On Apr 7, 11:52, Daniel Stenberg wrote:
>On Wed, 7 Apr 2004, Christophe Neumair wrote:
>
>> It would seem that AIX don't like nonblocking
>> sockets.
>> Then, I just run the configure with
>> --disable-nonblocking option and all request
>> are a complete success :)
>
>If I were you, I'd contact someone about this. It sounds like a serious flaw.
>Perhaps there's already a patch available for your OS that fixes this problem.
>
>libcurl won't be able to time-out properly in lots of places when you disable
>non-blocking sockets.

We don't have a test for this in the test suite I presume? For otherwise
this would have been tested thoroughly by now I believe :-)

That there should be a nonblocking-socket-problem in AIX sounds a bit
suspicious to me though, and I would like to see some example code that
shows this. We haven't seen any such problems in the systems we've
made for AIX 4.3 and 5.1 (we = I and the compnay I work for).
In fact I'm right now at a customer's site with a room full of IBM
SP2 node racks, all running software with nonblocking sockets somewhere
in there.. I'm not saying that there are no such problems, but I'm not
seeing any in our applications, at least.

(There are certainly *some* AIX-specific problems one should be aware of
though, the POSIX semaphores, for example, exist in the AIX 5.1 runtime
library and header files, but they're all just fake functions.
Your application will compile just fine, but will fail in runtime if
your're synchronizing your threads through posix semaphores.
For some reason IBM thought this non-solution was 'cleaner' than not
providing the header- and libc definitions at all. Go figure.)

-Tor
Received on 2004-04-07