On Sun, 31 Jan 2010, koettermarkus_at_gmx.de wrote:

> I've had similar problems with the multi_socket API, but I think this
> behaviour goes back beyond 7.19.7.

This bug I noticed probably always existed, it's nothing new.

> If there are multiple handles on the multi, some new timeout from other
> handles would resurrect the stale handle?

Yes, if there are actions on other handles or other timeouts that trigger
after that 1ms they would resurrect the stale handle. That helps disguising
this problem since it's rare that it actually cause real problems.

> I've had the same 'problem' with the multi_socket interface, but the problem
> was a stale handle after successful connect, afaik the handle would be
> 'revived' once other handles got active, and removed for some timeout
> conditions.

The same problem with CALL_PERFORM returned from a timeout call to
multi_socket() can of course happen in several states and I must admit I
haven't really tried to figure out all possible scenarios - I just focused on
the specific case I spotted. Of course there may also be other bugs lurking
that this patch doesn't fix...

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