I saw so much repeated code for parsing brigades, that I created a "readahead" API: ap_brigade_ra(). It is passed similar arguments as those to input filters, and additionally is passed a readahead struct and a readahead limit. This abstraction acts as a buffer and parses out bytes and lines, reading from upstream filters when it needs to.
Okay, I have no idea what this ap_brigade_ra() API is or what its intention is based upon your description. You keep referring to the 'readahead API' as the magic solution for all of the issues I raised. So, I respectfully ask that you please provide this API with examples to this list if you want to continue the discussion.
Please send me a private email if you would like to see the code; the design is there, and so is a lot of the code (too long to post), but it has not yet been tested. I'd have waited until I was a bit further along, but since it changes some behavior, I would like to get the minimal behavior changes into 2.4 if I can. Thanks!
Development and design discussion happens (or should happen!) on this list, not in private emails; so taking this off-list isn't an option: and especially so for filtering and brigade changes. This code needs as much input from as many people as possible. Making 'minimal behavior' changes without clearly understanding what that achieves long-term isn't justified, IMHO. -- justin
