[mythtv] Sluggish pause, waiting for a keyframe?

[Ed Wildgoose]

Doug Larrick wrote:

> Doug Larrick wrote:
>
>> A little more playing around reveals that pause now appears to wait 
>> for or reposition to a keyframe.  For material that has come in over 
>> ATSC, this is not good news, as the keyframes can be several seconds 
>> apart (typically 2-3, but I've seen as much as 10).
>
>
> Isaac fixed the jumping around when entering pause (thanks, Isaac), 
> but I still get a multi-second delay sometimes between pressing pause 
> and it actually pausing.
>
> A little debugging has led to the conclusion that NVP::Pause is 
> waiting a long time to acquire the (newish) decoder_lock.  I haven't 
> fully examined the code yet (gotta go to work), but it seems the loop 
> in NVP::StartPlaying holds this lock most of the time.


Having a lock debug flag would be extremely nice.  I have a couple of 
delays that I would like to debug including the channel change speed 
(which I am sure that I can at least half if I could just track down my 
bottleneck)

How easy is it to add some code into the lock create/drop code so that 
we could switch on a trace flag if required and just see the code point 
and exact time that locks are acquired and dropped.  This is obviously a 
time critical function in many cases, but perhaps there is an efficient 
way to generate a "MythMutex" which can optionally spew debugging info...?

Crazy idea?

Ed W