steve@vicom.com (Steve Maurer) (11/28/89)
While trying to polish off the ends of a project, I ended
up having to do a work-around for a mutual design deficiency between
our product and the SGI compiler. However, on the work around
itself, I found another, much less forgivable bug in the SGI
compiler; an outright violation of the C standard.
While the compiler allows pointers to functions to be
assigned, it does not allow them to be compared. It also doesn't
allow them to be cast to allow comparison. The following example
illustrates the problem. It was checked out on a number of other
C compilers (Sun being the main one), and it compiled and performs
flawlessly. But not on the SGI.
- - - - - - - - - - - - - - - - - - - - - - - -
void foo(), bar();
main()
{
void (*pfn)();
pfn = foo; /* SGI compiler allows this */
if ( pfn == foo ) /* but barfs on this */
printf("Pfn calls foo\n");
else
printf("Pfn calls bar\n");
}
void
foo()
{
}
void
bar()
{
}
- - - - - - - - - - - - - - - - - - - - - - - -
My question is: does anybody know a work around for this flaw?
Is there a new release fix for it? If that's the case, I need the
new stuff immediately, since the project I'm working on is already way
behind.
Steve Maurer
steve@vicom.comsteve@vicom.com (Steve Maurer) (11/28/89)
Steve Maurer: >- - - - - - - - - - - - - - - - - - - - - - - - > > My question is: does anybody know a work around for this flaw? >Is there a new release fix for it? If that's the case, I need the >new stuff immediately, since the project I'm working on is already way >behind. Never mind. I found a work-around by declaring a dummy procedure rk (for real kludge), which I pass the function pointer to, but declare as an integer argument. Gack. Never thought I'd have to resort to a pascal function pseudo-cast in a C program. Steve Maurer steve@vicom.com
davea@quasar.wpd.sgi.com (David B. Anderson) (11/29/89)
In article <1989Nov27.234730.25669@vicom.com> steve@vicom.COM (Steve Maurer) writes: > > > While trying to polish off the ends of a project, I ended >up having to do a work-around for a mutual design deficiency between ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ >our product and the SGI compiler. However, on the work around ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Please send e-mail on the original compiler problem. I'd like to know what it was so I can fix it. >itself, I found another, much less forgivable bug in the SGI >compiler; an outright violation of the C standard. > > While the compiler allows pointers to functions to be >assigned, it does not allow them to be compared. It also doesn't >allow them to be cast to allow comparison. The following example >illustrates the problem. It was checked out on a number of other >C compilers (Sun being the main one), and it compiled and performs >flawlessly. But not on the SGI. Steve: The problem is that void was not handled properly in released versions of the c compiler. Change the function return types to int and the problem disappears. In the next release (after 3.2) of IRIX void and void * are properly handled. My aplogies for the inconvenience. Regards, [ David B. Anderson Silicon Graphics (415)335-1548 davea@sgi.com ]
mtoy@sgi.com (Michael Toy) (11/29/89)
I compiled and ran the example verbatim from your message, and .... it works fine for me. My guess is that you are running an older release of IRIX. Under 3.2, no such problem exists. Michael Toy
mike@BRL.MIL (Mike Muuss) (11/30/89)
One workaround might be to create a union between the function pointer and a long. Assign into the pointer element, compare the long element. Ugly, but it will get you going. You can probably hide it with some macros if you have to do it a lot. Best, -Mike