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
<jamrial>
kasper93: f944a70fccd broke it, somehow
<jamrial>
can reproduce with mingw-w64 on msys2
<BtbN>
it might simply be the first checkasm test that uses one of those two remaining avpric things that get dllimported?
<jamrial>
linking complains about avpriv_cga_font and avpriv_vga16_font from xga_font_data.c in lavu
<jamrial>
neither the filter or the test use it
<BtbN>
Yeah, I don't see an immediate reason why it would cause that
rvalue has joined #ffmpeg-devel
minimal has quit [Quit: Leaving]
<jamrial>
BtbN: using getters worksaround it
<jamrial>
and these arrays are the only ones using av_export_avutil, so might as well just stop doing it
<kasper93>
IIRC dllimport is important for data importing from dll. For functions ot
<kasper93>
it's optional or basically optimization
<kasper93>
so getters may workaround this
<cone-194>
ffmpeg Daniel N Pettersson master:e56fd2af1a0b: avformat/tls_openssl: add hostname for verification
<BtbN>
How? You have normal repo write permissions.
<BtbN>
I don't see any permission I could set that sounds like it would allow that.
<kasper93>
what will happen to trac?
<kasper93>
stay as read only?
kasper93_ has joined #ffmpeg-devel
kasper93 is now known as Guest527
Guest527 has quit [Killed (silver.libera.chat (Nickname regained by services))]
kasper93_ is now known as kasper93
rvalue has quit [Read error: Connection reset by peer]
rvalue has joined #ffmpeg-devel
<Lynne>
BtbN: I don't see a button to create new issues
<Lynne>
err, labels
<Lynne>
I'd like to have one for each component, e.g. avcodec, avfilter
<Lynne>
since they won't just be used for issues but for MRs too
<Lynne>
also it would be really awesome to have some script running which reads MR titles and automatically tags them
System_Error has quit [Remote host closed the connection]
jamrial has quit []
System_Error has joined #ffmpeg-devel
cone-194 has quit [Quit: transmission timeout]
Flat has quit [Ping timeout: 268 seconds]
mkver has joined #ffmpeg-devel
Flat has joined #ffmpeg-devel
Flat has quit [Ping timeout: 245 seconds]
<Lynne>
BtbN: nevermind, I'm an idiot and forgot to login
<Lynne>
the auto-tagging feature would be nice though
mkver has quit [Ping timeout: 252 seconds]
Flat has joined #ffmpeg-devel
TheVibeCoder has joined #ffmpeg-devel
NotWarcop has joined #ffmpeg-devel
Warcop has quit [Ping timeout: 248 seconds]
kurosu has joined #ffmpeg-devel
jarthur has quit [Remote host closed the connection]
jarthur has joined #ffmpeg-devel
wyatt8750 has quit [Ping timeout: 276 seconds]
ngaullier has joined #ffmpeg-devel
wyatt8740 has joined #ffmpeg-devel
Teukka has quit [Read error: Connection reset by peer]
Teukka has joined #ffmpeg-devel
Teukka has quit [Changing host]
Teukka has joined #ffmpeg-devel
jarthur has quit [Quit: jarthur]
indecisiveturtle has joined #ffmpeg-devel
kwizart has quit [Quit: leaving]
sm2n has quit [Remote host closed the connection]
OctopusET has quit [Remote host closed the connection]
<kierank>
TheVibeCoder: move to forjego going well I see
<kierank>
professional and efficient ffmpeg infrastructure
<haasn>
what's the current status of AVIF support in ffmpeg?
<haasn>
seems like it relies on libdav1d for decoding
<haasn>
it also doesn't handle transparent AVIF correctly, just exposes the alpha channel and color image as separate video streams
OctopusET has joined #ffmpeg-devel
sm2n has joined #ffmpeg-devel
<JEEB>
haasn: same as other HEIF based formats, basically.
<JEEB>
and yea, since it's just AV1 it requires an AV1 decoder, dav1d being recommended
<Lynne>
alpha in avif is just a separate gray image
<JEEB>
yea, aux images in HEIF
<JEEB>
you'd have to do something like attach the aux image into a side data for the AVPacket, and then on the decoder level check if it exists and then do the whole yuva dance with two decoder instances if you want it to be old style handled (a la VP6 or the hackpatch I did for VP8)
<JEEB>
alternatively the whole auto-inserted filter approach
<JEEB>
but yea, it isn't automagically handled right now by any API client as far as I know
<haasn>
TheVibeCoder: hi, I did a comprehensive investigation of different premultiplication methods and came to the conclusion that adding the ((a >> 1) & 1) bias lowers the accuracy both on premultiply and on round-trip (premul+unpremul)
<haasn>
by an order of magnitude, for round trip
<haasn>
you can make it a bit better by adding another clamp FFMIN(asrc, dst) but both of these methods have worse error than just doing the naive calculation without the offset
<haasn>
in fact, the naive method gives only about 0.6% worse SAD than the correctly rounded result (using double math)
<haasn>
maybe it's needed only for the limited range case or YUV case
<haasn>
nope, also improves roundtrip PSNR in the limited/full range YUV cases by ~2 dB on average to remove the extra term
<TheVibeCoder>
haasn: cool
<haasn>
in related news, I'm revisiting alpha mode in AVFrame. for codecs like PNG that mandate straight alpha, should the codec itself perform the conversion, or should we set up the full avfilter machinery?
<haasn>
I am a bit wary of adding more parameters to the avfilter negotiation, especially given that the current architecture can't really handle alpha mode being its own negotiation category; it would have to be lumped in with vf_scale somehow
realies9 has quit [Quit: ~]
realies9 has joined #ffmpeg-devel
System_Error has quit [Ping timeout: 244 seconds]
<Lynne>
conversion to?
<haasn>
Lynne: premultiplied to straight alpha
<Lynne>
what decoder outputs premult alpha? texture compression ones?
<haasn>
Lynne: AVIF can in theory
<haasn>
though ffmpeg does not support it yet
<Lynne>
would it be too painful to add a flag like color_range?
<haasn>
more practically, many input files in the wild are premul alpha regardless of any codec signalling
<JEEB>
I think also MOV or Prores had a flag for alpha type
<haasn>
Lynne: I did; it was rejected by NG because I exit out with an error if you specify an incompatible color_range, rather than performing an automatic conversion
<JEEB>
since someone showed the option in a video editor in a screen capture
<haasn>
and I am wondering if it's okay to have said automatic conversion in the codec or if we really need to spend an extra dozen hours hacking it into the avfilter negotiation somehow
System_Error has joined #ffmpeg-devel
<Lynne>
haasn: was he wrong, or right, in your opinion?
<haasn>
in my opinion, I agree with him that automatic conversion would be better than erroring out
<haasn>
but I disagree with him that having working automatic conversion represents a requirement to merging what is ostensibly an improvement over the status quo
<haasn>
letting perfect be the enemy of good, etc
<JEEB>
I wonder if something like the "please insert this bsf before this module" kind of system would work for this. although it would be kinda fighting with the auto-conversion :D
<haasn>
other formats that can output premul alpha, according to wikipedia, includig jpeg xl and tiff
<haasn>
though I'm not sure if they have header tagging for it
<BtbN>
Trac can either be migrated into Forgejo somehow (but I'm not a fan of that), or set read-only
jamrial has joined #ffmpeg-devel
<BtbN>
It has a Wiki and Issues. Maybe migrating the Wiki is more realistic, but I don't really feel like dumping over 10k issues into Forgejo will make for a good experience
cone-386 has joined #ffmpeg-devel
<cone-386>
ffmpeg James Almer master:48ef4caac2d4: avcodec/ffv1dec: don't add offsets to more NULL pointers
<cone-386>
ffmpeg James Almer master:87cc21374860: avformat/hevc: don't print parameter_set_id for any NALU
<cone-386>
ffmpeg James Almer master:b0159af6bc76: avfilter/f_metadata: use the return value of vsnprintf() to write the argument list
<jamrial>
i could see p->data[2] being NULL for semi-planar, but that's not the case
<jamrial>
oh, this is rgb...
kwizart has joined #ffmpeg-devel
System_Error has quit [Remote host closed the connection]
System_Error has joined #ffmpeg-devel
jarthur_ has joined #ffmpeg-devel
<jamrial>
BtbN: is it normal to get emails for mrs and issues sent to my email?
<BtbN>
all members of the org do, unless you opt out
<jamrial>
ok
<jamrial>
ideally, a bot would post here like with trac
<cone-386>
ffmpeg James Almer master:86670fec9cc8: avcodec/ffv1dec: don't check chroma_planes for f->colorspace > 0
<jamrial>
BtbN: also, forgejo doesn't like my ssh key. it wants to be more than 2048bits :/
<another|>
you have a key with < 2048b ?
<jamrial>
no, exactly that
<another|>
ah
indecisiveturtle has quit [Ping timeout: 244 seconds]
<haasn>
is that an unreasonable requirement nowadays?
<haasn>
seems like NIST considers RSA 2048 secure until 2030
<BtbN>
I got surprised by that too, but didn't find it unreasonable
<BtbN>
I use ed25519 keys everywhere anyway now
andrewrk has quit [Remote host closed the connection]
andrewrk has joined #ffmpeg-devel
<nevcairiel>
Seems unreasonable to hardcode such a requirement, make an option
<nevcairiel>
my typical keys are rsa 4096, but i've had servers that now complain about rsa entirely already, linux people like to throw such requirements in their defaults :P
ngaullier has quit [Remote host closed the connection]
<Lynne>
I modernized all my keys at once more than a year ago, it wasn't at all painful
<nevcairiel>
i guess ed25519 is the recommendation still?
<beastd>
kasper93,BtbN: The old #NNNNN references and ticket confusion in general should probably be tackled (better sooner but at latest when we decide to switch). We should look into bumping the issue number in forgejo to say 20000 so no conflicts of the numbers are possible.
<BtbN>
I don't think that's possible
<beastd>
later we could generate all issues from trac as tickets in forgejo so that they contain the trac link in the description. anyway i would be willing to help
<BtbN>
I don't think throwing over 10000 issues into Forgejo will be nice
<BtbN>
we'll just disable issue creation on trac
<beastd>
BtbN: you mean not possible with UI knobs and configs? or impossible in general?
<BtbN>
I think it's just DB IDs
<BtbN>
wait no, that makes no sense. There is not one DB per Repo
<beastd>
BtbN: i think the issue number can be bumped without creating all the tickets, but i need to look into the db schema again
<beastd>
iirc the id is for all issues. but there should be another field index per repo that is the issue number
<beastd>
i should be able to have a look into implementation tomorrow. my memory is not really complete about the details right now.