[MPlayer-dev-eng] [PATCH] bad seeking in stream.c, buffer flushed to early

[Reimar Döffinger]

Hi,

>>> stream_seek_long first sets s->buf_pos=s->buf_len=0; (thus clearing the
>>> buffer) before attempting to seek, which effectively makes it seek
>>> forward instead of backwards on linear streams (e.g. when doing cat test
>>> | mplayer -nocache -).
>>> Simply commenting this out made it work for me and I couldn't detect any
>>> ill effects, but somebody who knows this code should check this.
>>
>> Don't think this is good idea. The name *_seek_long meen that we must
>> seek, thus empty the buffer.
> 
> The question is why do we have to empty the buffer that early and not 
> after we actually did seek successfully? As it is currently, if the seek 
> fails for some unexpected reason, the stream position will change 
> anyway, which I consider a bad thing...
> I also think that the buffer is emptied somewhere else anyway, as 
> seeking still works for me after this patch...

Can anyone help here? Otherwise I might soon consider applying and 
waiting for the bugreports...

Greetings,
Reimar Döffinger