ffmpeg-headless: uplift #354952
ffmpeg-headless: uplift #354952
Conversation
/nix/store/m5ghr321rsxpfl001c5fdh56y9f42lpd-ffmpeg-headless-7.1-bin 233072648 /nix/store/rjimh83l9pdpmd7qvanvfhkwr13cmbz5-ffmpeg-headless-7.1-bin 234001624
/nix/store/rjimh83l9pdpmd7qvanvfhkwr13cmbz5-ffmpeg-headless-7.1-bin 234001624 /nix/store/pk5bgvay9yb1hl6pcda2fs9d33s7gws0-ffmpeg-headless-7.1-bin 234002056
/nix/store/pk5bgvay9yb1hl6pcda2fs9d33s7gws0-ffmpeg-headless-7.1-bin 234002056 /nix/store/4gmnzizak5ca1skhf9r9m2m0jcbik8ap-ffmpeg-headless-7.1-bin 236116824
/nix/store/4gmnzizak5ca1skhf9r9m2m0jcbik8ap-ffmpeg-headless-7.1-bin 236116824 /nix/store/nqardac1g5h774bprz145fsl4qm2p92l-ffmpeg-headless-7.1-bin 236173272
/nix/store/nqardac1g5h774bprz145fsl4qm2p92l-ffmpeg-headless-7.1-bin 236173272 /nix/store/a9a7n7w0dhcjay5vi41xy685m6md1hkw-ffmpeg-headless-7.1-bin 294694384 This may look like quite an increase but if you combine it with a noteworthy package such as pipewire, the increase is only ~5MiB: /nix/store/cs43xb7x8bqy0igaip0vrsgjij5aafc9-foo 657507664 /nix/store/xcgc1k9mn8f9fqns9832dnnlhnxm6dka-foo 661376384 With actual closures, I believe this would be even less.
/nix/store/a9a7n7w0dhcjay5vi41xy685m6md1hkw-ffmpeg-headless-7.1-bin 294694384 /nix/store/2wxab02vqydrmm19m4wiffz1ic3v7528-ffmpeg-headless-7.1-bin 295755456
/nix/store/2wxab02vqydrmm19m4wiffz1ic3v7528-ffmpeg-headless-7.1-bin 295755456 /nix/store/0km9wmmxs4aasna3jvx6ksigbi5qlas8-ffmpeg-headless-7.1-bin 296508848
|
(Drafting because this should only land after breaking features on staging are unblocked. |
Atemu
added
the
2.status: wait for branch‐off
|
Wait, how is this breaking? Although |
|
That's a good point, not sure what I was thinking. |
Atemu
removed
the
2.status: wait for branch‐off
This pulls in clang as a build input but as it turns out this actually _reduces_ closure size slightly: /nix/store/0km9wmmxs4aasna3jvx6ksigbi5qlas8-ffmpeg-headless-7.1-bin 296508848 /nix/store/d0dl773klln23daxy1f24xcpbgch184a-ffmpeg-headless-7.1-bin 296402944 Fixes NixOS#344114
|
Added cuda-llvm which surprisingly decreases closure size. I'd like to move ahead with this one now. Any objections to the 60.3965 MiB stand-alone and 6.1656 MiB composite closure size increase? |
wegank
added
the
12.approvals: 1
|
No closure size worries from me. The only thing here that doesn’t seem like obviously a good idea is OpenMPT; Arch and Fedora don’t enable it and it’s unclear to me what their security track record is like. I’ll defer to your judgement on that one, but maybe we could wait and see if anyone really wants it? (I have to wonder how many people use Jellyfin to play tracker modules.) |
wegank
added
the
12.approved-by: package-maintainer
|
I'm not entirely sure what it's useful for because it seems that's for stuff from before I was born but if it's important enough for jellyfin to activate, it probably is for us too. The Wikipedia page also says that it's expected of media players to support this format, so I think it would be the least surprising behaviour if e.g. mpv (using ffmpeg/libav) supported it. |
|
The official mpv binaries don’t support it, and Debian and Fedora don’t ship support, so I’m not sure how widespread that expectation is. They’re quite elaborate sequencer instruction files rather than “audio” files per se; I expect that tracker enthusiasts mostly used specialized players for the task. I worry about exploits like this when turning on obscure formats by default. Looks like they put out a lot of security updates, but at least we seem to be keeping on top of them. |
With NixOS#354952, the regular ffmpeg-headless should now have all the deps required for frigate to make use of hardware acceleration on all hardware. This is progress towards NixOS#271863 This reverts commit 7e33e47. This may be surprising as it's quite soon after 7e33e47 was merged but we decided in NixOS#357717 (comment) to merge it quickly to unbreak master and backport to 24.11. Non-full ffmpeg can be pursued in master using a simple revert which is where we are now.
|
This broke building ffmpeg for musl. Fix is #360234. |
In #354936 we got a nice overview from the upstream jellyfin-ffmpeg maintainer over features that actually matter for real-world use-cases. This enables most of them in ffmpeg-headless by default.
Total diff:
Looking at a more realistic diff with something else in the closure however:
nix-build --expr 'with import ./. { }; buildEnv { name = "foo"; paths = [ ffmpeg-headless pipewire ]; }'I think <9MiB is an acceptable increase.
Things done
nix.conf? (See Nix manual)sandbox = relaxedsandbox = truenix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage./result/bin/)Add a 👍 reaction to pull requests you find important.