This will begin the process of breaking up the `useLLVM` monolith. That
is good in general, but I hope will be good for NetBSD and Darwin in
particular.
Co-authored-by: sterni <sternenseemann@systemli.org>
During d6d4228b39 I failed to notice that the current chromiumDev
version is older than the first one that contained the commit to fix the
dependency on opus in webcodecs.
This should hopefully fix build of chromiumDev (if there are no
additional issues).
The build was failing with:
In file included from ../../third_party/blink/renderer/modules/webcodecs/audio_encoder.cc:7:
In file included from ../../media/audio/audio_opus_encoder.h:16:
gen/shim_headers/opus_shim/third_party/opus/src/include/opus.h:5:10: error: 'opus.h' file not found with <angled> include; use "quotes" instead
#include <opus.h>
^~~~~~~~
"opus.h"
[...]
fatal error: too many errors emitted, stopping now [-ferror-limit=]
20 errors generated.
[42272/44233] CXX obj/third_party/blink/renderer/modules/webcodecs/webcodecs/decoder_template.oo[K
Note: This also fixes the ungoogled-chromium channel name in versionRange.
The "channel" variable shouldn't be part of the final derivation. This
also makes it possible to avoid unnecessary rebuilds for identical
channels (e.g. major updates are tested via the "beta" channel first and
usually neither the source-code archive nor the dependencies change when
the update makes it into the "stable" channel - this means we could
better use chromiumBeta to test major updates in advance).
continuation of #109595
pkgconfig was aliased in 2018, however, it remained in
all-packages.nix due to its wide usage. This cleans
up the remaining references to pkgs.pkgsconfig and
moves the entry to aliases.nix.
python3Packages.pkgconfig remained unchanged because
it's the canonical name of the upstream package
on pypi.
This also adds a dedicated channel for ungoogled-chromium that enables
us to update ungoogled-chromium independently of chromium.
TODO: Automate ungoogled-chromium updates via update.py (currently it
needs to be updated manually).
Note: Unfortunately this changes the ungoogled-chromium derivation
because common.nix passes the channel as an argument to
stdenv.mkDerivation (this makes it more difficult to verify this commit
but the result should remain the same).
I used nix-instantiate to verify that the derivations for chromium and
ungoogled-chromium remain unchanged (only the meta attributes change
slightly as I added myself as ungoogled-chromium to receive
notifications for PRs/issues).
The gn version depends on the channel and new gn versions aren't always
backward compatible. Therefore we should also include it in
upstream-info.json (I've scoped it under "deps" as we'll likely have to
add more like this in the future).
LLD: https://lld.llvm.org/
When you link a large program on a multicore machine, you can expect that LLD runs more than twice as fast as the GNU gold linker. Your mileage may vary, though.
Link-time optimization (LTO) is supported by default.
Some default settings have been tuned for the 21st century. For example, the stack is marked as non-executable by default to tighten security.
LTO & ThinLTO: https://clang.llvm.org/docs/ThinLTO.html
LTO (Link Time Optimization) achieves better runtime performance through whole-program analysis and cross-module optimization. However, monolithic LTO implements this by merging all input into a single module, which is not scalable in time or memory, and also prevents fast incremental compiles. ThinLTO is a new approach that is designed to scale like a non-LTO build, while retaining most of the performance achievement of full LTO.
PGO: https://llvm.org/docs/HowToBuildWithPGO.htmlhttps://blog.chromium.org/2020/08/chrome-just-got-faster-with-profile.html
Allows your compiler to better optimize code for how it actually runs. Users report that applying this to Clang and LLVM can decrease overall compile time by 20%.
Because PGO uses real usage scenarios that match the workflows of Chrome users around the world, the most common tasks get prioritized and made faster. Delivers up to 10% faster page loads.
CFI: https://clang.llvm.org/docs/ControlFlowIntegrity.htmlhttps://www.chromium.org/developers/testing/control-flow-integrity
Aborts the program upon detecting certain forms of undefined behavior that can potentially allow attackers to subvert the program’s control flow. These schemes have been optimized for performance, allowing developers to enable them in release builds.
By default, a program compiled with CFI will crash with SIGILL if it detects a CFI violation.
Additionally:
Use minizip instead of zlib. Chromium says zlib but actually uses minizip.
Remove old unused workarounds.
Make shell scripts POSIX compliant.
Update documentation URLs.
Prepare for using system libraries.
This should also fix VA-API for chromiumBeta (though that part needs
some cleanup). However, chromiumDev likely still fails due to the
absence of dirmd (not included in the tarball so far, we might have to
package and add it as a dependency).
Wanted to do this for a long time to collect important knowledge and
make it easier to pass maintainership.
Only time will tell if this'll be useful or become outdated instead.
Chromium 86.0.4240.75 builds fine without this patch. And since
WEBP_MAX_DIMENSION is the same in the system libwebp this patch should
not be required anymore (it was introduced in 06ec2a9f19, apparently to
fix the build).
By default GN produces a build with all of the debug assertions enabled (is_debug=true) and including full debug info (symbol_level=2). Setting symbol_level=1 will produce enough information for stack traces, but not line-by-line debugging. Setting symbol_level=0 will include no debug symbols at all. Either will speed up the build compared to full symbols.
This is done to avoid driver specific issues and restores the previous
behaviour. Like before video acceleration can be enabled without having
to rebuild Chromium.
This will additionally install the following files:
libEGL.so libGLESv2.so
libVkICD_mock_icd.so libvk_swiftshader.so libvulkan.so
libEGL.so and libGLESv2.so are required to fix our ANGLE support.
The rest should help with the Vulkan support (currently an experimental
feature that is disabled by default).
Michael Weiss