> I'm wondering how much added code and complexity non-blocking support
> would actually entail? An extra 2-5%, maybe? Couldn't this be supported
> as a compile-time option when building axTLS?
>
Cam started an axTLS mailing list on SourceForge at Dan Stenberg's request (http://sourceforge.net/mailarchive/forum.php?forum_name=axtls-general). Perhaps the issue could be raised there.

> We've used libcurl+axTLS for several years, and I don't recall ever
> seeing a failure due to this. In fact, when we use HTTPS, it is used
> as a logon transaction which is forwarded by the server to another
> authentication server, and if the credentials are invalid, the
> authentication server won't respond for 30 seconds, and when it does,
> the HTTPS successfully completes, without any peep from axTLS.
>
It could be that your application code's structure is more robust. Polling the connection, for instance, would probably expose this issue much more readily than waiting for an event on the socket or returning from "select" with a sufficiently long timeout.

Eric

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2011-01-12