[FFmpeg-devel] [PATCH] avformat/flvenc: Specify codec tag with MKTAG

[Michael Niedermayer]

Hi Zhao Zhili

On Sat, May 17, 2025 at 10:05:34PM +0800, Zhao Zhili wrote:
> 
> 
> > On May 17, 2025, at 19:14, Timo Rothenpieler <timo at rothenpieler.org> wrote:
> > 
> > On 17.05.2025 06:35, Zhao Zhili wrote:
> >>> 在 2025年5月17日，上午1:39，Timo Rothenpieler <timo at rothenpieler.org> 写道：
> >>> 
> >>> On 16.05.2025 19:24, Zhao Zhili wrote:
> >>>>>> On May 17, 2025, at 01:10, Zhao Zhili <quinkblack-at-foxmail.com at ffmpeg.org> wrote:
> >>>>> 
> >>>>> 
> >>>>> 
> >>>>>> On May 17, 2025, at 00:27, Timo Rothenpieler <timo at rothenpieler.org> wrote:
> >>>>>> 
> >>>>>> On 16.05.2025 17:59, Zhao Zhili wrote:
> >>>>>>>> On May 16, 2025, at 22:52, Timo Rothenpieler <timo at rothenpieler.org> wrote:
> >>>>>>>> 
> >>>>>>>> On 16/05/2025 16:24, Zhao Zhili wrote:
> >>>>>>>>> From: Zhao Zhili <zhilizhao at tencent.com <mailto:zhilizhao at tencent.com>>
> >>>>>>>>> ffmpeg -i input.mp4 -c copy -tag:v av01 output.flv
> >>>>>>>>> [flv @ 0x143204080] Tag av01 incompatible with output codec id '225' (10va)
> >>>>>>>> 
> >>>>>>>> I don't quite understand what causes this.
> >>>>>>>> Is this an issue when running on big endian architectures?
> >>>>>>>> I'm pretty sure I tested all combinations of codecs with muxing and demuxing, and never ran into that error.
> >>>>>>> The key point is when specify tag via command line, e.g., -tag:v av01, it’s
> >>>>>>> passed to AVCodecParameters codec_tag in little endian.
> >>>>>>> You didn’t see the error because without specify the tag explicitly, codec_tag is copied
> >>>>>>> from AVOutputFormat codec_tag to AVCodecParameters codec_tag, so they are
> >>>>>>> the same.
> >>>>>>> Another example is codec_mp4_tags.
> >>>>>> 
> >>>>>> This still irks me as wrong.
> >>>>>> There is _a lot_ of places all over flvenv.c, in all kinds of functions, that hard-depend on the values in par->codec_tag being from the _codec_ids tables at the top of the file.
> >>>>>> Like, they contain flv specific audio and video codec IDs for the pre-ext-flv codecs, those would all also break.
> >>>>>> 
> >>>>>> So there seems to be a deeper issue there if those values can be overridden from the commandline. The encoder clearly does not expect that.
> >>>>>> 
> >>>>>> Looking at the code this error comes from:
> >>>>>> https://github.com/FFmpeg/FFmpeg/blob/master/libavformat/mux.c#L314
> >>>>>> 
> >>>>>> It looks to to me like it's working exactly as intended and required by flvenc, protecting it from invalid tags.
> >>>>>> 
> >>>>>> So, when the user provides a custom tag that is invalid, isn't that kinda on the user?
> >>>>>> The check you're running into does what it's supposed to:
> >>>>>> It detects that the provided tag is invalid for this codec in this container.
> >>>>>> 
> >>>>>> Why do you want to override it anyway? There is only exactly one valid tag for each codec.
> >>>>> 
> >>>>> A user shows the error message to me. Because he know there are -tag option for mp4, and
> >>>>> enhanced-rtmp use fourcc for extended codecs, so he thought it should work.
> >>>>> 
> >>>>> The -tag:v av01 is redundant, it should be a NOP, not trigger error. The strings “av01”
> >>>>> is the right order of fourcc in spec. The endian issue should be limited to the internal.
> >>>>> Current error message is confusing, because it shows 10va instead of av01.
> >>>> Doc from Microsoft shows fourcc use small endian
> >>>> https://learn.microsoft.com/en-us/windows/win32/directshow/fourcc-codes
> >>>> While wiki and enhanced rtmp spec says it’s big endian
> >>>> https://en.wikipedia.org/wiki/FourCC
> >>>> It’s clear in av_fourcc_make_string
> >>>> that we use small endian.
> >>>> For normal codec id in flv, e.g, 7 for H.264, it’s not a big issue, since they’re not fourcc.
> >>> 
> >>> This patch just fixes it for a very small subset of codecs in flv though.
> >>> There's nothing indicating that -tag:v is needed or sensibly supported in flvenc.
> >>> If you'd want to pass h264 or aac as fourcc, it'd be flat out impossible, and no easy fix is available.
> >>> 
> >>> I'd rather not complicate flvenc, even if just a little bit, just to swap around the endianness of the few codecs that do use a fourcc based tag.
> >>> 
> >>> flvenv should probably just completely ignore -tag:v, since the option makes no sense for it anyway.
> >> I can remove setting tag list to AVOutputFormat, does that works for you?
> > 
> > Won't that break even more things?
> > The code heavily relies on the tags being what they are, and passing the tag list to AVOutputFormat would remove avformat enforcing that, with said error when trying to override it with something invalid.
> 
> The enforcing logic has a precondition, that codec_tag is in little endian.
> flv_video_codec_ids and flv_audio_codec_ids don’t match this requirement.
> 
> > 
> > I honestly don't think anything needs to be changed at all.
> > Passing custom tags is simply not something flv sensibly supports.
> 
> AVOutputFormat codec_tag is API exported to users. Even if users
> are not supposed to pass codec_tag to flvenc, the exported information
> should match the tradition, that codec_tag is in little endian according to
> the code in av_fourcc_make_string, mov, matroska, and so on.

You are correct but the code did use the opposit endianness for a long
time.
Changing this would need a api version bump,
should be documented in APIChanges
we would have to make sure nothing breaks

In the end iam not sure the work is worth given this seems not
really affecting anything except an error message for a case that isnt
really usefull.

That said, codec_tag is container specific, but it also often has
large numbers of values shared between containers.

If for example mov had a codec_tag that would match one in flv then
having endianness matching between the 2 would be nice
OTOH if flv is its own universe, iam not sure the work to change
endianness is worth the work.
But if someone wants to do that "cleanup" work, iam not against
it, assuming we dont cause others some headaches with it

thx

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

The real ebay dictionary, page 2
"100% positive feedback" - "All either got their money back or didnt complain"
"Best seller ever, very honest" - "Seller refunded buyer after failed scam"
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20250518/c181eb9e/attachment.sig>