[Python-Dev] Monterey (64-bit AIX) screws up the poll() implementation, I think

[Trent Mick]

test_poll.py currently fails on Monterey (64-bit AIX) because of a difference
in the system poll() function as compared to Linux.

On Linux (and as required by the Signle Unix Specification:
http://www.opengroup.org/onlinepubs/007908799/xsh/poll.html) if you "poll()"
a bogus file descriptor you get a POLLNVAL return value in the 'revents'
field of the structure sent to poll(). This is tested like so in
test_poll.py:

    # returns NVAL for invalid file descriptor
    FD = 42
    try:
        os.close(FD)
    except OSError:
        pass
    p = select.poll()
    p.register(FD)
    r = p.poll()
    assert r[0] == (FD, select.POLLNVAL)

Monterey decided to return -1 instead. [Aside: However, 'revents's type is
"short" so the Python interface to poll() (using PyInt_FromLong) actually
interprets that as -32768 instead.]

Questions:
1. Can anybody offer an opinion if this is Python's or Monterey's problem?

2. Can anybody tell me where I can browse the POSIX spec to see if this
   breaks POSIX compliance as well?

3. Could someone (Vladimir?) run this test program on a normal AIX box and
   tell me what the output is (does the return value == POLLNVAL?)?

----------------------------------------------------------------------
#include <stdio.h>
#include <poll.h>

#define NFDS 1

int main(void)
{
    struct pollfd ufds[NFDS];
    int pollResult;

    printf("hello\n");

    /* poll for any event on a bogus file descriptor */
    ufds[0].fd = 42;
    ufds[0].events = POLLIN | POLLPRI | POLLOUT;
    pollResult = poll(ufds, NFDS, -1);

    if (pollResult != 1) {
        printf("*** poll result was %d, I expected 1.\n", pollResult);
    }

    printf("ufds[0].revents = %hd\n", ufds[0].revents);
    printf("POLLNVAL = %ld\n", POLLNVAL);

    return 0;
}
----------------------------------------------------------------------


Thanks,
Trent

-- 
Trent Mick
TrentM@ActiveState.com