Jeffrey Horner <[EMAIL PROTECTED]> writes: > I'm not sure I like the idea of spooling the files to a tmp file just > to copy them to another tmp file, which will ultimately be copied to > another file by the script writer. Imagine this cost with a 500Mb file.
I understand it's a tricky balance. IMO the right thing for us to do in apreq-dev@ is add some kind of apreq_brigade_link_file() which understands hard links (when available). That's pretty much what we already do in the perl glue, and we could just promote that code to the C API. > My next idea is to create a hook to siphon off all the brigades for > file params and write to the tmp files myself. This way, I'm using > the libapreq api and not presuming what the internals do... > > How's that sound? Certainly doable, but I think it'd be overkill. We deal with these exact same issues (by cheating and peeking at the spool bucket implementation) in the perl glue, so having a common C API makes the most sense to me. -- Joe Schaefer
