[theora] example_encoder.c patch to read successfully from FIFOs

Mike Baynton mike at mbaynton.com
Sat Oct 5 18:41:49 PDT 2013

Some years back I tried to use example_encoder.c with mkfifo FIFO 
buffers to encode content as it was being decoded by mplayer. This 
deadlocks, because to do this you need two FIFOs (one for decompressed 
audio and one for video) that mplayer alternates writing output to, and 
that encoder_example similarly alternates reading chunks of audio and 
video input from. The trouble is, mplayer will fill one of the two 
buffers (say the video FIFO) and still have more to write to that buffer 
before alternating back to the other one (audio), but encoder_example 
will remain in the "reading from the audio input" part of its code, 
blocking on a read from the empty audio buffer, and so will not empty 
any data from the full video buffer until it receives a bit more audio 

This seems like a useful thing to support, as the alternative of first 
dumping complete raw a/v to files on disk and then consuming them with 
encoder_example requires massive disk i/o and space, and does not allow 
decode and encode to run in parallel. Looks like other people have tried 
it in the distant past as well:
  - http://lists.xiph.org/pipermail/theora/2003-February/000335.html
  - http://lists.xiph.org/pipermail/theora-dev/2002-September/001308.html

At the time, I patched encoder_example to basically replace all freads 
to instead read from dynamically allocated internal memory buffers, 
which in turn read from fread, but would automatically cut over to and 
buffer data some data from the other stream when one stream became 
empty. I was then able to encode on-the-fly with encoder_example 
successfully. This was in 2006; I just looked and there have been 
substantial changes to encoder_example.c since then, but none that fix 
this issue. So, I am writing to the mailing list to see if there would 
be interest in incorporating an updated patch if I were to take the time 
to sync and test it with the current code. I'm not sure if this would be 
seen as beyond the scope of an example encoder...but I personally think 
it'd be quite useful.

Mike Baynton

More information about the theora mailing list