On Sat, 2008-04-12 at 01:09 +0200, Henrik Nordstrom wrote: > When would one want to (in the protocols we deal with) cancel a write > without closing the fd?
Never! It is not a protocol issue. There was a case in the pre-async code where I wanted to cancel a write just before closing the fd because I did not want comm to call my write handler "recursively" (reentering the transaction) with an ERR_CLOSING or some such. It was not possible to avoid that call and I had to work around that. I now realize that with async calls in place, this should not be an issue because the writer can just cancel the callback and because the call will not be reentrant anyway. Thanks, Alex.
