We can't just edit binary wrappers in place because of a length
mismatch, so we have to parse the generating makeCWrapper call out of
the binary, extract wrapper arguments from it and add them to the
Firefox wrapper.
All these contortions are needed because Firefox looks for its runtime
in argv0, so the proper argv0 needs to be set by wrappers to always
point to the "final" runtime. I think this could be avoided by wrapping
/lib/$libName/firefox instead of /bin/firefox, and I'd like to look into
that in the future, but for now I'm just fixing the immediate problem.
Firefox uses a sandboxing model that only allows access to paths that
were previously explicitly granted. We can only add granular permissions
to a specific sandbox by patching, because setting LD_LIBRARY_PATH would
affect all of them.
To use hardware decoding via VA-API with Firefox 98.0.2 one needs to
head to `about:config` and enable `media.ffmpeg.vaapi.enabled`.
Closes: #157061
Both LTO and PGO require the use of clang so I think its easier to just
stick with clang for all builds, so PGO and LTO could in theory be used
without each other.
Lo and behold, we're finally catching up with Mozillas very own firefox
build in terms of speed.
PGO is an optimization technique in which in a first step we create a
build that supports instrumentation, meaning we can use it to create a
profile of how the browser behaved during usage. Then in a second pass
we create the final build that uses the acquired profiling data to
optimize the browser for the workload it actually received during
profiling.
The downside is that with PGO we now need to build Firefox twice, which
increases the build time from around 20 minutes to roughly 50 minutes.
In the Speedometer 2.0 benchmark multiple tests could see a
responsiveness improvemeant around 20-25%, which makes the increased
build time well worth it.
Sadly this benefit seems limited to x86_64-linux, builds on
aarch64-linux get stuck during profiling and I haven't found out why.
Finally, after a long time, we can say:
Closes: #76484
Supersedes: #129503
In #109133@alyssais discovered that the drmSupport flag stopped
working. This is because Mozilla decided around Firefox 51
(mozbz#1289634) to swap the default values and our flag was asking for
the wrong thing all along.
Since this flag has now been enabled for multiple years, disabling it
would mean a regression for our users. Leaving it enabled should be
unproblematic since it only controls whether Firefox shows the EME nagbar,
that allows to enable Widevine CDM, when a site requests it. The choice is
therefore completely up to the enduser.
Disabling this nagbar is still possible at runtime by setting
`browser.eme.ui.enabled` to `false`. If Widevine CDM was inadvertently
enabled it can be disabled at `media.gmp-widevinecdm.enabled`.
Supersedes: #109133
It is normal for such a large derivation to accumulate mixed codestyles
over all those years making it harder to maintain.
By splitting up dependencies to one per line I noticed two unused
dependencies (hunspell, sqlite) that were removed in this change.
Also most lists are now sorted alphabetically, so reasoning about them
has become much easier.
Lists now use the same style for indentation and concat, and environment
variables are all defined in preConfigure as opposed to bare in the
derivation.
The object dir, firefox build directory, is now in a predictable
directory at ${pwd}/mozobj as opposed to ${pwd}/obj-@CONFIG_GUESS@,
saving us one wildcard expansion. This is also a good prepration for the
PGO build where we need to reference bits inside this directory.
Added a few comments and clarified others from what I've learned how the
firefox build works.
Since Firefox 51 the --enable-system-ffi flag was renamed to
--with-systemd-ffi when it moved to the javascript toolkit parts. Rename
the flag accordingly and reintroduce the dependency.
mozbz#1294803
Darwin support was marked broken in 2019 with Firefox 69 and has missed
therefore missed out and not been tested on the following 29 major
releases since.
It cannot be supported again without a darwin user stepping up to take
care and work on and test every major release, which hasn't happened
since I took over maintainership.
The recommendation of the people that tend to the firefox source build
is for darwin users to use firefox-bin instead.
Firefox does not support passing pipewire as a system library and
instead relies on a vendored copy it ships. We keep the flag because it
is tied into the wrapper, because we still need to have access to its
libraries at runtime.
Tied to the geolocation support flag. On Linux this uses the DBus
interface of NetworkManager to query visible accesspoints and relay them
to Mozilla location service in return for position approximation.
This is for consistency with other extra arguments that can be handed to
firefox common.nix.
Also pull the patch phase of the derivation closer together and use the
same list concat style as elsewhere.
With Firefox 98.0 the remoting name now depends on the update channel
(mozbz#1752418), which resulted in a weird app_id/wmclass of
`firefox-default`, which broke window association in GNOME and likely
other desktops.
Fixes: #165107
Naïm Favier