[issue1471] ioctl request argument broken on 64-bit OpenBSD or OS X

[Gregory P. Smith]

Gregory P. Smith <greg at krypto.org> added the comment:

I am unable to reproduce this problem at all on Mac OS X 10.4 or 10.5
with 32-bit or 64-bit python trunk builds.

I have however checked in a fix that'll prevent negative code values
passed in from being sign extended to 64-bits when converted to an
unsigned long on platforms that use those.

A unittest for fcntl.ioctl behavior when passed both positive and
negative values is included.

trunk r61652

while the Linux man page describe's ioctl's code parameter as an int,
the sys/ioctl.h file on linux has it as an unsigned long int in the
function prototype.  So its quite possible it is actually an unsigned
long on all platforms and we should change unsigned int code to unsigned
long code and the 'I' to a 'k'.

However if that is done we -might- have a sign extension problem again
if anyone is passing a negative value in and expecting it to act like a
32-bit value.   If the OS kernels only look at the lower 32-bits of the
ioctl code during the syscall it'd work, but if it tested the 64bit
value -123 64bit != -123 32bit.  The better answer to that would be to
just reject all negative values and fix any defined ioctl code constants
to be positive on all platforms.

fbvortex.  Can you apply the change from svn to your python and test
that it works properly on your 64-bit OpenBSD system?  Thanks.

paste a note here once you have with the results.

----------
resolution:  -> fixed
status: open -> pending

__________________________________
Tracker <report at bugs.python.org>
<http://bugs.python.org/issue1471>
__________________________________