Re: HP/UX 11 test_socket failure

> I've never been good at ifdefs, but I parse internal_setblocking in
> socketmodule.c as executing:
> 
>   #else /* !PYOS_OS2 */
>       delay_flag = fcntl(s->sock_fd, F_GETFL, 0);
>       if (block)
>           delay_flag &= (~O_NDELAY);
>       else
>           delay_flag |= O_NDELAY;
>       fcntl(s->sock_fd, F_SETFL, delay_flag);
>   #endif /* !PYOS_OS2 */
> 
> for HP/UX. However, man recv says:
> 
>   If the O_NDELAY flag is set using fcntl() (defined in <sys/fcntl.h>
>   and explained in fcntl(2) and fcntl(5)), nonblocking I/O is enabled.
> 
>   In this case, the recv() request completes in one of three ways:
> 
>     ·  If there is enough data available to satisfy the entire request,
>        recv() completes successfully, having read all the data, and
>        returns the number of bytes read.
> 
>     ·  If there is not enough data available to satisfy the entire
>        request, recv() completes successfully, having read as much data
>        as possible, and returns the number of bytes it was able to read.
> 
>     ·  If there is no data available, recv() completes successfully,
>        having read no data, and returns 0.
> 
> whereas it says:
> 
>   If the O_NONBLOCK flag is set using fcntl() (defined in <sys/fcntl.h>
>   and explained in fcntl(2) and fcntl(5)), POSIX-style nonblocking I/O
>   is enabled.  In this case, the recv() request completes in one of
>   three ways:
> 
>     ·  If there is enough data available to satisfy the entire request,
>        recv() completes successfully, having read all the data, and
>        returns the number of bytes read.
> 
>     ·  If there is not enough data available to satisfy the entire
>        request, recv() completes successfully, having read as much data
>        as possible, and returns the number of bytes it was able to read.
> 
>     ·  If there is no data available, recv() completes, having read no
>        data, and returns -1 with errno set to [EAGAIN].
> 
> ... which seems to be what recv expects (function sock_recv):
> 
>     n = recv(s->sock_fd, PyString_AS_STRING(buf), len, flags);
>     Py_END_ALLOW_THREADS
> 
>     if (n < 0) {
>             Py_DECREF(buf);
>             return s->errorhandler();
>     }
> 
> I'll patch and try this tomorrow, unless someone else does it before
> me.

Good sleuthing!  I didn't know there were platforms where O_NDELAY and
O_NONBLOCK differed. :-(  On Linux, they're two names for the same flag.

Looking forward to your patch.

--Guido van Rossum (home page: http://www.python.org/~guido/)