On Sep 4, 2008, at 12:14 PM, Eli Friedman wrote:
>> + /// BlockVoidPointer - The assignment is between a closure
>> pointer and
>> + /// void*, we accept for now.
>
> I don't see any good reason to accept this; it'll just destroy the
> block pointer.
void * is the canonical opaque type, and block pointers can be pushed
into and pulled out of a void * per the usual C rules.
> Just to double-check, the comparison is roughly equivalent to memcmp
> for block pointers?
Yes, I think so.
> The code as-is won't work correctly for cases like
> "int a(int(^b)(void)){return b == (void*)0}".
gcc does this:
$ cat t.c
int a(int(^b)(void)) {return b == (void*)0; }
mrs $ gcc -S t.c -fblocks -fomit-frame-pointer -O9
mrs $ cat t.s
.text
.align 4,0x90
.globl _a
_a:
xorl %eax, %eax
cmpl $0, 4(%esp)
sete %al
ret
.subsections_via_symbols
clang should do something similar when done.
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
