On Fri, Sep 25, 2009 at 12:56:45PM -0700, Constantine Sapuntzakis wrote:
> Attached is a merged version of hostthre.c. I've also cleaned up the
> macros to be functions  and the types to be prefixed with curl_. I
> haven't tested it on a Mac, so it probably doesn't compile but I think
> it's close.

Hey, this is nice. Much cleaner and nicer than my implementation (mainly
because I was shooting for wrapping pthreads around the existing logic,
which depended on Windows-specific threading quirks. Some notes:

- curl_thread_t_null should be used in curl_thread_join for windows,
  instead of a literal NULL
- The definition of *tsd in gethostbyname_complete looks truly evil.
  I think that's worth splitting out. (Also getaddrinfo_complete)
- You've reverted to itoa() in getaddrinfo_thread. Please use snprintf
  with NI_MAXSERV - this is more portable. itoa() doesn't exist in
  POSIX.

   char service[NI_MAXSERV];
   snprintf(service, sizeof(service), "%d", tsd->port);
- The "shouldn't be reached" bit in Curl_is_resolved should probably be
  a DEBUGASSERT as well.

I'll probably have a couple more tweaks to make once I get around to
testing this on my Mac and Linux machines.

> This does not address non thread-safe resolvers. That's probably best
> dealt with in the configure mechanism.

I can write a follow up patch for that in a bit. Jamie's advice seems
sound.

Do we have any sense of what we're going to do about the dummy_sock
stuff? Is there a better way?

-- 
Joshua Kwan
-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette:  http://curl.haxx.se/mail/etiquette.html