time of day timestamps handling in gstreamer (v4l2 source)

[Nicolas Dufresne]

Le 2014-08-27 17:17, Philippe De Muyter a écrit :
> On Wed, Aug 27, 2014 at 12:19:16PM -0400, Nicolas Dufresne wrote:
>> Le mercredi 27 août 2014 à 16:58 +0200, Philippe De Muyter a écrit :
>>> How can a source having known timeofday timestamps (perhaps even old timestamps
>>> as with dv1394 or matroska files) ensure that they are preserved and made
>>> available for the sink ?
>> Would it possible to clarify this question ? Initially you mention
>> v4l2src, other sources behave differently. User can't really assume what
>> is the meaning of a timestamp, but can transform it in one of the known
>> form using the segment. Known forms are stream time, running time, clock
>> time.
> I have two use cases :
>
> - the first one is real-time acquisition with a v4l2 driver producing
> timeofday timestamps.  In addition to my own processing where I must get access
> to the timeofday timestamps, I would like to use standard gstreamer plugins
> to produce video files containing those timestamps.
>
> - I must be able to replay those video files to feed them again to my own
> processing and here also I need to get the original timeofday timestamps.
>
> Thus, of course, I need a container format able to save those timestamps.

I've given a second look at the timestamp algorithm (with the delay) and 
it's in fact correct. The rule has been fixed a long time ago that live 
source produce timestamp that are running time. To get the original 
timestamp back, you would have to force (probably implement) a GstClock 
base on timeofday (see gst_pipeline_set_clock(), currently only 
CLOCK_MONOTONIC, what correct v4l2 driver produce, are supported). Then 
the original timestamp will be timestamp + base_time (see 
gst_element_get_base_time()). This only works for v4l2src and element 
doing the same thing.

Note that to me this is all a bit awkward. I'd would forget about 
getting the v4l2 timestamp, implement and set a UTC/NTP clock on my 
pipleine and save the base_time for each generated streams. This way, 
when I play back the file, I know that the NTP time I want is the 
streaming + ntp-base-time.

What you could do also, which would seem more correct, is to propose a 
GstV4L2Meta that would expose original timestamp and maybe sequence 
number along that. That would mean adding public library to v4l2 
backend. This, though, would be even more specific.

About your request of having a container that store exactly these 
timestamp, you'll have to read containers internal specification and 
implementation. They rarely store time-stamp in 64bit form (sometime 
they only store duration).

Nicolas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/gstreamer-devel/attachments/20140827/2fed5f91/attachment.html>