firefox, thunderbird, librewolf: Enable wayland support by default

[mweinelt]

Please read the commit message for details.

TL;DR: Always set MOZ_ENABLE_WAYLAND=1 to prevent XWayland usage on Wayland.

Description of changes

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 22.11 Release Notes (or backporting 22.05 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
  • (Release notes changes) Ran nixos/doc/manual/md-to-db.sh to update generated release notes
• Fits CONTRIBUTING.md.

Fixes #163206

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/firefox-all-black-when-first-launched-after-login/21143/15

[bjornfor]

So using MOZ_ENABLE_WAYLAND=1 also on X11 is fine?

Does anyone know why firefox doesn't enable this by default itself?

I've run with MOZ_ENABLE_WAYLAND=1 for a while on Wayland, and it does fix the occasionally black screen / crash on startup, so +1 from me.

[Tungsten842]

So using MOZ_ENABLE_WAYLAND=1 also on X11 is fine?

Does anyone know why firefox doesn't enable this by default itself?

I've run with MOZ_ENABLE_WAYLAND=1 for a while on Wayland, and it does fix the occasionally black screen / crash on startup, so +1 from me.

Some bugs supposedly:
https://bugzilla.mozilla.org/show_bug.cgi?id=1752398

[vcunat]

Funnily, there are both kinds in there – bugs in running on Xwayland ("Blocks") and bugs in running on Wayland ("Depends on").

resolved

[Atemu]

On the one hand, firefox-wayland is superior to the experience on xwayland and I've been using it because of that but OTOH Mozilla explicitly decided against enabling wayland by default and it's kind of their decision whether they deem their wayland support to be production-ready.

What Fedora did here was to enable wayland by default in firefox but offer a firefox-x11. I think we should do the same.

[lovesegfault]

Ah, right, this might have licensing implications since we'd be distributing a branded non-standard build.

Sigh.

[rhendric]

All this does is set an environment variable in the wrapper script we use to invoke Firefox; does that really constitute changing the build for purposes of licensing?

[mweinelt]

I don't think this affects licensing unless we severely degrade the UX.

[Atemu]

Well, the thing is, we could severely degrade UX by enabling wayland. You and I are comfortable with that but some of our users wouldn't even know that the terrible bug they just experienced could have been caused by wayland being enabled or that that's even the case.

I don't think Mozilla will be coming after us in any capacity if we did a minor thing like this (even if they'd rather we didn't) but, as a packager, it's a general courtesy to follow upstream projects explicit UX decisions wherever possible.

Fedora stoke a nice balance I think but, from Mozilla's perspective, they'd probably rather have us keep it the way it was (convenient yet explicit wayland opt-in).

[piegamesde]

I was baffled to find out that I'd been using Firefox with XWayland all the time. As of 2022, I'd expect applications to use Wayland by default if they support it. Since this is just an environment variable and not a compile flag, and that setting it does not affect X11 support, I am not concerned about licensing stuff.

However, I think we should still provide a way to opt-out of Wayland, i.e. to force the usage of XWayland if desired. Since we are setting the environment variable in the wrapper, setting MOZ_ENABLE_WAYLAND=0 won't work. Having a firefox-x11 package as discussed above is a viable option.

An alternative would be to set this environment variable globally, for example as part of a programs.firefox.wayland option that defaults to true. The obvious disadvantage is that this would then only apply to NixOS systems.

[squalus]

An alternative would be to set this environment variable globally, for example as part of a programs.firefox.wayland option that defaults to true. The obvious disadvantage is that this would then only apply to NixOS systems.

A nixpkgs config option would work, and apply to non-NixOS systems.

[ajs124]

Since we are setting the environment variable in the wrapper, setting MOZ_ENABLE_WAYLAND=0 won't work

if we use --set-default instead of --set, this won't be an issue

[jtojnar]

I was baffled to find out that I'd been using Firefox with XWayland all the time. As of 2022, I'd expect applications to use Wayland by default if they support it.

Then that should be taken up with Mozilla.

Since this is just an environment variable and not a compile flag, and that setting it does not affect X11 support, I am not concerned about licensing stuff.

The method does not really matter. GTK_USE_PORTAL was also an environment variable and it lead to a massive slowdown of some users’ boot that was difficult to pinpoint. The issue is that we would be potentially making a non-trivial change in behaviour.

And since Mozilla only allows us to use the Firefox trademark as long as our changes do not break their product, it is our responsibility to make sure that any changes are in fact non-breaking, if we want to keep using the Firefox branding.

At minimum, we should probably find out why exactly it is not enabled by default. But final decision whether the change would be fine with Mozilla can only be done by Nixpkgs Firefox maintainers, hoping that their mental model of Mozilla is good enough.

[colemickens]

(fwiw, I generally agree with @jtojnar)

[mweinelt]

The basic change here is that nobody will use XWayland by default anymore. Nobody will be able to pick the wrong firefox package anymore. XWayland is most certainly the variant that gets the least maintenance at this point. Accessibility features like the onscreen keyboard don't seem to work under XWayland.

I think keeping a firefox-x11 variant around is pretty much pointless, when Firefox with Wayland enabled will gracefully fallback to X11 when the session is based on X11. There is also the possibility to force the usage of X11 by overriding MOZ_ENABLE_WAYLAND as outlined by @ajs124.

We're not the first distro to enable Wayland by default, Ubuntu went first.

I do not expect any licensing issue with this.

[vcunat]

I believe this PR is good as it is now. The variable is there.

I've also been using this PR on pure X11 since the moment it was open. The wayland variant has been heavily tested by other nixers for a long time, so it's not anything new.

The variable override sounds sufficient at this point, as so far there's no evidence that it would be commonly needed and it works across all the packages using the wrapper.

[maralorn]

Have been using this for years, works great.

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/breaking-changes-announcement-for-unstable/17574/16