I'll see if I can figure out something to do based on Nathan's suggestion.
Looking a little closer today, it looks like maybe there is more of a
compiler difference (or -W flag difference) that determines if you see
the warning, rather than a kernel version?
For example, on RHEL4 (x86_64) I do not see the warning. However, if I
add this to the pvfs2-sysint.h header:
#ifdef __KERNEL__
#ifndef INTPTR_MIN
#error unable to find INTPTR_MIN
#endif
#endif
... then it does not compile. So maybe the #if checks don't always
generate a warning if some of the defines are missing? That's kind of
disappointing if so, because it means we have the opportunity to
silently get the size wrong at build time. I guess the
"#if INTPTR_MIN == INT32_MIN"
is always getting processed as true if those values are undefined.
-Phil
Sam Lang wrote:
Hi Phil,
I'm not able to reproduce that warning with a 2.6.11 kernel. I'm not
sure why as the limits.h that gets included doesn't seem to define
INTPTR_MIN either. Could you just send me a patch with Nathan's
proposed fix or something similar that gets rid of those warnings for you?
Thanks,
-sam
On Apr 4, 2006, at 3:30 PM, Phil Carns wrote:
Actually, it does compile now that I look closer (I had a second
problem confusing me), but it does generate quite a few warnings like
this:
pvfs2/include/pvfs2-sysint.h:44:5: warning: "INTPTR_MIN" is not defined
pvfs2/include/pvfs2-sysint.h:44:19: warning: "INT32_MIN" is not defined
Not sure what the ramifications are of the warning, though.
-Phil
Phil Carns wrote:
Is INTPTR_MIN defined in the kernel headers somewhere as well? I am
having a hard time compiling the kerne module at the moment because
/kernel/linux-2.6/pvfs2-utils.c ends up pulling in pvfs2- sysint.h.
I am using the 2.6.15.4 kernel.
-Phil
Sam Lang wrote:
This seems to work everywhere we tried, so I went ahead and
committed that change. Thanks Pete!
-sam
On Apr 4, 2006, at 12:15 PM, Pete Wyckoff wrote:
[EMAIL PROTECTED] wrote on Tue, 04 Apr 2006 11:32 -0500:
Hm...actually I didn't notice before but the use of __WORDSIZE or
BITS_PER_LONG will break on darwin (which doesn't define either).
Previously, I fixed this by defining SIZEOF_LONG_INT and
SIZEOF_VOID_P in pvfs2-config.h, but the PVFS_sys_attr_s struct now
includes padding and that struct is defined in pvfs2-sysint.h, which
is an external header.
Posix defines intptr_t as an int that can hold a pointer, hence you
can get your word size out of that. Then this should work and uses
only constants that POSIX requires in <stdint.h>:
#if INTPTR_MIN == INT32_MIN
32-bit
#else
not 32-bit
#endif
Appears visually to be okay on darwin, x86, and x86_64. But I've
never tried it in real life. Too bad there appears to be no direct
wordsize in posix anywhere.
Another approach would be with unions:
struct PVFS_sys_attr_s {
...
union {
char _s[8];
char *link_target;
};
...
}
But I haven't tried that either. On ancient gcc you need to give
the union a name then "#define link_target _u.link_target", e.g.
-- Pete
_______________________________________________
Pvfs2-developers mailing list
Pvfs2-developers@beowulf-underground.org
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
_______________________________________________
Pvfs2-developers mailing list
Pvfs2-developers@beowulf-underground.org
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
_______________________________________________
Pvfs2-developers mailing list
Pvfs2-developers@beowulf-underground.org
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers