[libcvd-members] Semantics of frame_pending?

[Edward Rosten]

Anyone want to weigh in on the semantics of frame_pending()?


frame_pending was originally used to indicate if there was still a frame 
waiting in the kernel's buffer for V4L2 video capture devices. It extended 
well to DV buffer.

It is sometimes used in the following way:

while(buffer.frame_pending())
{
        //Burn frames
        buffer.put_frame(buffer.get_frame());
}

to reduce latency on live feeds. This doesn't work with the new 
ServerPushJpegBuffer, snice it can't tell if there is a frame available.
Flushing code will simply keep on flushing forver. Note that flushing code 
also breaks on recorded video. It also gets very strange when DiskBuffer 
or VideoFileBuffer have the unset_pending flag set.

Note that ReadAheadVideoBuffer makes the frame_pending() flag work 
properly when used with ServerPushJpegBuffer.


My thoughts so far are to add two new flags, live and flushable.

Flushable implies live, but not the reverse. If flushable is true, then 
the above flushing code will work. If flushable is not true, then 
frame_pending will always be 1. A ReadAheadBuffer then will automatically 
turn a live, nonflushable stream in to a live flishable stream.

So the code would become:

while(buffer.flushable() && buffer.frame_pending())
   ...


Thoughts?

-Ed