[isidrogar]

I made a quick test using my dxvarenderer branch which is essentially isidrogar's patch with some code reshuffle. Played a DVD iso for 1 hours, free memory was ~2745k when starting XBMC and went to ~2891k after playing for an hour, so does not look like there is a memory leak.

However there is a bug in isidrogar's code that can lead to XBMC freezes during video playback/end of playback. There is no check in CProcessor::Render() to see if the surface buffer contains the frame (or its backward references). See line 1382.

Hi a11599,

I suppose you are talking about the long check after the "or" in this line:

if ((time > m_time) || (time - (m_caps.NumBackwardRefSamples + m_caps.NumForwardRefSamples) * 2 <= m_time - m_size * 2))
{ ... }

This is not a bug. My code intentionally hasn't this check because I think isn't needed:

In case you play a DVD which starts with a still frame not detected properly in DVDVideoDecoder you will have a black screen render (the "black screen" issue discussed previously in this thread). For this case the "Add" method assures the surface buffer has no NULL pointers and you can present the image properly.

In other cases if Render() method tries to present a frame not included in surface array the only issue is the line which calculates index start frame will not select the correct one, but will never go out of bounds (note the operator modulus m_size) nor NULL surfaces (because they are always filled) so in worst case you only could get an invalid VideoProcessBlt() related to presentation time (screen not refreshed, the same behavior as if you return after the check in your code).

In conclusion, include or not the check only has sense for debug purposes but has nothing to do with freezes or similar. Maybe I'm wrong and I'm missing something so in case correct me, but if there is a bug probably will be in other side.