On Thu, 14 Nov 2013, Björn Stenberg wrote:

> > Looking at the code the only problem I can see is in singleipconnect()
> > when the default clause gets called in "switch (error)".
>
> This indeed looks wrong, though I fail to provoke the socket leak on any
> of my machines.

Cheers for the patch Björn.

> My attempts at getting access to a sparc machine are still unsuccessful,
> so I'm submitting this patch untested on sparc. It doesn't feel terribly
> good doing debugging via the autobuilders, but on the other hand it
> doesn't feel good leaving the issue unattended either.

This is also my view. I feel the current code is simply wrong and the socket
descriptor should be returned - although you and Daniel know this area of
code better than I ;-)

With that in mind, I have pushed the patch for two reasons a) in an attempt
to fix the Solaris FD leak and b) correct what we both feel is incorrect
code.

The only thing I'm not too sure about and would like to run by you is:
Should we be setting *sockp when the default clause in the switch returns
CURLE_COULDNT_CONNECT?? That doesn't quite feel right to me.

Cheers again

Steve

-------------------------------------------------------------------
List admin: http://cool.haxx.se/list/listinfo/curl-library
Etiquette: http://curl.haxx.se/mail/etiquette.html
Received on 2013-11-14