* Gets rid of a lot of unnecessary synchronization
  * Fix known bug #64 by running addrinfo_callback in the same thread as
multi_perform/socket_action
  * Ask caller to select on dummy socket for read instead of write -- this
actually blocks the process
  * Set timeouts with exponential back-off so caller doesn't block for too
long or spin too much

The timeout stuff is the part that makes me the most nervous. There is only
one timeout per curl handle - I checked and I believe the code isn't
clobbering an existing one but....

-Costa

Attachment: hostthre.diff
Description: Binary data

Reply via email to