outline of upcoming ap_r* patch

[Greg Stein <gs...@lyra.org>]

There are two problems that I think need to be solved:

1) ap_r* should feed all output immediately into the output chain so that
   proper ordering is preserved across both output mechanisms.
   
   Problem: a bazillion buckets get shoved into the chain and generate poor
            downstream performance (processing, chunking, network output).

2) optimize how ap_r* feeds data into the filter chain.

   Problem: each call to an ap_r* function allocates a brigade and a bucket.
            we need to be able to avoid that.


Solution:

  At the first ap_r* invocation, insert a special filter at the top to
  buffer the ap_r* output. ap_r* will write directly to the filter as long
  as it remains "on top" (if another filter is inserted in front of it, then
  we fall back to a less-optimal scheme so that output can flow through the
  whole chain).



I've got my system up and running with the latest CVS, strace working as
expected, and will work on the patch now.

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/