michaelni changed the topic of #ffmpeg-devel to: Welcome to the FFmpeg development channel | Questions about using FFmpeg or developing with libav* libs should be asked in #ffmpeg | This channel is publicly logged | FFmpeg 7.1.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
<BtbN> For PRs it can kinda be done with a properly maintained CODEOWNERS file. At least I think getting assigned still sends you an E-Mail, even if not watching
<BtbN> but for an issue, someone has to manually assign you. Or we actually need to employ an LLM, that reads issues, and somehow guesses who might care
<fjlogger> [FFmpeg/FFmpeg] Pull request #20091 opened: avcodec/h264_parser: remove incomplete heuristic (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20091) by B​alling
Compn has quit [Read error: Connection reset by peer]
Compn has joined #ffmpeg-devel
Kimapr_ is now known as Kimapr
Guest96 has quit [Quit: Client closed]
<kasper93> BtbN: I changed lint script to use api, still in python though
<BtbN> kasper93: so did I, see my PR
<BtbN> I pinged you there
<BtbN> I basically made it capable of running both locally (which it couldn't before, it needed GITHUB_* vars), and on Actions
<kasper93> I didn't see, I would comment there.
<kasper93> I think my version is simplier, I removed all the unnecessary crap
<kasper93> also by the looks of commit report we used the same tools
secondcreek has quit [Ping timeout: 252 seconds]
<kasper93> also I made it so it gets only one file, doesn't even checkout repo at all
<BBB> code lint? I hope it's not require to pass before code gets merged?
<BBB> (I hade code linters)
<BBB> hate*
BtbN has quit [Remote host closed the connection]
BtbN has joined #ffmpeg-devel
iive has quit [Quit: They came for me...]
\\Mr_C\\ has joined #ffmpeg-devel
<BBB> oh I see it's not a code linter; it's a git commit message linter
secondcreek has joined #ffmpeg-devel
<BtbN> There is also a code linter and typo squatter
<Lynne> I can't help but think that it will be a MAJOR PIA for specific commits
<Lynne> sure, it may work for 90% of commits, but for the last 10% you may get death threats
<Lynne> and no, stop forcing adding sign-offs when we've never used them
<Lynne> its like requiring formal business attire at our irl meetings in the hopes of looking like we're synergizing important business decisions
<kasper93> I think we can revisit commit msg linter at another time
<kasper93> With proper documentation update, with resoning on some of the restictions. Because right now I agree that some people might get very unhappy that ci is not passing their commits
<kasper93> Though from experience, this kind of check really helps with drive-by contributors. Almost certenly they will do something that doesn't much existing "style" of commit messages
<kasper93> and sure it is not relevant, but reading consistent looking git log is easier to follow along
<kasper93> (ffmepg is pretty good about it already)
<Kimapr> so i made a demuxer for a format that is half something i made up and half a niche demoscene thing (bytebeat). should i submit this to upstream ffmpeg or keep it on my personal soft-fork?
<Kimapr> bytebeats are typically shared as URLs to a web player with data encoded in the URL or short snippets of code, but i made them into a self-contained file format with metadata embedded as a comment at the top of the file so that it's possible to conveniently play them in a media player
jamrial has quit []
<Lynne> you can submit a PR
<Lynne> IMO I think it would be fine, we have various fun demoscene formats like 8088 corruption
<Yalda> I fed ffplay a 50GB block device with unrecognizable partition today which was fun. Video came out!
Traneptora has joined #ffmpeg-devel
<Traneptora> BtbN: can you ban balling from the new tracker? they've already been banned from trac and from IRC
<Traneptora> it's the same guy as Valzpod
<Traneptora> already banned from nearly every multimedia tracker
<kasper93> not banned on trac
<Kimapr> Lynne: by PR do you mean like on the new forgejo instance? the main site says to submit patches to the mailing list, but lurking here i learn that there is also a forgejo instance now
<kasper93> they are the main user
<Lynne> Kimapr: use the forgejo instance, everyone that matters is on there now
<Kimapr> i see .. that is also preferrable to me, as i am not well-versed with email
<Kimapr> though i did set up a personal email server the other day
<Kimapr> should i also move my bugfix for the libopenmpt demuxer to forgejo? i submitted it to the mailing list
mkver has joined #ffmpeg-devel
<Lynne> sure, it would make merging easier
<Lynne> does "RGhA" mean anything to anyone?
<Lynne> as well as "bp16"/"bp64"
<Yalda> RGhA sounds like a fourcc
<Lynne> found them, CVPixelFormatType FOURCC
<Yalda> Ah, Core Video
<Lynne> oh god
<Lynne> if I'm reading this right, apple have integrated some sort of partial debayering in their prores raw decoder
<Lynne> it looks like plain bilinear debayer
<Lynne> actually this looks like the qscale matrix and some decompiler jank
<Lynne> still, I think I've finally found the dequant function
<Lynne> all of them, they apply different dequant for various output formats (rgha/bp16/bp64 (whatever downscaled bayer is?))
<Lynne> with 4 different variants
_whitelogger has joined #ffmpeg-devel
<Traneptora> kasper93: they are banned on trac, they just keep making alt accounts
<Traneptora> whack-a-mole
Warcop has quit [Remote host closed the connection]
andm100 has joined #ffmpeg-devel
<Kimapr> i have another question: what's a decent way to pass purely textual "video" data from my demuxer? bytebeat in principle is an audio-only format, but the commonly used web player displays errors thrown from the code in a neat monospace box which some songs exploited to display animations by throwing a specially-crafted error once in N samples
<Yalda> subtitle stream ?
<Yalda> sounds like a plain old timed text
<Yalda> oh animations exploited it...
<Kimapr> subtitles seem like a nice way to do this (though this is definitely an unconventional use of them), as i won't have to be responsible for turning text into pixels then
<Yalda> yes, i think its still timed text falling into the realm of subtitles
<Yalda> may not be visually ideal without some ASS markup to force monospace and alignment, but it sounds like the easiest path
<Kimapr> now the problem is figuring out how to make a subtitle stream (idk how lol)
<Kimapr> for audio stream i was lucky to know of a demuxer that happened to have the same exact structure as i would want mine to be, but there's no examples for subtitles that i can comprehend
<Yalda> it should not be too bad. here is some boilerplate to create the actual stream itself: https://paste.debian.net/hidden/7870407f/
<Yalda> here is example of AVI demuxer feeding text packets from old school embedded SRT/ASS (notice GAB2 function above but probably irrelevant for you): https://code.ffmpeg.org/FFmpeg/FFmpeg/src/commit/master/libavformat/avidec.c#L1194
<Lynne> Kimapr: black frame + subtitles
<Yalda> yea black frame is good idea
<Yalda> just know/plan for that internal text-based subtitle format is ass
<Yalda> (advanced substation alpha)
andm100 has quit [Quit: Client closed]
TheVibeCoder has joined #ffmpeg-devel
_whitelogger has joined #ffmpeg-devel
derpydoo has joined #ffmpeg-devel
mkver has joined #ffmpeg-devel
<fflogger> [editedticket] StevenAllanSpielber: Ticket #11653 ([avformat] -c copy does not copy track names) updated https://trac.ffmpeg.org/ticket/11653#comment:12
kepstin has quit [Remote host closed the connection]
kepstin has joined #ffmpeg-devel
kepstin has quit [Remote host closed the connection]
kepstin has joined #ffmpeg-devel
<fflogger> [editedticket] rgr2: Ticket #11683 ([avcodec] "libx265" generated videos abnormal frames parsing?) updated https://trac.ffmpeg.org/ticket/11683#comment:4
<fflogger> [editedticket] rgr2: Ticket #11683 ([avcodec] "libx265" generated videos abnormal frames parsing?) updated https://trac.ffmpeg.org/ticket/11683#comment:5
derpydoo has quit [Ping timeout: 276 seconds]
<BtbN> Traneptora: I could, but I don't really wanna get into said whack-a-mole. There is no good way to ban someone from just making a new account over and over.
<BtbN> Unless there's actually extreme transgressions, I think it might be better to at least know for sure who's on the other side of the screen.
\\Mr_C\\ has quit [Remote host closed the connection]
<fjlogger> [FFmpeg/FFmpeg] Pull request #20092 opened: swscale/swscale_unscaled: use 8 line alignment for planarCopyWrapper with dithering (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20092) by c​us
moonmuaaz has joined #ffmpeg-devel
jamrial has joined #ffmpeg-devel
moonmuaaz has quit [Quit: Client closed]
<fjlogger> [FFmpeg/FFmpeg] Pull request #20069 merged: forgejo/workflows: add pre-commit based linting (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20069) by B​tbN
mkver has quit [Ping timeout: 265 seconds]
<fjlogger> [FFmpeg/FFmpeg] Pull request #20093 opened: fix a bunch of uninitilized uses (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20093) by m​ichaelni
<fjlogger> [FFmpeg/FFmpeg] Pull request #20094 opened: tests/fate/hlsenc: add testcase of hls fragment mp4 named cmfa (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20094) by s​tevenliu
mkver has joined #ffmpeg-devel
minimal has joined #ffmpeg-devel
<kasper93> I'm also not a big fan of auto-merging, but maybe it's for the best to automate things, especially for not a big patches
<kasper93> the banner `michaeln i scheduled this pull request to auto merge when all checks succeed 16 minutes ago.` is big enough to see that it's been scheduled
<kasper93> Can I configure forgejo "quote and reply" to not include @XYZ wrote in <link> part?
<fjlogger> [FFmpeg/FFmpeg] Pull request #20093 merged: fix a bunch of uninitilized uses (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20093) by m​ichaelni
s55 has quit [Quit: Bye]
s55 has joined #ffmpeg-devel
mkver has quit [Ping timeout: 248 seconds]
mkver has joined #ffmpeg-devel
<BtbN> kasper93: I merged the two scripts we wrote and pushed the result to your PR
<kasper93> thanks, will look later
<kasper93> forgejo doesn't have ability to filter notifications by mentions
<kasper93> that's big deal for me, because some notifications are more important than others, especially if you watch high trafic repo
<BtbN> I mean, if you unwatch the FFmpeg repo, you only get mentions
<kasper93> and lose log of what's happening in ffmpeg repo
<BtbN> How does this work on GitHub or other sites? Isn't it the exact same problem there?
<BtbN> An idea I'd have it marking the mails as important it they directly mention you
<kasper93> On github you have filters https://0x0.st/s/WcrxPadYbX8HPTPSFmR03w/870_.png (and even can add custom ones)
<BtbN> kasper93: can you assign me to https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20080 and then mention me in a comment? Want to check something.
<kasper93> done
<BtbN> nope, nothing in the Mail-Headers to distinguish those
<BtbN> Though making a sieve-filter on "mentioned you" and "assigned you" would work
<beastd> i forgot on who mentioned it already (wbs iirc) but ideally we would make forgejo post the important stuff to the ffmpeg-devel ML and forgejo would only email individuals about mentions and for issues or PRs they already were part of.
<BtbN> Yes, but someone needs to write such a feature
<kasper93> BtbN: For emails, I can do it myself. For javascript Play-Doh website, seems to be not so flexible
<BtbN> Well, I've now installed a sieve filter for "mentioned you" and "assigned you"
<BtbN> that should get me most pings to a seperate folder
<Lynne> please don't, I don't want double notifications from both ffmpeg-devel and code.ffmpeg.org
<BtbN> We absolutely need to at least forward new PRs to the ML
<BtbN> But not much else imo
<BtbN> Also, a few days ago didn't you say we should just drown the ML in notifications? :D
<BtbN> kasper93: one downside of the simplified script is that it now only works on PRs, not pushes
<BtbN> though I disabled it for pushes anyway, cause at that point it's too late and would just be red forever
<kasper93> I know, that was explicit
<kasper93> there is no point to checking pushes, cant fix those anyway
<BtbN> Well, it was still enabled for pushes
<kasper93> I know I forgot to disable it
<kasper93> I was thinking about it at some point, but never did it ;p
<fjlogger> [FFmpeg/FFmpeg] Pull request #20096 opened: libopenmpt: fix seeking weirdness (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20096) by k​imapr
kurosu has joined #ffmpeg-devel
mkver has quit [Ping timeout: 265 seconds]
<fjlogger> [FFmpeg/FFmpeg] Pull request #20097 opened: Fix more warnings (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20097) by k​asper93
<fjlogger> [FFmpeg/FFmpeg] Pull request #20076 merged: avformat/mov: free AVBPrint on error when parsing infe boxes (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20076) by j​amrial
<fjlogger> [FFmpeg/FFmpeg] Pull request #20088 merged: avformat/mov: move AVC-Intra extradata generation to earlier in the stsd parsing process (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20088) by j​amrial
<BtbN> The whole signed commits thing confuses me
<BtbN> how did they work with E-Mail based patches?
<BtbN> You GPG signed the E-Mail, and it somehow carried into the commit?
<jamrial> no, you sign the commit
<BtbN> But you can only do that if you push it yourself then
<jamrial> yeah
<BtbN> So that's the same with Forgejo still then
<jamrial> or pull the commit from some public repo
<BtbN> push yourself, and mark PR as manually merged
<BtbN> There is some support for GPG in Forgejo, but I doubt you can or should upload your private key
<jamrial> yes, that's how to workaround it, but still more work than a "rebase and merge" click :p
<BtbN> Like, you can add your GPG pubkey in your account
<BtbN> If you want signed commits, I see no other way to do that. You need to rebase yourself
<jamrial> people that care about signing can do it, so not a problem really
<BtbN> I could enable the "FF Only" merge option
<BtbN> so if it's available, you can at least use it and the signature survives
<BtbN> I guess a proper merge-commit would also retain it?
<BtbN> But we never used those before ever
<jamrial> yes, but i don't like the idea of doing branch merges
<jamrial> it makes bisecting a pain
<BtbN> If you force-push upgrade the PR branch first, before manually merging to master, it'll also auto-detect the PR as merged and close it
<BtbN> needs to be the same commit hash for that to work
<kasper93> yes, if forgejo don't rebase internally, signature will survive
<kasper93> signed tags / releases is important, for each commit it's less relevant
<kasper93> unless every single commit require signature... it's just easy to bypass it. Of course there is impersonation aspect to it.
fjlogger has quit [Remote host closed the connection]
fjlogger has joined #ffmpeg-devel
<fjlogger> [FFmpeg/FFmpeg] Pull request #20098 opened: avformat/mov: set primary extradata based on the first Sample only if it's not already in place (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20098) by j​amrial
fjlogger has quit [Remote host closed the connection]
fjlogger has joined #ffmpeg-devel
<fjlogger> [FFmpeg/FFmpeg] New comment on issue #20081 avformat/avio: clarify that AVSEEK_FORCE has no effect (https://code.ffmpeg.org/api/v1/repos/FFmpeg/FFmpeg/issues/20081) by B​alling
<fjlogger> [FFmpeg/FFmpeg] Pull request #20098 opened: avformat/mov: set primary extradata based on the first Sample only if it's not already in place (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20098) by j​amrial
<kasper93> BtbN: rebased lint branch, with small adjstemnts to script, mostly types, and assert.
<kasper93> I don't like `os.environ["GITHUB_REF"].split("/")[2]` but it will work untill it doesn't, probably not a big deal
mkver has joined #ffmpeg-devel
fjlogger has quit [Remote host closed the connection]
<BtbN> That's the standard way of getting the PR numbers in A LOT of workflows
<BtbN> they can't break that
fjlogger has joined #ffmpeg-devel
<fjlogger> [FFmpeg/FFmpeg] Pull request #20098 opened: avformat/mov: set primary extradata based on the first Sample only if it's not already in place (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20098) by j​amrial
<BtbN> last time this one re-plays. It should have sent an E-Mail of the PR to ffmpeg-devel
<fflogger> [editedticket] waider: Ticket #11606 ([avcodec] avcodec/dvbsubenc: Broken dvb_encode_rle8) updated https://trac.ffmpeg.org/ticket/11606#comment:1
<BtbN> There is some weird stuff going on with E-Mail formating. That ">" in front of the From line is not there anywhere.
<BtbN> But git am does not seem to mind
<BtbN> It is using the wrong From: though, despite the correct one being in the mail. Guess if you want it to get it right you gotta download the linked mbox format patch
<fflogger> [editedticket] waider: Ticket #2909 ([avcodec] Make dvbsubs more compliant among stb's) updated https://trac.ffmpeg.org/ticket/2909#comment:9
Kimapr_ has joined #ffmpeg-devel
Kimapr has quit [Remote host closed the connection]
<fjlogger> [FFmpeg/FFmpeg] New comment on issue #20080 forgejo/lint-commit-msg: add script for commit message linting (https://code.ffmpeg.org/api/v1/repos/FFmpeg/FFmpeg/issues/20080) by k​asper93
<fjlogger> [FFmpeg/FFmpeg] Pull request #20099 opened: avformat/webvttdec,webvttenc: update documentation (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20099) by G​ewoonLeon
<fjlogger> [FFmpeg/FFmpeg] Pull request #20078 merged: avcodec/srtenc: don't produce SRT files with mixed line endings (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20078) by k​asper93
<fflogger> [editedticket] pbk: Ticket #11606 ([avcodec] avcodec/dvbsubenc: Broken dvb_encode_rle8) updated https://trac.ffmpeg.org/ticket/11606#comment:2
<kasper93> self-approve are not allowed anymore? What should be done about the patches that are not interesting for others to review?
<kasper93> s/are/is/
<kierank> self approve was never allowed
<Yalda> "rebase then FF" button works or I should merge manually still? small change https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20099#issuecomment-1365
<kasper93> kierank: how so?
<kasper93> Documentation clearly states
<kasper93> > If no one answers within a reasonable time-frame (12h for build failures and security fixes, 3 days small changes, 1 week for big patches) then commit your patch if you think it is OK.
<kierank> when was that added
<kasper93> I don't know, but for years, patches are merged like that.
<kasper93> it's just redirect to main page
<kierank> it's HSTS
<kierank> you can't proceed
mkver has quit [Ping timeout: 276 seconds]
<kasper93> where do you want to proceed to?
<kierank> it's impossible to go ahead
<kierank> 18:53:47 <kasper93> unresponsive maintainer policy added here https://github.com/FFmpeg/FFmpeg/commit/9fcee5853de3642dfa0798e39a506219db98dd4e
<kierank> 20 years ago when the project was completely different
<kierank> there was literally a fork over that policy lol
<kasper93> So you say it's not longer applicable today and no one follows this rule?
<TheVibeCoder> mini is the only rule here
<kierank> lol
Kimapr_ has quit [Remote host closed the connection]
Kimapr_ has joined #ffmpeg-devel
<Yalda> i just don't want to break anything by merging that way, sorry for not being in the loop
GewoonLeon has joined #ffmpeg-devel
<GewoonLeon> Yalda: I can rebase the branch in my fork if that makes things easier
<jamrial> kierank: if you want the web interface for the non forgejo repo, it's https://git.ffmpeg.org/gitweb/ffmpeg.git
<jamrial> GewoonLeon: if you want your commit to remain signed, then do that, yes
<kierank> jamrial: which one is the true repo
<jamrial> i think they are in sync
<Yalda> GewoonLoon; that would probably be best, although this is also a learning exercise for me to merge with foregjo
<jamrial> if you push to source.ffmpeg.org, bit will be reflected in both
<fjlogger> [FFmpeg/FFmpeg] New comment on issue #20095 The video contrast is reduced when using SDL playback (https://code.ffmpeg.org/api/v1/repos/FFmpeg/FFmpeg/issues/20095) by B​alling
<kierank> Then source.ffmpeg.org should have a valid certificate, no?
<TheVibeCoder> Balling is alive again?
<Yalda> ^ also issue comment IRC bot messages link to a JSON output
<Yalda> TheVibeCoder: not only alive but making PRs
<GewoonLeon> Yalda: i pushed a rebase
<fflogger> [editedticket] Balling: Ticket #11606 ([avcodec] avcodec/dvbsubenc: Broken dvb_encode_rle8) updated https://trac.ffmpeg.org/ticket/11606#comment:3
<jamrial> kierank: you're not supposed to access that domain from a web broser, so not really
<jamrial> the repo at g.v.o is outdated because the hooks are rejecting a commit that was pushed after the switch
<TheVibeCoder> tell kierank to not visit avcodec.org
<fjlogger> [FFmpeg/FFmpeg] Pull request #20101 opened: Small assortment of configure improvements for MSVC (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20101) by k​asper93
fjlogger has quit [Remote host closed the connection]
fjlogger has joined #ffmpeg-devel
<fjlogger> [FFmpeg/FFmpeg] Pull request #20102 opened: avformat/mov: add support for multiple decryption keys (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20102) by k​asper93
rvalue- has joined #ffmpeg-devel
rvalue has quit [Ping timeout: 248 seconds]
Guest73 has joined #ffmpeg-devel
rvalue- is now known as rvalue
Kimapr_ is now known as Kimapr
<fjlogger> [FFmpeg/FFmpeg] Pull request #20103 opened: forgejo/workflows: run fate with assert level 2 (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20103) by k​asper93
<jamrial> BtbN: why is CI building ffmpeg with --enable-nonfree? it's not including any relevant external dependency
minimal has quit [Quit: Leaving]
<kasper93> I've added both gpl and nonfree. Is it an issue?
<jamrial> gpl is fine, nonfree is a noop in there
<jamrial> it's only for external dependencies that have libraries that make the resulting binary non distributable
<kasper93> it doesn't harm, no?
<BtbN> I think with enable-nonfree you get stuff that otherwise get disabled cause not gpl compatible, right?
<BtbN> so if you want to enable _everything_, you need it
<BtbN> and we're not redistributing the binaries there after all, so it at the very least doesn't hurt
<jamrial> BtbN: yes, but everything affected by it is not autodectected, and we're not enabling them either
<jamrial> anyway, it's harmless i guess
<kasper93> We could buff build image with some deps
<kierank> TheVibeCoder: avcodec has an actual certificate
<kasper93> it's the same cert and is invalid
<fjlogger> [FFmpeg/FFmpeg] Pull request #20090 merged: avcodec/dvbsub{enc,dec}: fix rle256 handling (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20090) by c​us
<kasper93> BtbN: this shouldn't happen with labels ^
<BtbN> What shouldn't happen?
<kasper93> ah, it's because forgejo got confused about pr commit range after merge
<kasper93> shrug
<BtbN> The workflow should probably only run on open PRs
<BtbN> cause people might re-use branches and stuff
<BtbN> But yeah, it somehow got confused what the merge base is there
<kasper93> reusing after close/merge should work ok
<kasper93> but maybe doing it once at open is sane in general
fjlogger has quit [Remote host closed the connection]
<BtbN> Wenn you can extend the scope of an existing PR
fjlogger has joined #ffmpeg-devel
piolad has joined #ffmpeg-devel
TheVibeCoder has quit [Ping timeout: 240 seconds]
<fflogger> [editedticket] cus: Ticket #11606 ([avcodec] avcodec/dvbsubenc: Broken dvb_encode_rle8) updated https://trac.ffmpeg.org/ticket/11606#comment:4
<kasper93> if anyone is bored take a look at https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20025 thanks
<fjlogger> [FFmpeg/FFmpeg] Pull request #20080 merged: forgejo/lint-commit-msg: add script for commit message linting (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20080) by B​tbN
<kasper93> BtbN: if forgejo will sign commits on rebase, it will rewrite commiter field of commit, no?
<BtbN> No idea, but why would it?
GewoonLeon has quit [Ping timeout: 260 seconds]
<BtbN> The way I'm setting it up is that it'll only sign when the commits has a valid signature before, i.e. machting the key configured by the user
<kasper93> I though commiter must much signer, but maybe that's not the case, dunno
<BtbN> I have no experience with that stuff
<BtbN> "Forgejo also needs to be told who the committer of the commit is, which requires a name and email and should be configured as followed:"
<BtbN> I guess it will
funman has quit [Changing host]
funman has joined #ffmpeg-devel
<Yalda> GewoonLeon: sorry; I had stepped away. it already went stale. I can rebase and merge via forgejo button if you're good with that
<kasper93> BtbN: looks like you can trust the signature and re-sign commit.
<BtbN> That's what collaborator mode seems to do
<kasper93> I remembered correctly, that in GitHub-like signing, it would change commiter
<BtbN> Yeah, comment there seems to indicate that
<kasper93> I think it's fine, but it should be documented somewhere in docs, that such (re)signing happens.
Guest73 has quit [Quit: Client closed]
piolad has quit [Quit: Client closed]
<Yalda> oo BtbN did you just add it by chance? I got a 502 then refreshed, and now instead of telling me it won't sign, it implies to sign as long as I add my public key
<BtbN> Yeah, just restarted it to enable the settings
<Yalda> neat
<Yalda> gonna try the rebase/merge
<fjlogger> [FFmpeg/FFmpeg] Pull request #20099 merged: avformat/webvttdec,webvttenc: update documentation (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20099) by M​arth64
<Yalda> easy enough
<Yalda> i added the key and set up mfa before doing so
<BtbN> It will only do it though if 1. you have 2FA set up and 2. the commits are already signed with a key from a collaborator of the repo
<wbs> is 2fa relevant if authenticating through github too?
<BtbN> yes
<wbs> ok
<wbs> will set it up tomorrow then
<BtbN> if you log in via github you will be prompted for a OTP after auth via github
<Yalda> yep added 2fa and locked down. it said, before i clicked merge, it would sign it with a key at /some/path/i_forgot
<Yalda> and i see a lock by the commit hash so it appears to have worked
<another|> l
<Traneptora> kierank: source.ffmpeg.org is an alias for git.ffmpeg.org
<Traneptora> just use git.ffmpeg.org and the cert will be fine
<Traneptora> by an alias I mean a CNAME
<kierank> somebody fixed the certificate
<kierank> Sun, 03 Aug 2025 18:17:02 GMT
<kierank> just now
<Traneptora> so like 13 minutes ago lol
<Traneptora> good timing
<Traneptora> oh wait 18:00 UTC nvm
scat117 has quit [Ping timeout: 252 seconds]
<kierank> 20 minutes after i posted on irc
<Yalda> kasper93: LRC sub id lyric file expressed as subtitles?
<BtbN> The ACME setup on ffmpeg.org is hyper weird
<Yalda> s/id/is
<BtbN> I mess it up almost every time
scat117 has joined #ffmpeg-devel
<BtbN> I need to migrate that to dehydrated at some point
<BtbN> it's some shell script that appears to be self-written bei reimar and lives in his homedir
<BtbN> editing the domain list does not cut it. You need to generate a new csr, which you can only do as root, and then need to chown it...
<kasper93> Yalda: LRC is a lyric file, but it's no different than subtitle in practice, not sure if that's what you ask
<Yalda> kasper93: Yes, just wanted to confirm the use case is lyric. thank you
indecisiveturtle has joined #ffmpeg-devel
<Yalda> it looks good kasper93, i will do a quick test and approve
<Yalda> makes sense
<kasper93> thanks
<BtbN> Yalda: I see no locks?
<Yalda> its directly to the right of the commit message
<Yalda> center of page
<BtbN> but not on master
<BtbN> the lock there on the PR would have had to be green in order to get carried over
<Yalda> i get it now, that probably ties in to the fact that when I click the lock from the MR it tells me "Signed by untrusted user:", which matches your statement earlier of the commit being signed with a collaborator's key
<Yalda> needing to be signed with*
<BtbN> I'm not sure how it would have went if their key would have been added to their account
<BtbN> Cause by getting their PR merges they also are a collaborator
<BtbN> so maybe a green lock there, and you having 2FA, would have been enough
<Yalda> i think the user has another MR if that one gets merged down the road it could be a good case study
<BtbN> We should tell them to add their signing key to their account
<Yalda> it is this one. https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20062 Traneptora is reviewing. once/if blessed makes sense to request the user try that and see what happens
<BtbN> so maybe the user is just not trusted, hm
<Yalda> pross: I can re-test this one on M3/M1 and approve later as well https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20089
<Yalda> though I only did g728 last time it looks straightforward
<kasper93> I think "collaborator" mode is misleading, I think it means "people with push access"
<BtbN> Yeah
<kasper93> else it's not trusted, even if verified.
<BtbN> Would changing the model to committer help?
<kasper93> I think it would yes
<BtbN> This collaborator model seems quite limited
<BtbN> basically nothing from an external person could ever be signed
<kasper93> Though then commiter is rewritten by forgejo. I think they idea is that you don't want to put trust to commits with foregin commiter.
<BtbN> There's collaboratorcommitter, which seems to demand _both_
<Traneptora> Yalda: should I ping you when I'm ready to approve the change before merge?
<Traneptora> so you can test whatever you want to test
DauntlessOne498 has quit [Read error: Connection reset by peer]
DauntlessOne498 has joined #ffmpeg-devel
<Yalda> Traneptora: I will wait since BtbN looks like might be considering changing config ^ I think the goal was just to chime in to ask the user to add a signing key to their account so we can tease out this commit signing workflow is all
<Yalda> ty
<BtbN> I'm not sure if we should change config there
<BtbN> I'll leave it as it is for now, it at least enabled michael to keep signed commits via PRs :D
Kimapr has quit [Remote host closed the connection]
Kimapr_ has joined #ffmpeg-devel
<Yalda> cool. yeah I see his are nice and green. then Traneptora yes sure whenever you approve just let me know and I'm happy to add a comment there - doesn't make sense now in the middle of the review. I will tag you BtbN
<Yalda> so we are asking them to add their public key and thats it ?
<Yalda> that they signed the commit with
<fjlogger> [FFmpeg/FFmpeg] Pull request #20058 merged: vf_bwdif: add AVX512 implementation (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20058) by k​ierank
<BtbN> I think it's already added
<BtbN> otherwise it wouldn't have the yellow lock
<Yalda> you are probably right, it says signed by untrusted user
<kasper93> you can check thier keys in the profile
<BtbN> git and forgejo actually recommend against gpg keys, and to use ssh keys instead
<jamrial> i don't think anyone is signing with gpg
<kasper93> it shows those commits as gpg signed though
<jamrial> right, misread
<BtbN> jamrial: you can push to forgejo btw. if you prefer, it'll then also show up as from the right user :D
<kierank> jamrial: can we make avx512 include avx512bw
<kierank> that was certainly my intention
<kierank> I think skylake-sp in practical terms is useless for us
<kierank> actually the header is wrong
<kierank> #if HAVE_AVX512 /* F, CD, BW, DQ, VL */
<pross> Yalda: ok thanks
<jamrial> is it? maybe i was wrong and it does cover those, with icl including a check for popcntdq/ifma/vbmi
<jamrial> i can't say from looking at x86/cpu.c because it just uses magic numbers to check flags
<kierank> grok says the magic numbers are f, d1, bw
<BtbN> Quick reboot of the Forgejo server for kernel update
fjlogger has quit [Remote host closed the connection]
fjlogger has joined #ffmpeg-devel
fjlogger has quit [Remote host closed the connection]
fjlogger has joined #ffmpeg-devel
<kierank> jamrial: anyway I'm not sure how to tell everyone they should be using avx512icl
<jamrial> we could make sure AVX512 implies vl/bw/dq (so skylike-sp, cannon lake, etc) and should only be used for ymm or lower
<jamrial> any other extension or zmm -> use ICL
<kierank> that's disappointing
<kierank> Intel(R) Xeon(R) Gold 6238R CPU @ 2.20GHz
<kierank> maybe on xeon Silver it was faster a few years ago
mkver has joined #ffmpeg-devel
<kasper93> #20096 is another untrusted user
Kimapr_ is now known as Kimapr
<Kimapr> hii that's me yes
<BBB> kierank: try amd zen5
<BBB> but yes sometimes avx512 just doesn't help, it's ok
<kierank> BBB: this is avx512 non icl
<kierank> avx512 on ymm
<kierank> Which was faster on the cpu at time of commit