Here's my first diff: http://www.creative.net.au/diffs/20060926-squid3-p1.diff
The aim is to: * cut back on calls to strlen() which aren't needed in the client_side request parser path * rework the client_side parser to not require a seperately-malloc()ed 'prefix' buffer during the header parsing stage I think I've identified one of the reasons the current http parser wouldn't function well as any kind of "incremental" parser. I've documented it with a comment. I'll commit this tomorrow night if noone objects or finds a way to break it. Adrian
