2008/8/22 Adrian Chadd <[EMAIL PROTECTED]>: > G'day again, [snip]
> (Now, for those who are scared - _yes_, they stay stable in > client_side.c[c] in all of Squid-2.6, 2.7 and 2.HEAD as far as I can > tell (and thus they should be in 3.x) - not by API contract, but by > sheer luck. ;) Actually, for 2.6, 2.7 and 3.x it -should- be ok - the copy-into-temporary-buffer-and-write trick should be relatively consistent. This -may- not hold for 2.HEAD as I'm taking a buffer reference and passing a {const char *, int} from that to comm_write() and deref'ing it when the object is destroyed. This may go south if there's a write scheduled but not cancelled implicitly by a comm_close() when the object free's its own cbdata pointer. Gah, this will all make much more sense when proper reference-counted buffers and buffer regions make an appearance. Soon.. Adrian