[FFmpeg-devel] [PATCH] probe max read size
Tue Jun 2 06:50:47 CEST 2009
Michael Niedermayer wrote:
> On Mon, Jun 01, 2009 at 03:37:55PM -0700, Baptiste Coudurier wrote:
>> Michael Niedermayer wrote:
>>> On Mon, Jun 01, 2009 at 12:36:54PM -0700, Baptiste Coudurier wrote:
>>>> Michael Niedermayer wrote:
>>>>> On Sun, May 31, 2009 at 02:53:34AM -0700, Baptiste Coudurier wrote:
>>>>>> After some tests, it seems more reasonable to stop probing after some
>>>>>> max size to avoid consuming to much memory.
>>>>> Maybe we then should replace probe_packets by a size based thing,
>>>>> having 2 is uglier than one if one is sufficient
>>>> Sure. Problem is per stream counting is not reasonable since streams
>>>> might not have many packets present this will cause much buffering.
>>> but without per stream counting streams starting once the threashold is
>>> reached should fail to be probed or some streams would only be returned
>>> at eof ...
>> I think that, in memory constrained systems, user will want to limit:
>> - raw_packet_buffer size
>> - probing data size (which is btw still a problem, pd->buf must be
>> released at some point if codec is not probed).
>> To achieve this, I think both sizes must be globally bounded.
>> Maybe by s->probesize or another field, but ideally it must be user
> Either way the value that counts the amount remaining must be in a context
> not on the stack of av_read_*
> also if it is in the context it can be set by the user
>>> If we want a packet limit of PL and a byte limit of BL then
>>> counter= BL
>>> counter -= packet.size + BL/PL;
>>> if(counter > 0)
>>> could be used, per stream, iam not sure if this is a good idea or if
>>> we rather should keep 2 counter
>>> Geometrically, this limit check has an area of a triangle that is within the
>>> limit while the normal 2 variable check uses a rectangle, i feel that a triangle
>>> is supperior ;)
>> Sure, is packet limit still wanted if we have global byte limit though ?
> assuming you have a 1mb size limit and a stream containing gsm
> it wont ever be probebed successfully because we have no gsm probe
> it has 13kbit/sec which means that no packet will be returned to the
> user app for ~10 minutes when then finally all the packets will be
> returned for these 10min
Patch attached. I believe using probesize is a good opportunity and I
plan to replace MAX_READ_SIZE by probesize too, changing probesize
default to 5000000.
I find the FFMAX a bit ugly but I feel it would be safer.
Baptiste COUDURIER GnuPG Key Id: 0x5C1ABAAA
Key fingerprint 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer http://www.ffmpeg.org
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 2292 bytes
Desc: not available
More information about the ffmpeg-devel