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
mkver has joined #ffmpeg-devel
realies9 has quit [Ping timeout: 272 seconds]
realies9 has joined #ffmpeg-devel
minimal has quit [Quit: Leaving]
iive has quit [Quit: They came for me...]
<fflogger>
[newticket] jr_clifton: Ticket #11685 ([undetermined] ffmpeg fails to copy all metadata from webm files) created https://trac.ffmpeg.org/ticket/11685
<jamrial>
BtbN: derek tried to resolve a comment/review but couldn't. i assume he has no privileges for that, but it's weird considering it was his own comment
<BtbN>
I think he needs to be a member for that. Not sure if I made him one yet
<BtbN>
He has push access, right?
<jamrial>
yeah
<jamrial>
but does that mean normal users can't mark their own comments are resolved?
<BtbN>
No idea, never tried that
<BtbN>
I'd guess so?
<wbs>
how do I request membership on the repo on forgejo?
<wbs>
(is there a UI option for doing it, or is it just asking to be added?)
<jamrial>
register an account there, and then BtbN can add you to the org
<wbs>
I have the account already since a couple of months
<BtbN>
Me or michael can add people so far
<BtbN>
What's the username?
<wbs>
mstorsjo
<BtbN>
added you
<wbs>
thanks!
<BtbN>
can't do too much with it yet, but working on it
<fflogger>
[newticket] flyfish101: Ticket #11686 ([swscale] [Security] signed integer overflow on libswscale/output.c) created https://trac.ffmpeg.org/ticket/11686
<fflogger>
[newticket] flyfish101: Ticket #11689 ([swscale] [Security] Division by zero on libswscale/utils.c:1408:60) created https://trac.ffmpeg.org/ticket/11689
<BtbN>
Is he just going to flood the tracker with every random thing the fuzzer found?
indecisiveturtle has quit [Ping timeout: 276 seconds]
LainIwakura has joined #ffmpeg-devel
indecisiveturtle has joined #ffmpeg-devel
<kasper93>
from openssl docs
<kasper93>
> Note that BIO_get_new_index() can only be used 127 times before it returns an error.
<kasper93>
it would make sense, why streaming fails for them after "a while"
<JEEB>
huh
<JEEB>
so they have a static array of indices
<kasper93>
so, should we just globaly init one BIO_TYPE | idx
<kasper93>
and use it always
<kasper93>
we cannot repeatedly call BIO_get_new_index()
deeyes has joined #ffmpeg-devel
deeyes has quit [Ping timeout: 248 seconds]
<JEEB>
yea, something like that
<JEEB>
kasper93: check if you can utilize the same index concurrently. if not, you may need to do something ilkewhether you need to pool them for threads or something.
<JEEB>
perkele, partial buffers and rewriting :/
<kasper93>
it's just a type
<kasper93>
idx is there only to allow custom type setup
<JEEB>
ok
<JEEB>
and if we have just one type, one is enough
LainIwakura has quit [Ping timeout: 272 seconds]
<frankplow>
BtbN: Could I also be added to the forgejo org (I'm also frankplow on forgejo)
<Lynne>
TheVibeCoder: the qscale value is indeed split in 2 8-bit parts, the second part is a multiply (for hq, 15 ~= 16 == x >> 4, e.g. 12-bit coefficients)
<Lynne>
the first part is nothing more than header length (5 bits), plus maybe padding or reserved
<Lynne>
that's why the implementation sums up the 3 component encoded lengths plus the header length
<Lynne>
this is bad news, where the hell do the stairsteps go away in the ref decoder
<Lynne>
dequant is just a multiply, with the factor directly taken from the header
any1 has quit [Server closed connection]
any1 has joined #ffmpeg-devel
jarthur has joined #ffmpeg-devel
LainIwakura has joined #ffmpeg-devel
indecisiveturtle has quit [Ping timeout: 265 seconds]
<cone-061>
ffmpeg Michael Niedermayer master:3afae0f4409a: avcodec/sanm: Check w, h for subversion < 2
<cone-061>
ffmpeg Andreas Rheinhardt master:15cec7166537: checkasm/h264dsp: Fix stack-buffer-overflow, effective-type violations
<cone-061>
ffmpeg Michael Niedermayer master:33ae6cda71e6: avformat/dhav: fix reference point mixup
<cone-061>
ffmpeg Michael Niedermayer master:a408b7497776: avformat/dhav: Remove end_ prefix from variable names
<michaelni>
do we have to close prs or do they detect that the patches where applied ?
<kepstin>
if it was done with a merge commit, the github pr will automatically close. if the commits were rebased, iirc it usually won't close the pr.
indecisiveturtle has quit [Read error: Connection reset by peer]
indecisiveturtle has joined #ffmpeg-devel
<fjlogger>
[FFmpeg/FFmpeg] New comment on pull request #20038 avfilter: add vf_scale_d3d11 filter and mfenc support for d3d11 frames (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20038) by dashsantosh-mcw
kasper93_ has joined #ffmpeg-devel
kasper93 is now known as Guest9850
Guest9850 has quit [Killed (molybdenum.libera.chat (Nickname regained by services))]
Yalda has quit [Remote host closed the connection]
<fjlogger>
[FFmpeg/FFmpeg] New comment on pull request #20038 avfilter: add vf_scale_d3d11 filter and mfenc support for d3d11 frames (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20038) by dashsantosh-mcw
<wbs>
kasper93: if it works like gitlab, then it should also detect it, if you push the exact same commit that is in the PR. so if you need to rebase before pushing, then first forcepush to update the PR, then push to master and it should autoclose the PR
<JEEB>
yea I'd hope that forgejo can do that same thing, if the hash matches the HEAD of that PR, consider it merged
indecisiveturtle has quit [Ping timeout: 260 seconds]
indecisiveturtle has joined #ffmpeg-devel
LainIwakura has quit [Ping timeout: 272 seconds]
<kasper93>
what are the advantages of using openssl over schannel on windows?
mkver has joined #ffmpeg-devel
LainIwakura has joined #ffmpeg-devel
LainIwakura has quit [Ping timeout: 272 seconds]
<BtbN>
it used to be much more fully-featured
<BtbN>
but I have pulled the schannel tls handler up to speed by now
<kasper93>
I noticed all mpv builds from those popular build farms are using openssl
<kasper93>
I would think it's not neccessary.
<BtbN>
It's also a pain, license wise
<BtbN>
For Windows I'd always just use schannel
<BtbN>
security updates via OS updates alone are worth it
<kasper93>
+1
<BtbN>
Have you seen my latest comment on the PR?
<BtbN>
I don't think we need an index to begin with
<kasper93>
I haven't read that deep into this
<kasper93>
Is BIO_TYPE_NONE ok?
<kasper93>
Or BIO_TYPE_SOURCE_SINK without index?
<BtbN>
The only use of those indices I can find is to set per-type data pointers
<BtbN>
And we don't use those
<BtbN>
like, the FD type stores the FD in that data slot
<BtbN>
BIO_TYPE_NONE is just defined to 0, but I think should still be used to be explicit about what we're doing
<kasper93>
> type can be set to either BIO_TYPE_NONE or via BIO_get_new_index() if a unique type is required for searching (See BIO_find_type(3))
<kasper93>
we don't use any find
<kasper93>
so I think you are right NONE is ok to use
<BtbN>
BIO_get_ex_data and BIO_set_ex_data also use them
<BtbN>
there's apparently "data slots" which are per type
<BtbN>
very confusing system, but yeah, relatively sure we don't use any of it, so the index is entirely superfluous
<BtbN>
Is that actually the cause as to why those streams fail "after a while"?
<BtbN>
Cause that's not a new thing. That code has been in ffmpeg for ages
<kasper93>
that's a new thing
<BtbN>
The BIO idx? No
<BtbN>
The issue, yes
<kasper93>
index was added last month in 816dad231f5b312fd98e81cc6a6cbaf47c1e26a5
<fjlogger>
[FFmpeg/FFmpeg] Pull request opened: #20055 BIGSLEEP-433502298 (dashdec) and BIGSLEEP-434637586 (sanm) and a unrelated issue in subfile i stumbled across while testing (https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20055) by michaelni
<kasper93>
the issues is BIO_get_new_index() is incrementing global var and fail after 127
<BtbN>
hm, yeah. I don't think that commit is accurate
<BtbN>
ePirat: can you elaborate?
<jamrial>
i think he's on vacations until next week or so
<BtbN>
In theory a clean revert of that commit would return it to proper working state
<BtbN>
But I'd like to know why he thought/thinks it's neccesary
<BtbN>
Cause the documentation clearly says you can just use NONE
<kasper93>
> Note that BIO_get_new_index() can only be used 127 times before it returns an error.
<kasper93>
and we fail everything because of the index being -1 after some time
<jamrial>
if there's a clear breakage, maybe we can revert that commit and come back to it later
<jamrial>
unless other changes he added depend on it
<BtbN>
I'd just really like to know how that commit came to be, but it looks 100% unneccesary to me, and broken cause of the 127 times limit
<kasper93>
If we agree that index is not needed we should revert this
<BtbN>
I'm always unsure and careful when dealing with OpenSSL code
<kasper93>
and for consistency change to NONE, but this can be done separatelly
<BtbN>
cause their API is quite something
indecisiveturtle has quit [Ping timeout: 260 seconds]
<kasper93>
so google build infra around oss-fuzz and now using this for BIGSLEEP which is closed?
<fjlogger>
[FFmpeg/FFmpeg] Pull request review comment: [#20055 BIGSLEEP-433502298 (dashdec) and BIGSLEEP-434637586 (sanm) and a unrelated issue in subfile i stumbled across while testing](https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20055) by kasper93
<kasper93>
and I've seen this cooking for some time, with all the changes to oss-fuzz build / rebuild scripts