[FFmpeg-devel] movenc: add write_btrt option

[Jan Ekström]

On Fri, Apr 8, 2022 at 5:47 AM "zhilizhao(赵志立)" <quinkblack at foxmail.com> wrote:
>
>
>
> > On Apr 8, 2022, at 4:36 AM, Jan Ekström <jeebjp at gmail.com> wrote:
> >
> > On Thu, Apr 7, 2022 at 11:42 AM Eran Kornblau <eran.kornblau at kaltura.com> wrote:
> >>
> >>>
> >>>
> >>> -----Original Message-----
> >>> From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of "zhilizhao(???)"
> >>> Sent: Wednesday, 6 April 2022 11:46
> >>> To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> >>> Subject: Re: [FFmpeg-devel] movenc: add write_btrt option
> >>>
> >>>> On Apr 3, 2022, at 1:07 PM, Eran Kornblau <eran.kornblau at kaltura.com> wrote:
> >>>>
> >>>> Trying my luck in a new thread…
> >>>>
> >>>> This patch is in continuation to this discussion –
> >>>> https://eur02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fffmp
> >>>> eg.org%2Fpipermail%2Fffmpeg-devel%2F2022-March%2F294623.html&data=
> >>>> 04%7C01%7C%7Cb9907958f97048f5645708da17a9f3c8%7C0c503748de3f4e2597e268
> >>>> 19d53a42b6%7C1%7C0%7C637848315958196733%7CUnknown%7CTWFpbGZsb3d8eyJWIj
> >>>> oiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&am
> >>>> p;sdata=flQc21b5EVWTy%2Bkmw%2FncIWzdLqUxY5XFislPRs5Ij6o%3D&reserve
> >>>> d=0
> >>>>
> >>>> supports forcing or disabling the writing of the btrt atom.
> >>>> the default behavior is to write the atom only for mp4 mode.
> >>>> ---
> >>>> libavformat/movenc.c | 30 +++++++++++++++++++-----------
> >>>> libavformat/movenc.h |  1 +
> >>>> 2 files changed, 20 insertions(+), 11 deletions(-)
> >>>>
> >>>> diff --git a/libavformat/movenc.c b/libavformat/movenc.c index
> >>>> 4c868919ae..b75f1c6909 100644
> >>>> --- a/libavformat/movenc.c
> >>>> +++ b/libavformat/movenc.c
> >>>>
> >>> […]
> >>>>
> >>>> -    if (track->mode == MODE_MP4 &&
> >>>> -            ((ret = mov_write_btrt_tag(pb, track)) < 0))
> >>>> -        return ret;
> >>>> +    if ((mov->write_btrt == -1 && track->mode == MODE_MP4) || mov->write_btrt == 1) {
> >>>> +        if ((ret = mov_write_btrt_tag(pb, track)) < 0) {
> >>>> +            return ret;
> >>>> +        }
> >>>> +    }
> >>>
> >>> I prefer to handle the auto mode (mov->write_btrt == -1) in a single place, so we don’t need to change multiple lines if the condition changed, e.g., enable btrt for MODE_MOV. Please correct me if I’m wrong, mov_init() has all of the contexts to overwrite mov->write_btrt.
> >>>
> >> Makes sense, thanks for the feedback!
> >> Updated patch attached
> >>
> >> Eran
> >
> > Generally speaking I am not against this patch. Would have possibly
> > came up with something similar myself in case the actual output of
> > libavformat would have caused issues, which surprisingly enough it
> > wasn't.
> >
> > I know you have just copied the logic from tmcd or so, but I think the
> > -1 logic is unnecessary. We don't need to force writing of this
> > structure no matter what, so you either have it enabled (by default),
> > or disabled. If additional formats such as QTFF have since added the
> > btrt box into their specification, that doesn't need forcing, but
> > rather addition of it into the logic later (if you wanted forcing then
> > you'd have to deal with strict_std_compliance being
> > unofficial/experimental or higher etc, and if this was not set -
> > warning the user that a not officially defined functionality was being
> > written into the container and exiting with AVERROR_EXPERIMENTAL).
> >
> > Additionally, I thought new options go to the end of the AVOption
> > array, but then saw 1dddb930aaf0cadaa19f86e81225c9c352745262 where
> > James added "crf" into the middle of an array... so I guess since it's
> > an array and not a struct the location no longer matters as much?
> > ┐(´д｀)┌ Although the struct integer should definitely go to the end of
> > it, otherwise you are breaking existing offsets? Although thankfully,
> > the struct isn't externally exposed so someone else could chime in
> > regarding this, I am unfortunately quite tired throughout this week :P
> > .
>
> The order of options and the offset of fields in private struct have no
> effect on ABI. I take these into consideration:
> 1. Readability. Related options and fields should be put at the same
>    place.
> 2. Memory footprint. Reduce struct padding.
>

Yes, this is a minor thing within my comment, my comment was mostly
regarding the -1 case being unnecessary (since I don't think we need
to actually force-force this, just controlling whether this box is
written or not under the general rules of where it is defined).

And yes, if the order doesn't matter then grouping makes sense. It's
just that for very long "add to the end" was the general principle, so
I was mostly utilizing this as a base to request comments from others
regarding this.

Jan