[FFmpeg-cvslog] In retry_transfer_wrapper, do not check url_interrupt_cb,

[Ronald S. Bultje]

Hi,

On Mon, Mar 7, 2011 at 7:55 AM, Martin Storsj? <martin at martin.st> wrote:
> On Sun, 6 Mar 2011, Michael Niedermayer wrote:
>> On Sun, Mar 06, 2011 at 11:48:22AM +0100, Nicolas George wrote:
>> > Le sextidi 16 vent?se, an CCXIX, Michael Niedermayer a ?crit?:
>> > > a 2 or 3 level abort should be used. Like unix TERM & KILL
>> >
>> > It is an interesting solution, but is it necessary?
>> >
>> > With the check for url_interrupt_cb at the end of the loop, the code works,
>> > just as it has worked for a long time. It just needs a little clarification
>> > in the documentation.
>> >
>> > Adding this multi-level interrupt system would not solve any problem
>> > currently known. And when a problem will actually arise, we do not know
>> > whether this will be enough to solve it.
>> >
>> > Designing a solution without a problem is rarely a good idea.
>> >
>> > For now, I think it would be best to just restore the check of
>> > url_interrupt_cb at the end of the loop and document that url_interrupt_cb
>> > makes the protocol it has interrupted unusable.
>>
>> we cannot make the writing protocol unuseable as a first level meassure to quit
>>
>> when q is pressed we need to write, trailer + index + seek back to begin +
>> update filesize and frame number in header for some formats.
>> when q is pressed we didnt even yet start writing the index and trailer
>>
>> multiple levels are the obvious solution.
>> level 1, simply stop in the user app cleanly (this needs no call back value)
>> level 2, hard abort any reading protocols returning incomplete data if needed
>> level 3, hard abort any writing protocols this will cause inconsistence between
>> ? ? ? ? ?what is written and what the muxer wanted to write. Depending on file
>> ? ? ? ? ?this may render the file unplayable or some frames at the end missing
>
> Yes, I think something like this would be useful. A simple first step
> would be without a distinction between level 2 and 3, only interpreting
> the first 'q'/ctrl+c as a graceful "please don't transcode any more data,
> start writing the trailer now", and only triggering url_interrupt_cb() on
> the second one.

Before we put any more effort in url_interrupt_cb(), can we please
kill it and replace it by something that works across instances,
threads and so on? What if I have an application that has 1000
URLContexts and I only want to quit one of them?

Ronald