[FFmpeg-devel] [PATCH] avformat/avio{, buf}: introduce public AVIOContext::bytes_read

[Michael Niedermayer]

On Mon, Oct 11, 2021 at 10:24:47PM +0300, Jan Ekström wrote:
> On Mon, Oct 4, 2021 at 12:12 PM Jan Ekström <jeebjp at gmail.com> wrote:
> >
> > On Mon, Oct 4, 2021 at 1:06 AM Michael Niedermayer
> > <michael at niedermayer.cc> wrote:
> > >
> > > On Mon, Oct 04, 2021 at 12:25:26AM +0300, Jan Ekström wrote:
> > > > On Sat, Oct 2, 2021 at 2:51 PM Michael Niedermayer
> > > > <michael at niedermayer.cc> wrote:
> > > > >
> > > > > On Sat, Oct 02, 2021 at 02:42:52PM +0300, Jan Ekström wrote:
> > > > > > On Sat, Oct 2, 2021 at 1:32 PM Michael Niedermayer
> > > > > > <michael at niedermayer.cc> wrote:
> > > > > > >
> > > > > > > On Sun, Sep 26, 2021 at 06:48:18PM +0300, Jan Ekström wrote:
> > > > > > > > Such a field can be seen as generally useful in cases where the
> > > > > > > > API user is not implementing custom AVIO callbacks, but still would
> > > > > > > > like to know if data is being read even if AVPackets are not being
> > > > > > > > returned.
> > > > > > > > ---
> > > > > > > > Originally I thought about making an accessor for the private field, to
> > > > > > > > not grow the public struct's size (and have a duplicate field, as well
> > > > > > > > as making sure the value was read-only). But an objection was raised
> > > > > > > > that such accessors should be refrained from as they unnecessarily
> > > > > > > > filled the function symbol space or so. Together with the objection, a
> > > > > > > > proposal of making it a field on the public struct that was only written
> > > > > > > > to was proposed.
> > > > > > > >
> > > > > > > > This patch follows that proposal.
> > > > > > > >
> > > > > > > >  doc/APIchanges        | 3 +++
> > > > > > > >  libavformat/avio.h    | 5 +++++
> > > > > > > >  libavformat/aviobuf.c | 2 ++
> > > > > > > >  libavformat/version.h | 2 +-
> > > > > > > >  4 files changed, 11 insertions(+), 1 deletion(-)
> > > > > > >
> > > > > > > There are 3 statistics, read, write and seek
> > > > > > > shouldnt all 3 be provided to the user?
> > > > > > >
> > > > > > > thx
> > > > > > >
> > > > > >
> > > > > > I added one which I have seen actually utilized by at least one API
> > > > > > client, and then others could be added as per responses.
> > > > > >
> > > > > > That is why I pinged, as I had not received any responses - either
> > > > > > positive or negative.
> > > > > >
> > > > >
> > > > > > Writing I can see a use for, seek I am not as sure of. But if you
> > > > > > believe all of them should be exposed I am fine with that.
> > > > >
> > > > > seek is timeconsuming especially if its over a network due to
> > > > > latency.
> > > > > So for example if suddenly the number of seeks changes that
> > > > > could be interresting.
> > > > >
> > > > > thx
> > > >
> > > > I would prefer to add fields which were noted as specifically private
> > > > and then cleaned up when there are actual API client users that would
> > > > see them as useful, or if there are clear use cases where they'd be
> > > > useful. I have seen the read bytes statistic actually being utilized
> > > > by an API client with a comment:
> > >
> > > Assume a network protocol, TCP, UDP, HTTP, RT*P whatever
> > > how do you tune the buffer sizes ?
> > > Can the number of seeks be used ?
> > > or from a different point of view, if there are alot of seeks should
> > > a user app try to increase the buffer sizes ?
> > >
> > > maybe iam missing something but when playing a not perfectly interleaved file
> > > over the network the buffer size should be what makes the difference between
> > > that working or not working
> > > ideally a user app shouldnt need to mess with this, of course and these values
> > > should all be automagically adjusted
> > >
> > > If a user app fails to get packets in realtime over the network, it would
> > > fail to play that stream. Some user apps could display a warning message to
> > > the user about it.
> > > If now the user app has access to the number of seeks it could be more
> > > specific in the warning to the user.
> > > "Unable to play network is maybe too slow"
> > > "Unable to play buffer is maybe too small or file is poorly interleaved"
> > > ...
> > >
> > > Maybe iam just seeing all this from the wrong side i dunno but to me it seems
> > > usefull to a user app to have access to the number of seeks and these seem
> > > non contrived use cases to me ... Ive gotten random point to nowhere
> > > warnings about playback issues and restarting the computer obviously that
> > > never was the issue.
> > >
> > > thx
> > >
> >
> > OK, I think this is now focusing on the wrong point, sorry.
> >
> > I think it's just better for me to note that I am not the best person
> > to post (and thus be the one to argue for the usefulness in reviews if
> > someone asks why I am bringing those private entries that were once
> > removed back to the public struct) of those other entries.
> 
> Ping? Is this now in a purgatory state due to me not wanting to be the
> one to argue for the other options in review?

Well, iam a bit undecided about what is best
a more broader IO statistic API that provides more details and that can be
extended and maybe can be disabled so as not to waste resources when the
statistics are not needed was a thought i had.
Simply adding one field because you need it, then the next person adds
one field he needs and so on. In the end the result is called a mess.
a few fields scattered over the structs and copied around between them
every time they change.

IIUC you are not interrested in implementing and arguing for a more
complete IO statistics API. And I have not enough time to implement
and argue for it myself ATM. There are no objections against your
patch so nothing stops you from applying 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/20211012/7ee1e322/attachment.sig>