
The -Zgcc-ld=lld flag makes rust use the rust-lld
linker that is distributed as part of rust toolchain.
However, this flag doesn't work on nixos correctly
as
1) rust-lld needs to be patched to have the correct rpath
to find libz.so
2) the bin/gcc-ld/ld.lld wrapper which calls rust-lld also
needs to be patched to use the correct dynamic loader
3) rust-lld doesn't respect NIX_LDFLAGS which contains
the additional search path derived from buildInputs.
The system linkers on nixos are wrapped so that
NIX_LDFLAGS is added as the rpath to the final binary.
Signed-off-by: Mukilan Thiyagarajan <mukilan@igalia.com>
It's unclear if the compiler flag was doing anything, but I've verified
(with otool) that the environment variable does affect the minimum
version of the MacOS set in the binary. We could examine later if this
is still necessary.
This was added in #23163 when switching CI from gcc to clang.
There are a few motivations for this change:
1. lld is demonstrably faster than gold, but is really only stable on
Linux at the moment. There's a good chance that it will be ready for
all platforms soon though.
2. Most people do not have gold installed on MacOS and Windows. You'd
have to do this manually through homebrew. I think it's a safe
assumption that this probably won't be slowing things down much on
those platforms.
3. We need to remove all configuration of the build that happens while
running `./mach build` if we ever hope to make `cargo build`
equivalent to the mach build. This unlocks static configuration of
the rustflags. One of the big blockers for proper `cargo build`
support.
There were some issues with the way that the `--release` and `--dev`
arguments were handled in mach commands.
- Not all commands accepted them in the same way. For instance `./mach
test-wpt` didn't really accept them at all.
- If you did not pass either of them, mach would try to guess which
build you meant. This guess was often quite surprising as it wasn't
printed and it depended on the state of the your target directory,
which is difficult to remember.
- The `dev` profile is colloquially called a "debug" profile and some
commands accepted `-d` or `--debug...` like arguments, but `--debug`
with `./mach run` meant run in a debugger. It was easy to mix this
up.
This change:
- Centralizes where build type argument processing happens. Now it the
same shared decorator in CommandBase.
- Uses a `BuildType` enum instead of passing around two different
booleans. This reduces the error checking for situations where both
are true.
- Be much less clever about guessing what build to use. Now if you
don't specify a build type, `--dev` is chosen. I think this behavior
matches cargo.
- Makes it so that `./mach test-wpt` accepts the exact same arguments
and has the same behavior as other commands. In addition, the suite
correct for `test-wpt` is removed. There are only two suites now and
it's quite unlikely that people will confuse WPT tests for rust unit
tests.
This change replaces OpenSSL with rustls and also the manually curated
CA certs file with webpki-roots (effectively the same thing, but as a
crate).
Generally speaking the design of the network stack is the same. Changes:
- Code around certificate overrides needed to be refactored to work with
rustls so the various thread-safe list of certificates is refactored
into `CertificateErrorOverrideManager`
- hyper-rustls takes care of setting ALPN protocols for HTTP requests,
so for WebSockets this is moved to the WebSocket code.
- The safe set of cypher suites is chosen, which seem to correspond to
the "Modern" configuration from [1]. This can be adjusted later.
- Instead of passing a string of PEM CA certificates around, an enum is
used that includes parsed Certificates (or the default which reads
them from webpki-roots).
- Code for starting up an SSL server for testing is cleaned up a little,
due to the fact that the certificates need to be overriden explicitly
now. This is due to the fact that the `webpki` crate is more stringent
with self-signed certificates than SSL (CA certificates cannot used as
end-entity certificates). [2]
1. https://wiki.mozilla.org/Security/Server_Side_TLS
2. https://github.com/briansmith/webpki/issues/114Fixes#7888.
Fixes#13749.
Fixes#26835.
Fixes#29291.
Type inference was incorrectly inferring that our `check_output()`
helper was returning `str` when in reality, it returns `bytes`. This
fixes the caller that was no longer decoding those bytes and fixes the
type annotation on the function.
