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
Marth64 has quit [Remote host closed the connection]
Kei_N has joined #ffmpeg-devel
Kei_N_ has quit [Ping timeout: 245 seconds]
<Lynne>
is the ML dead?
<Lynne>
tried to send patches, got OK on all, but it was way too fast and there's nothing on the ML
<BtbN>
mailserver looks generally happy
<BtbN>
There's a ton of postfix/smtp[2440504]: 9C0E068C96C: to=<vm0b0t@computernewb.com>, relay=mail.computernewb.com[23.148.232.34]:25, delay=33961, delays=33959/0.03/0.8/1.3, dsn=4.3.5, status=deferred (host mail.computernewb.com[23.148.232.34] said: 451 4.3.5 <vm0b0t@computernewb.com>: Recipient address rejected: Server configuration problem (in reply to RCPT TO command))
<BtbN>
But I think that's referring to _their_ server having a config problem
MyNetAz has quit [Remote host closed the connection]
MyNetAz has joined #ffmpeg-devel
thilo has quit [Ping timeout: 276 seconds]
thilo has joined #ffmpeg-devel
thilo has quit [Changing host]
thilo has joined #ffmpeg-devel
minimal has quit [Quit: Leaving]
av500 has quit [Ping timeout: 252 seconds]
_av500_ has joined #ffmpeg-devel
c1480_ has quit [Quit: WeeChat 4.1.1]
Warcop has joined #ffmpeg-devel
jamrial has quit []
mkver has quit [Ping timeout: 260 seconds]
^Neo has quit [Ping timeout: 244 seconds]
rvalue has quit [Read error: Connection reset by peer]
<fflogger>
[editedticket] Gyan Doshi <ffmpeg@gyani.pro>: Ticket #11469 ([ffmpeg] ffmpeg_demux: readrate plays "catch up" if output is blocked, then later resumed) updated https://trac.ffmpeg.org/ticket/11469#comment:5
rvalue has quit [Read error: Connection reset by peer]
<cone-089>
ffmpeg Andreas Rheinhardt master:37348e1b2089: fftools/ffprobe: Switch to AVMutex
Albo-texime has joined #ffmpeg-devel
abdu8 has joined #ffmpeg-devel
abdu has quit [Ping timeout: 240 seconds]
<shesha4572>
Hey guys! Im trying to ingest some video files stored on my video server to ffmpeg. My server supports ranged http requests. The first range req sent by ffmpeg is with Range as 0- which my server responds with a header Content Range: bytes 0-63999999/fileSize and Content length set correctly as 64M but ffmpeg never recognizes that the response
<shesha4572>
doesnt have the whole video and gives an error of Stream ends prematurely at 64000000, should be {fileSize}. I suppose this logic is here https://github.com/FFmpeg/FFmpeg/blob/master/libavformat/http.c#L1667. There is never a check for the end offset sent by the video server. variable target_end is always 0 on first request as s->end_off is
<shesha4572>
initialized to 0
Albo-texime has quit [Remote host closed the connection]
Tanay has joined #ffmpeg-devel
<cone-089>
ffmpeg Michael Niedermayer master:4ff1d1f3d641: add tools/compare-cvelists.sh
abdu30 has joined #ffmpeg-devel
<BBB>
kierank: re: profiling, I believe I've typically had better callgraph results when omitting -fomit-frame-pointer, does that make any sense?
abdu8 has quit [Ping timeout: 240 seconds]
<kierank>
Probably but in his current command he has no debug symbols at all
abdu17 has joined #ffmpeg-devel
abdu30 has quit [Ping timeout: 240 seconds]
putacho has joined #ffmpeg-devel
putacho has quit [Read error: Connection reset by peer]
putacho has joined #ffmpeg-devel
microchip_ has quit [Ping timeout: 248 seconds]
shesha4572 has quit [Quit: Client closed]
AgencyLeader has joined #ffmpeg-devel
<AgencyLeader>
librempeg just got new demuxer and decoder!
rvalue has quit [Read error: Connection reset by peer]
rvalue has joined #ffmpeg-devel
<kierank>
AgencyLeader: someone on X complaining vgm support is bad
<AgencyLeader>
vgm is ?
<AgencyLeader>
X is closed social network
<AgencyLeader>
i do not consume closed social networks
<kierank>
Some codec on sega
<AgencyLeader>
get discord, it does not need your phone
<AgencyLeader>
kierank: that is not streamlike format, its more like sequenced or worse, executable like music
<AgencyLeader>
and such sequenced, to some extent even sound bank formats are pure mess
<AgencyLeader>
and adding support for generic sequenced music, including programmed music, is so huge that is more complex than whole FFmpeg current codebase
<devinheitmuell-1>
Question: there’s been a longstanding limitation where you can have multiple side data elements with the same type for AVFrame, but not for AVPacket. Does anyone know if anyone is working to improve that situation?
<jamrial>
devinheitmuell-1: not really. there are afaik no packet side data types where having more than one entry makes sense
<devinheitmuell-1>
jamrial: In my case it’s an issue with AV_FRAME_DATA_SEI_UNREGISTERED (for stuff like proprietary messages). For A53 captions it was worked around with years ago by appending multiple entries to the same type, but that only works if all the side data types are the same size (with SEI unregistered data, you can have each one with a different size)
<JEEB>
(51
<JEEB>
whoops
Tanay has quit [Ping timeout: 268 seconds]
ccawley2011 has quit [Ping timeout: 252 seconds]
gnafu has quit [Quit: "Rebooting for kernel update..."]
<kierank>
BBB and astrange fighting the good fight on there
HarshK23 has joined #ffmpeg-devel
c1480 has quit [Quit: WeeChat 4.1.1]
c1480 has joined #ffmpeg-devel
mindfreeze has quit [Ping timeout: 265 seconds]
mindfreeze has joined #ffmpeg-devel
ccawley2011 has quit [Ping timeout: 252 seconds]
bhaskar has joined #ffmpeg-devel
<AgencyLeader>
BBB and astrange are irrelevant
<AgencyLeader>
LLM, AI and compilers are winning the war
<BBB>
as long as dav1d is faster than any AI-produced av1 decoder [with the caveat that you're not allowed to train it on dav1d], I'm ok
<AgencyLeader>
that is unfair
<AgencyLeader>
train on dav1d all day and year
<BBB>
ok, so you want your AI to write you an av2 decoder before you spend money on a team of experts to write dav2d. how?
<BBB>
if the result is "we'll pay both", then I don't see the point
<BBB>
double the cost, same outcome. why bother?
<AgencyLeader>
are you against the AI ?
<BBB>
I don't see the point if it can only regurgitate stuff we already know
<AgencyLeader>
if something can be coded: faster + better + more efficient
<AgencyLeader>
or for automated object detection?
<BBB>
in rust?
<AgencyLeader>
if its the fastest, best, and most correct implementation, yes.
<BBB>
that rust-avfilter email felt like it was genAI'ed
<BBB>
so much text
<AgencyLeader>
yes, modern man cant write lot of text at all.
<AgencyLeader>
too much tiktok/yt/fb/... usage per day
ccawley2011 has joined #ffmpeg-devel
ramiro has quit [Ping timeout: 244 seconds]
ramiro has joined #ffmpeg-devel
<BBB>
but honestly, can you read that much text? it's not like I can't focus, I'm able to understand complete video standards etc.
<BBB>
but that much text, it's just very ... confusing is not the right word. distracting?
witchymary has quit [Ping timeout: 272 seconds]
ccawley2011 has quit [Ping timeout: 260 seconds]
<AgencyLeader>
yes, its too much verbose to send actual meaning, big word salad.
<BBB>
kierank: do you need more word fights? I thought the hackernews one was ok in the end
<BBB>
I've seen worse
<BBB>
the hackaday one looks like good ol' slashdot, as in I'm not going anywhere near there
<AgencyLeader>
but does anyone here use copilots/ai/whatever when coding ?
putacho has quit [Quit: There is no spoon!]
<JEEB>
I have utilized locally executable LLM for tab-autocompletion. generally harmless.
<AgencyLeader>
yea, but i mean more advanced stuff
<JEEB>
I've tested deepseek's qwen-32b model by posting a quote from a spec and requested it to implement a basic parser in rust
microchip_ has joined #ffmpeg-devel
<JEEB>
(that's just qwen that was taught against the main deepseek model)
<JEEB>
and it did indeed generate some valid code
<Lynne>
I've had no good experiences with AI generated anything
<JEEB>
but I would not utilize AI-generated code, probably better to have it attempt to analyze your stuff
<Lynne>
image generation has stagnated for a year now, deepseek distills are lobotomized, real deepseek needs at least a hundred gigs of vram
<AgencyLeader>
huh
<AgencyLeader>
really?
<AgencyLeader>
that its still useless
<JEEB>
it's fun when the stuff gets worse the further you continue. where its initial version of something utilized nice early exits etc, but then it gets more and more derp
<AgencyLeader>
but there is something called RIFE
<AgencyLeader>
that does video stuff
<Lynne>
code-wise, I've never got anything good out of AI, even basic functions have errors that I point out, and the AI goes "oh, yeah, if you say so"
<Lynne>
if AI learned from errors, we'd have something in like a week, but for worse and better, its all stateless between sessions
<AgencyLeader>
how/what you use?
<JEEB>
AgencyLeader: I'm pretty sure due to your commits in open source you can also access the paid tier of copilot for free if you want
<JEEB>
since even I have access
<JEEB>
so you could just poke into it yourself
<AgencyLeader>
yea, but i need something to tweak in settings
<AgencyLeader>
or it already works
<AgencyLeader>
but isnt bussiness pro plan better quality?
<AgencyLeader>
or its just unlimited numbers?
<AgencyLeader>
in usage^
<JEEB>
the open source developer thing is the full paid thing
<AgencyLeader>
looks like i'm not eligible for free full pro copilot
<AgencyLeader>
As a maintainer of a popular open-source repository.
<AgencyLeader>
librempeg is not "popular"
c1480 has joined #ffmpeg-devel
c1480 has quit [Read error: Connection reset by peer]
<BBB>
you could cheat and mention you've done some minor contributions to ffmpeg
<AgencyLeader>
lol
c1480 has joined #ffmpeg-devel
<BBB>
ffmpeg is a popular cheat code in many such circumstances
<BBB>
like xyzzy in minesweeper
c1480 has quit [Ping timeout: 248 seconds]
<JEEB>
pretty sure you are since you have commits in FFmpeg. after all, it's an automated check
<JEEB>
that said I recall their docs were crap and the way I enabled it was funky
c1480 has joined #ffmpeg-devel
c1480 has quit [Ping timeout: 248 seconds]
AgencyLeader has quit [Quit: Client closed]
realies9 has joined #ffmpeg-devel
realies has quit [Ping timeout: 248 seconds]
realies9 is now known as realies
halloy7570 has joined #ffmpeg-devel
halloy7570 has quit [Ping timeout: 245 seconds]
<j-b>
good morning folks
Guest67 has quit [Ping timeout: 240 seconds]
witchymary has joined #ffmpeg-devel
c1480 has joined #ffmpeg-devel
* microchip_
farts at j-b :P
<microchip_>
oops :)
<j-b>
microchip_: you are too far.
<microchip_>
haha
<j-b>
microchip_: try inventing Farts-over-Ip
<microchip_>
lol
witchymary has quit [Remote host closed the connection]
<jamrial>
devinheitmuell-1: what's your usecase? You need multiple AV_PKT_DATA_A53_CC entries per packet?
<devinheitmuell-1>
jamrial: No, A53_CC is already handled by simply appending the data. My use case is “unregistered” SEI metadata. I want to be able to pass it through the v210enc so I can put it out over SDI using decklink.
<jamrial>
ok, will think how it could be done
<devinheitmuell-1>
The AVFrame can contain multiple side data entries for a given type, but I can’t simply do a conversion to the equivalent AVPacket type because that API doesn’t allow multiple entries for a given packet (calling it more than once just deletes the original entry)
abdu17 has quit [Quit: Client closed]
abdu has joined #ffmpeg-devel
<devinheitmuell-1>
If all the messages were the same size, then we could do it just like we did A53_CC (where it just assumes an array of structs all of the same size), but with unregistered SEI metadata each may be a different length.
<BtbN>
nvenc seems to encode the timecode for h264 just fine btw.
<devinheitmuell-1>
jamrial: Thanks. Any ideas in this area are welcome.
<devinheitmuell-1>
BtbN: Cool. Now if we could just get x264 patched I would be very happy. :-)
<BtbN>
Judging by the comment in their code, they actively don't want to add support
<JEEB>
BtbN: what was the magical sequence of actions to have code.ffmpeg.org send an email? decided to test forgejo since I've not utilized it yet.
abdu has quit [Quit: Client closed]
<devinheitmuell-1>
Really? Dekker sent me a proposed patch at once point. An API change is required because the picture_timing needs to be modified (i.e. you can’t simply append an extra SEI blob). I hadn’t seen anything to suggest they specifically didn’t want it supported.
abdu has joined #ffmpeg-devel
<BtbN>
There is some weird bug going on, where it just doesn't even try to send out activation mails I think...
abdu has quit [Quit: Client closed]
abdu has joined #ffmpeg-devel
Traneptora_ has joined #ffmpeg-devel
Traneptora has quit [Quit: Quit]
Traneptora_ has quit [Client Quit]
Traneptora has joined #ffmpeg-devel
Traneptora has quit [Quit: Quit]
Traneptora has joined #ffmpeg-devel
witchymary has joined #ffmpeg-devel
<BtbN>
JEEB: I manually activated your account now and opened a bug with Forgejo about this
<JEEB>
gotcha
<BtbN>
To me it looks like Forgejo is unable to mail on aarch64?!
<BtbN>
However that makes sense. It's the only thing that changed
<JEEB>
yea that doesn't sound like usual unless some bins/libraries etc are literally missing
<BtbN>
It's go
<BtbN>
literally a signel binary sitting there
<BtbN>
If a library was missing, that binary would not exist
<JEEB>
that's what I'd expect, yes
<JEEB>
but if it's using the container OS's functionality that it's calling out for
<JEEB>
which for whatever reason is not there in the aarch64 base image