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 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
<Daemon404>
hey haasn i just wanted to point out you msised a possible outcome in your email: rather than fork, this time people just leave. cheers.
Daemon404 has left #ffmpeg-devel [#ffmpeg-devel]
Daemon404 has joined #ffmpeg-devel
<Daemon404>
i also have to say jamrial you should be ashamed of yourself, admonishing naughty words from one side while the other gaslights and bullies/
Daemon404 has left #ffmpeg-devel [#ffmpeg-devel]
thilo has quit [Ping timeout: 272 seconds]
thilo has joined #ffmpeg-devel
rodgort has quit [Quit: Leaving]
rodgort has joined #ffmpeg-devel
<another|>
FTR, I think jamrial did the right thing admonishing the tone.
<jamrial>
dunno, Daemon404 is not the only one that doesn't think so
^Neo has quit [Ping timeout: 260 seconds]
<another|>
Welp, that's my 2 eurocents
<Lynne>
haasn: I think it would be fine to remove yuvj formats
<Compn>
remi said near the same thing to BBB the other day in this channel
<Compn>
bully + ashamed
<another|>
I noticed
jamrial has quit []
<fflogger>
[editedticket] Balling: Ticket #6531 ([avformat] MPEG-TS encoder may stop emitting SDT/PAT/PMT after DTS jump or rollover) updated https://trac.ffmpeg.org/ticket/6531#comment:1
HarshK23 has quit [Quit: Connection closed for inactivity]
Martchus_ has joined #ffmpeg-devel
Martchus has quit [Ping timeout: 248 seconds]
bencoh has quit [Ping timeout: 244 seconds]
bencoh has joined #ffmpeg-devel
Guest70 has joined #ffmpeg-devel
Guest70 has quit [Client Quit]
^Neo has joined #ffmpeg-devel
^Neo has quit [Changing host]
^Neo has joined #ffmpeg-devel
^Neo has quit [Ping timeout: 272 seconds]
System_Error has quit [Ping timeout: 264 seconds]
<j-b>
good morning
mkver has joined #ffmpeg-devel
<thardin>
morning
___nick___ has joined #ffmpeg-devel
<beastd>
moin :)
<beastd>
just FTR as it seems all things are getting mixed up again. BtbN already mentioned it as well. AFAICT there is no opposition from me or michaelni or who else gets accused of it to try and use things like web UIs for PRs and have CI running for them
<beastd>
it's "only" about what tech to use, make a poc, and if it works do the necessary implementation and migtiation work
<beastd>
however i do not agree this shortened "sth sth gitlab" will fix all things simplification. it will not solve all problems and it will bring new problems as well(it's natural).
<JEEB>
yes, for self-hosted stuff I've even seen a comment from michaelni on IRC recently with mention on him being OK with "what the community wants" being set up, so I did not consider him being against the system itself (esp. since at the base of it there's a git repository after all). I think the largest opposition I've seen was regarding utilizing already existing infra at VideoLAN or so.
<beastd>
i would also say it's definitely not to replacing mailing lists. it's complementary. and the question on what and how those two techs/processes should be linked.
<JEEB>
and yes, naturally either git forge will not fix AllTheProblems.
witchymary has quit [Remote host closed the connection]
<beastd>
short addition to mailing lists. basically it boils down to the argument elenril has been presenting here before. complex discussions are complex and they are mostly reflected worse in usual forum/comments/issue trackers. email has the advantage of decentralized storage and having technically everything to represent comlex discussion threads better. not all MUAs leverage it enough tho.
<beastd>
for code contributions comments on code in PR web UIs are usually better and also iterations and integrations in CI work better there
cosimone has joined #ffmpeg-devel
cosimone has quit [Remote host closed the connection]
rvalue has quit [Read error: Connection reset by peer]
rvalue has joined #ffmpeg-devel
DauntlessOne4 has quit [Ping timeout: 244 seconds]
<ePirat>
beastd, I dont think anyone expected that GitLab or Forgejo would replace the ML completely but "just" for code submission, review and possibly bug tracking.
elvis_a_presley has quit [Quit: smoke-bomb ; grapple-hook]
elvis_a_presley has joined #ffmpeg-devel
<BtbN>
I could see it replacing it in the VERY long run. But that's after a decade or longer, when everyone just got used to it
<BtbN>
ML ain't going away until nobody is using it anymore
<BtbN>
Seperating discussions from code review alone would be nice though
<Marth64>
my only practical issue with ML is that in certain situations it would be nice to upload diagrams and have rich text
<Marth64>
when trying to explain something complicated every once in a while
<kierank>
git-send email is a nightmare, inline code review is difficult
<Marth64>
yeah for code reviews its a nightmare
<BtbN>
inline code review is fine imo if send-email was used
<BtbN>
when a patch file was attached it's annoying
<Marth64>
its too much process overhead
<Marth64>
imo
<Marth64>
for patch submission and stuff
<BtbN>
Is it? It feels lamost the same as reviewing code on Github and friendy
<BtbN>
The most annoying part about E-Mail is when _reading_ a review of a long patch
<Marth64>
after getting used to it its ok. but i look back and think about others trying it and being like "nope"
<BtbN>
There is no easy way to jump to the next comment
<nevcairiel>
the problem with email code review isnt the first review, its following up with changes through several cycles where it gets really bad
<Marth64>
if patchwork didn't exist then it would be totally unusable. at least that gives syntax coloring
<BtbN>
Often times gotta slowly scroll over hundreds of lines of nothing, just to make sure to not miss a comment
<Marth64>
+1 on iterations
<Marth64>
i just wish i could attach a diagram sometimes. i guess a external link would work for that but then it loses archival value if external link goes down
<BtbN>
you could straight up attach a picture
<Marth64>
is it allowed?
<BtbN>
I've never seen anyone do it, but the ML does not block attachments
<Marth64>
from a rules perspective
<Marth64>
hm
<BtbN>
Why would attaching an explanatory picture be forbidden?
<j-b>
ML is completely obsolete for code review
<j-b>
- git send mail is a nightmare to set up
zane has joined #ffmpeg-devel
<Marth64>
BtbN: I've never seen anyone do it so I thought it was a no-go
<j-b>
- almost impossible from a quick view to know which MR have been merged or not
<BtbN>
It's more like nobody takes the time to make them :D
<BtbN>
But if you want to, nobody is stopping you
<Marth64>
j-b: yes agreed. discussion OK. code review not as OK.
<j-b>
- very hard to see which part of the patch is under discussion and which threads are open
<j-b>
- diagram are a mess (vs plantuml, mermaid)
<j-b>
- set of large commits is beyond hard to manage, and requires resending the whole set (like v12 of 27 patches, makes 300 emails)
<j-b>
- CI+fate is often broken on weird architectures
<j-b>
etc...
<BtbN>
The fate CI being broken is mostly because the runners are externally managed by volunteers, and they all fell into disrepair
<j-b>
lol no.
<BtbN>
it used to work quite okay, but often Patchwork would lose track of the state of a series
<j-b>
It's because if you send a patchset to the mailing lists it does not tell you that you break fate
<j-b>
a contrario from any other sane system of CI
<BtbN>
It does. It sends you a mail if your patch has a CI failure
<j-b>
Unless it is broken, like most of the time.
<BtbN>
If there are no CI runners online, it won't, indeed
<j-b>
and it works SOOO well SOOO often, that very often Michael sends an email afterwards saysing "this commit broke fate"
<j-b>
Like how the hell does this still happens in 2024?
<BtbN>
Michael runs a lore more than just fate
<j-b>
nice argument.
<BtbN>
I think he has a whole private test suite on top of fate
<j-b>
Like totally backward one.
<BtbN>
which is... a bit odd really
<j-b>
You broke the private CI...
<j-b>
That you know nothing about
<BtbN>
iirc the argument for that was that the needed samples can't be shared?
<j-b>
Oh, and I'm not even talking about fuzzer integration, code coverage
<j-b>
but just the first few arguements should be enough to explain which ML are obsolete for modern projects
<BtbN>
Stuff like Patchwork only works well if everyone strictly follows the workflow it expects
<BtbN>
which is utopian
<j-b>
"Oh, but no, fate+private_fate+perl scripts + patchwork + trac + ML + perl scripts parsing the ML" are totally fine.
<j-b>
exactly
<j-b>
So, patchwork is a dead solution.
<michaelni>
<Marth64> if patchwork ... gives syntax coloring <--- my mutt shows fully colored patches
<BtbN>
Patchwork is nice to apply patches via curl + git am :D
<j-b>
Oh, and how many times "your MUA/MTA broke line endings, we cannot apply your patches"
<michaelni>
tools/unwrap-diff if its a problem for the sender to fix
<j-b>
you don
<j-b>
you don't understand
<j-b>
"<@BtbN> Stuff like xx only works well if everyone strictly follows the workflow it expects. Which is utopian."
<j-b>
A set of uncoupled complex tools are not usable for normal developers in 2025.
<j-b>
"It works for me" is just Stockholm Syndrom, because you are used to that.
<Marth64>
I don't get emails from patchwork. I believe its because of Gmail IIRC so I wasn't able to get an account to work. (I just want to close old tasks there since I treat it like a queue)
<BtbN>
Yeah, gmail flat out rejects mails from patchwork
<michaelni>
j-b, i just said how to fix the wraping and that diff coloring is easy to do in mutt. I didnt say anything else
<kierank>
Marth64: you can't just make up things like "DVB 0502"
<kierank>
closed captions are unrelated to DVB
<Marth64>
Hi kierank. It's a closed caption coding transmitted specifically by some DVB-S providers. The intent wasn't to say its part of DVB but that its a coding specifically used by the providers using 0x0502 as a magic number
<Marth64>
I'm not attached to it, if it needs to change thats fine. but its a real format and that is its only use
<kierank>
it's a real format yes but the name is completely made up
<Marth64>
I don't know what else to call it besides the name of the provider. I don't think its standard
<Marth64>
if I knew a standard name for it, of course
<kierank>
there is no standard, exactly so it shouldn't be named after a standard
<Marth64>
It's no biggie. I can write a patch to change it.
<Marth64>
kierank: since you are here. do you know what is the practical purpose of padding byte pairs in CC?
<kierank>
to maintain cbr
<Marth64>
context: I am working on a CBS to split and assemble between the different codings
<kierank>
and to make sure captions are not sparse
<Marth64>
i see
<j-b>
Marth64: congratulations
<Marth64>
thank you. it actually works between DVD/A53 formats
<Marth64>
both ways
<kierank>
imo the behaviour of ffmpeg auto selecting field 2 when there are no captions just nulls is weird
<Marth64>
it is. but all of CC in ffmpeg is weird
<Marth64>
i'm working on a proposal deck and the CBS to propose what it could be
<Marth64>
oh. the actual CC vote, not closed captions
<Marth64>
thank you!
<j-b>
lol
<BtbN>
Marth64: just noticed you were missing +o still. Do you also want an ffmpeg/developer cloak?
<Marth64>
thank you, BtbN. I think I have a cloak but that one sounds cooler, I'll take if its not a lot of hassle
<BtbN>
not at all
<BBB>
(for those wondering: I did indeed forget to CC root, but forwarded it to root right after)
<Marth64>
I've never tried mermaid, only plant, but I'll check it out
<Marth64>
I made this for my deck and documentation
<BtbN>
BBB: I did not receive a mail to root, but saw it anyway and updated the cc@ alias
Marth64 has quit [Changing host]
Marth64 has joined #ffmpeg-devel
<BBB>
ok thank you
<BBB>
should we forward ongoing business to the new CC members so they can continue these discussions?
<BtbN>
mail to root just arrived. Guess it got graylisted or something
<BtbN>
Forwarding the latest discussions that are still ongoing to the new CC seems like a good idea to me
<jamrial>
maybe wait until after new year's
zane has quit [Quit: WeeChat 4.4.4]
Everything has joined #ffmpeg-devel
<kierank>
The irony was it was quite peaceful from the 16th to 19th with everyone preparing for Xmas
<kierank>
Then a certain email was sent
steven-netint has quit [Ping timeout: 252 seconds]
putacho has quit [Ping timeout: 244 seconds]
steven-netint has joined #ffmpeg-devel
microchip_ has joined #ffmpeg-devel
microchip_ is now known as putacho
* another|
opens ML
<another|>
....
<another|>
WTF? What kind of kindergarten is this? Some flimsy justification to remove access? And arbitrary rules to regain it? Is this a joke or abuse of power?
PowerAbuser has joined #ffmpeg-devel
<Marth64>
kierank: is null byte pair (0x0000) significant in a CC stream? i.e. why would encoder not just use padding pair 0x8080?
<Marth64>
assuming a53 part 4
<BBB>
another|: yeah that's not great... </sarcasm>
<BBB>
I really don't know what to do anymore, but fortunately it's not my problem anymore \o/
<PowerAbuser>
It never was 'Your' problem
<BtbN>
It's a problem of a very few people, that shits up the entire Project
<Marth64>
(nvm on the padding question, I see now)
___nick___ has quit [Ping timeout: 246 seconds]
<another|>
hmm.. jamrial: Did you resign from the old CC or the new one ?
<kierank>
Lol
Everything has quit [Ping timeout: 252 seconds]
<PowerAbuser>
Power to The All Mighty Leader!
<kierank>
PowerAbuser: welcome back!!
<kierank>
I knew you would miss the flames
<kierank>
PowerAbuser: can you review asm lessons
<PowerAbuser>
the X86 ASM SIMD lessons?
<PowerAbuser>
add more videos/visualizations to engage zoomers, its hard to take someone's time that watches tik-tok videos all the time
<PowerAbuser>
yes, that would make much more time and preparations...
<kierank>
should I put tiktok music
<PowerAbuser>
was link to lessons on hacker news?
<PowerAbuser>
or it got removed?
<jamrial>
kierank: add family guy funny moments compilations
<kierank>
PowerAbuser: draft lessons on google doc and twit
<kierank>
waiting for michaelni to give me github.com/ffmpeg access
<PowerAbuser>
link to bluesky, discord, youtube, telegram, substack, locals, rumble all possible and impossible social networks
<kierank>
yes when it's finished
<kierank>
I have to create the assignments as well
<PowerAbuser>
but yes, this is very niche, and stuff is even more fragmented, some useful SIMD can be found in rustc and related projects
<kierank>
goal is to get more simd in ffmpeg
<kierank>
compared to dav1d ffmpeg has ground to a halt
<jamrial>
vvc is getting some at least
<kierank>
true
<PowerAbuser>
later if there is interest, add other like, aarch/neon/risc of whatever it is called, but nowdays compilers under -march=native should be smart enough for moderately optimize complex loops
<kierank>
sure but I don't know that language
<another|>
PowerAbuser: What about Tinder? /s
<PowerAbuser>
tinder does not allow text content, and it is for lazy and short range pleasures.
<another|>
so SIMD is for long lasting pleasure?
<PowerAbuser>
kierank also it would be useful, it not done already, to show example how to write c support code to call SIMD functions for FFmpeg framework
<kierank>
PowerAbuser: yes assignment will cover that
<PowerAbuser>
later you could rewrite is slightly to be less FFmpeg-centric
<kierank>
there are 700 people in ffmpeg asm discord
<kierank>
actually 800 now
<PowerAbuser>
how many bots/clones?
<kierank>
dunno
<PowerAbuser>
this is new fresh discord you just made few days ago?
<kierank>
yes
<PowerAbuser>
its full of Chat-GPT like boots, they await fresh input so can be enhanced later
<kierank>
someone has made a chatgpt asm group
<PowerAbuser>
chatgpt uses asm?
<BtbN>
kierank: what exactly do you need to publish the courses on Github? Just a repo you can push to/write wiki pages? Or you got something more specific in mind?
<PowerAbuser>
well, they want to optimize every single chain of processing but its still dark there
<kierank>
BtbN: just a repo yes
<kierank>
FFMPEG Assembly Language Lessons
<kierank>
By Jonathan Craig
<kierank>
This CustomGPT is meant to be an educational study buddy to pair with the Assembly Language lessons hosted by FFMPEG (https://x.com/FFmpeg)
<kierank>
it's interesting to see how gen z learns
<BtbN>
"learns"
<Marth64>
gen iPad*
<PowerAbuser>
the ChatGPT can learn ASM coding?
<BtbN>
It can pretend that it knows everything about it at least
<kierank>
to have someone be able to reframe something you don't understand is interesting to be fair
<PowerAbuser>
there is some recompilers from C<->Rust<->python online on web but they are pretty good but not 100% correct all the time
<BtbN>
The problem with any LLM is, that they can sound convincingly correct about stuff they are flat out wrong
<BtbN>
Like the guy who "learned" how to do audio analysis from ChatGPT, and then falsely accused people of cheating in speedruns, cause ChatGPT thought him nonsense
<PowerAbuser>
I would make distinction from Text interface and model itself
<BtbN>
Took multiple actual audio engineers and forensics to debunk his findings
<PowerAbuser>
people trusting text displayed on screen
<michaelni>
kierank, what should i name the repository ?
<kierank>
michaelni: asm-lessons
<kierank>
or assembly-lessons
<kierank>
whichever you prefer
<PowerAbuser>
nasm-lessons
<michaelni>
kierank, created asm-lessons and added you with full admin permissions, i hope that works, if not tell me and ill see if i can fix it but the github member and team thing is confusing
<kierank>
thx
<kierank>
I don't have access afaik
<michaelni>
ill also try to setup a team so more people can maybe be added simpler/cleaner but the thing confuses me
<michaelni>
it says Awaiting kierank’s response
<kierank>
ah I have to accept invitation
<kierank>
You now have admin access to the FFmpeg/asm-lessons repository.
<michaelni>
good
<michaelni>
what permissions should the asm-lessons-team have ? and who should i add ? (or am i doing it wrong?)
<kierank>
don't know
<PowerAbuser>
full control on asm-lessons repo, 0-control on ffmpeg github mirror-clone-repo
<jamrial>
does github still force mrs to be enabled on all repos? would be nice to disable those
<PowerAbuser>
please reconsider making ffmpeg mirror-repo on github archive only :)
<BtbN>
Pull Requests cannot be disabled
<BtbN>
Making it archive only would make it impossible to update the mirror.
<BtbN>
That's a hard read-only
<kierank>
BtbN: #thatsthejoke
<BtbN>
What I do it put it into "emergency mode", which prevents users who are not team members from opening Pull Requests
<BtbN>
Can only enable that for a maximum of 6 months though
<PowerAbuser>
woa i need to memo that
<PowerAbuser>
how many asm lessons there will be?, it need to cover every major instruction with >=1 useful example
<kierank>
only 4 planned at vdd
<PowerAbuser>
hire someone with STF money to do visualizations in blender
<kierank>
potentially chatgpt can draw
___nick___ has joined #ffmpeg-devel
Teukka has quit [Read error: Connection reset by peer]
<ePirat>
Marth64, whats the reason the CC extraction is using the lavfi inout with the movie source filter?
<ePirat>
*input
<JEEB>
because you used to not be able to map possible sub-streams in ffmpeg.c itself, the movie source thing adds a subtitle stream based on it possibly always :D (which you can then map in ffmpeg.c)
<Marth64>
yes pretty much and it gives you the bytes in a53 format as AV_CODEC_ID_EIA_608 then from there you can pass to the ccaption_dec decoder or remux to SCC, RCWT, etc.
<Marth64>
but AV_CODEC_ID_EIA_608 within ffmpeg is a fallacy, because the a53 bytes must be split to EIA608 fields and/or CEA708 byte array, which then need to be demuxed into their individual substreams (since for each EIA608 field or CEA708 array there can be multiple caption text streams)
Everything has joined #ffmpeg-devel
mkver has quit [Ping timeout: 248 seconds]
<Marth64>
better rephrased: EIA608 and CEA708 are different codec but they are carried together as "EIA608" in ffmpeg
<beastd>
ePirat: About Git Forge not replacing MLs: I hoped so, but it wasn't really clear for me...
<beastd>
j-b, Marth64, BtbN: yes i think git forge stuff is superior to ML code review, quoting myself from earlier today: <beastd> for code contributions comments on code in PR web UIs are usually better and also iterations and integrations in CI work better there
<fflogger>
[newticket] dex: Ticket #11383 ([ffmpeg] Unable to achieve specified fps with AVFoundation camera on iOS) created https://trac.ffmpeg.org/ticket/11383