Playwright: browser improvements, update

[phaer]

Description of changes

• Support Firefox & Webkit tests on NixOS (and Ubuntu, i.e. in github actions) - previously only Chromium was supported.
• Add data from browsers.json to nixpkgs via update.sh
  in order to use patched upstream browsers without IFD
• Use autoPatchElfHook to make those browsers run on nixos
• Reuses attrs from pkgs.firefox-bin for firefox
• browser hashes are updated via update.sh
• theres a new nixos vm test for all 3 supported browsers.

TODO long term

• Consider using the same approach on darwin, where browser builds are currently impure (using playwright install)
• check playwrights browsers patches and consider using nixpkgs browsers to build from source, instead of pulling binaries for firefox, webkit and ffmpeg

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandboxing enabled in nix.conf? (See Nix manual)
  • sandbox = relaxed
  • sandbox = true
• 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/)
• 24.05 Release Notes (or backporting 23.05 and 23.11 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
• Fits CONTRIBUTING.md.

---

Add a 👍 reaction to pull requests you find important.

[phaer]

EDIT: Nevermind, I messed up. Going to push a fix right after local tests pass.

[nixos-discourse]

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

https://discourse.nixos.org/t/running-playwright-tests/25655/28

[nixos-discourse]

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

https://discourse.nixos.org/t/prs-already-reviewed/2617/1632

[wmertens]

LGTM - any reason why this isn't merged? (apart from the conflict)

[phaer]

@wmertens None that I am aware of. Just resolved the merge conflict which was due the recent python reformating PR.

[gungun974]

Just a question like that, but I see in that PR you update @playwright/test to version 1.42.1.

Should not update to 1.44.1 instead or it's okay you think ?

[phaer]

Yes, this PR is 2 months old already. Could upgrade if needed, but as written in the resolved thread in #298944 (comment) I'd prefer to do so in #302759 in to keep churn to a minimum:)

[SuperSandro2000]

Please cleanup the commit history

[teto]

the test times out locally too (x86)

[phaer]

the test times out locally too (x86)

Hm...I wonder what the issue is here. It just ran successfully for me on two different x86_64-linux nixos machines; one of them not managed by to check whether it's some obscure setting or so ;) It also succeeded on on aarch64-linux in ofborg. Going to ask ofborg to run again to see whether it's flaky?

EDIT: It's either flaky or i misunderstand how the checks are run in CI; can re-produce after a rebase.

@teto At this point I am guessing, but could you check whether adding --disable-gpu to chromiums flags in the test changes something for you?

Just wanted to use firefox with playwright half a year ago, wrote a test and now I am trying to get the test to pass with chromium 😅

[phaer]

@ofborg test playwright-python

[phaer]

Rebasing on top of the new playwright release in master and using --disable-gpu seem to have fixed the test, @teto?

The webkit minibrowser is broken inside the test, but seems to work outside of it. I excluded it from the test for now, as my priority was to get firefox working and a nixos vm test for the browsers to begin with :)

[SuperSandro2000]

looks fine to me otherwise, didn't test

[SuperSandro2000]

We could make this unconditional but whatever. Fine by me.

[phaer]

I think it won't hurt to be explicit about supported platforms. :)

[phaer]

Applied review suggestions; hope this is finally good to merge 🙏 😉

---

short rant about late-stage style reviews and workflows here:

I really long for a future where we either have a linter (in CI and locally) for things like sha256->hash when its deemed important enough.
And/or a platform which makes it easier to review rebased commits efficiently. The problem with such reviews late in the cycle is that it technically invalidates prior reviews (and produces additional overhead for a pull, fix, rebase, push cycle compared to a linter).

I also believe we could have done those changes in the next PR (pure macOS browsers are planned and I am interested in browsers built from source, but let's see how long my energy lasts for this 😅 )

---

the innocent change from sha256 -> hash caused:

• ofborg failing to eval because i forgot to update the hashes
• me running update.sh to check if it works for the full pipeline
• seeing no git diff
• discovering that nix-prefetch-url seemingly doesn't support a format fitting hash
• rewriting it to use update.sh `nix store prefetch-file --json --hash-type sha256 --unpack "$1" | jq -r .hash
• waiting for the downloads
• rebasing to squash the commits

and all that for no functional difference while there's about 56k instances of sha256 usage
in master according to rg. 😕

[teto]

this comment is outdated but let's remove it in future PRs.

[teto]

the test still times out hre for some reason but let's not penalize contributors for adding tests. Considering the effort here, I assume you are in for the long run :)

[altano]

🎉🙏