On Fri, Aug 2, 2013 at 1:56 PM, Sam Panzer <[email protected]> wrote on
http://llvm-reviews.chandlerc.com/D405 :
>
> I disabled those checks, then realized that BO_Xor might cause some false
> positives too.
> char c = 0x1ff ^ Y;
> is also fine if Y is between 0x100 and 0x1ff. It's even worse with something
> like
> char c = 0x0f00 ^ Y
> where Y is between 0xf080 and 0xf0ff, since the second byte of Y is known to
> start with a 1. I think that truncation leaves the result the same for signed
> inputs,
> since the result should be between -1 and -256.
>
> Should I also remove the non-Assign Xor check for now?
FWIW, I disagree; if I ever wrote
assert(0x100 <= Y && Y <= 0x1ff);
char c = 0x142 ^ Y;
I would like to be warned about that. The proper way to write that
code would be one of
char c = 0x42 ^ (char) Y;
char c = 0x42 ^ Y;
However, unrelatedly, I think you *should* try to avoid a diagnostic
on compile-time constants featuring "~". For example,
char highbits = 0xffffffe0 | Y; // should warn
char highbits = ~0x1f | Y; // should definitely not warn
my $.02,
–Arthur
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
