On 20.02.2015 14:30, Michael Niedermayer wrote:
On Fri, Feb 20, 2015 at 02:15:30PM +0100, Hendrik Leppkes wrote:
On Fri, Feb 20, 2015 at 1:38 PM, Michael Niedermayer <michae...@gmx.at> wrote:
On Fri, Feb 20, 2015 at 12:55:14PM +0100, Hendrik Leppkes wrote:
What else could it be?
http, ftp, who knows
also ff_rename() would interpret a "http://"; differently from
avio_open2(), so something more is needed to not end up with odd
rename() calls
It writes to separate segment files as well,
not sure that would work on anything but local files.
i dont know either
but if someone passes "ftp://..."; it should either work or fail
with a error message or mostly work while printing a warning
Some ftp servers claim thay allow to upload atomically. Who knows whats
the true. I wouldn't do that probably. But real regression is in case
someone doesn't really care to have live stream, but just want to
generate manifest and chunks on remote server directly for later use.
For me you patch is OK, but maybe it is worth to add rename as part of
protocol implementation to support these cases too. Assuming tempfile
could have hardcoded name like ".filename" instead of "filename" and
rename it when is ready.
For the record, this is the same method used in dashenc for writing
its manifest file, and it can avoid a sort-of race condition when
serving the files through a web server directly (ie. web server
reading while in the middle of re-writing it).
And the same in smoothstreaming encoder.
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
