[FFmpeg-devel] [RFC] seek when streamed

[Baptiste Coudurier]

Hi,

Michael Niedermayer wrote:
> On Tue, Nov 13, 2007 at 12:31:55AM +0100, Baptiste Coudurier wrote:
> 
>>Hi,
>>
>>Michael Niedermayer wrote:
>>
>>>On Sun, Oct 07, 2007 at 10:59:33PM +0200, Baptiste Coudurier wrote:
>>>
>>>
>>>>Hi
>>>>
>>>>I noticed a weird behaviour (bug ?) when seeking is requested when file
>>>>is streamed:
>>>>
>>>>in url_fseek:
>>>>   } else if(s->is_streamed && !s->write_flag &&
>>>>        offset1 >= 0 && offset1 < (s->buf_end - s->buffer) + (1<<16)){
>>>>       while(s->pos < offset && !s->eof_reached)
>>>>           fill_buffer(s);
>>>>       s->buf_ptr = s->buf_end + offset - s->pos;
>>>>
>>>>if eof_is_reached then s->pos might be < offset, therefore buf_ptr might
>>>>be > buf_end. Next get_buffer will memcpy negative size:
>>>>
>>>>int get_buffer(ByteIOContext *s, unsigned char *buf, int size)
>>>>{
>>>>   int len, size1;
>>>>
>>>>   size1 = size;
>>>>   while (size > 0) {
>>>>       len = s->buf_end - s->buf_ptr;
>>>>       if (len > size)
>>>>           len = size;
>>>>       if (len == 0) {
>>>>	   [...]
>>>>	} else {
>>>>	    memcpy(buf, s->buf_ptr, len);
>>>>           buf += len;
>>>>	[...]
>>>>
>>>>Note that url_fseek clear eof_reached.
>>>>
>>>>That will only happen when reading a streamed file.
>>>>Im submitting the attached patch, but Im not sure what the correct
>>>>behaviour should be. Any suggestion welcome.
>>>
>>>
>>>i think returning some error for attempts to seek after the end of the
>>>file seems appropriate
>>>
>>
>>Ok, patch attached.
> 
> 
> i think this is buggy, that is if the seek is slightly to before EOF
> fill_buffer() could still set EOF as the end of the buffer is further
> ahead
> 

Im not sure I understood the sentence correctly.

According to fill_buffer code, eof_reached will be set only when
URLContext read_packet will return <= 0, so when nothing could be read.

If the seek is close to EOF fill_buffer may read less than
s->buffer_size but still > 0, update s->pos which may still be < offset,
then fill_buffer another time, and maybe set eof_reached this time
though s->pos will still be < offset since nothing could be read, so we
would end with s->pos < offset, so seek failed and reporting error seems
appropriate.

I see that s->eof_reached would still be set to 1, that would not be
false as we are at eof effectively.

What would be buggy exactly ?

-- 
Baptiste COUDURIER                              GnuPG Key Id: 0x5C1ABAAA
SMARTJOG S.A.                                    http://www.smartjog.com
Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
Phone: +33 1 49966312