#10045 HIGH 1.5-sof: XO-1.5 Record audio/video are out of sync with each other

[Zarro Boogs per Child]

#10045: XO-1.5 Record audio/video are out of sync with each other
---------------------------------------+------------------------------------
           Reporter:  wad              |       Owner:  dsd                              
               Type:  defect           |      Status:  new                              
           Priority:  high             |   Milestone:  1.5-software-update              
          Component:  record-activity  |     Version:  Development build as of this date
         Resolution:                   |    Keywords:  camera record XO-1.5             
        Next_action:  diagnose         |    Verified:  0                                
Deployment_affected:                   |   Blockedby:                                   
           Blocking:                   |  
---------------------------------------+------------------------------------

Comment(by dsd):

 So, having looked at the last 2 comments above, is the conclusion that
 writing to /tmp is the solution to A/V sync issues, or at least to
 decrease it by a few hundred ms?

 (It's of course understandable that it decreases sync issues over time,
 given that we know our SD card goes slow which is v likely to cause buffer
 overruns, but it's really surprising that it helps sync *at the start*
 because I checked, the buffers do not overflow, so the backing media
 should not make any difference...)

 I started taking Jon's changes and putting them in official SL git.
  * quality selection changes -- this isn't applicable for the SL activity
 because it'll break on cameras that have different resolutions. Instead of
 hardcoding the info, the activity should parse the caps of the v4l2src and
 offer a selection of resolutions based on what is supported by the camera.
 if the code is rewritten in this form we can take that change upstream
  * sync=False change for video preview: applied
  * queue in video bin readded (albeit in more sensible place)
  * queues in transcoding pipelines not taken (should not be necessary,
 since data on an open file handle doesn't expire, and also didnt result in
 any difference in performance in my quick test)
  * do-timestamp change not taken since I don't really know what it does -
 is there some good documentation/reference somewhere that justifies it, or
 has someone confirmed that it makes a visible improvement to
 synchronization in a side-by-side test?
  * /tmp writing: not taken yet as it looks like this is still cooking, but
 I'll be happy to take this later

-- 
Ticket URL: <http://dev.laptop.org/ticket/10045#comment:35>
One Laptop Per Child <http://laptop.org/>
OLPC bug tracking system