Re: [PATCH] Do not sanitize left shifts for -fwrapv (PR68418)

[Jeff Law <law at redhat dot com>]

On 12/04/2015 01:48 PM, Paolo Bonzini wrote:
> > > > gcc:
> > > > 	PR sanitizer/68418
> > > > 	* c-family/c-ubsan.c (ubsan_instrument_shift): Disable
> > > > 	sanitization of left shifts for wrapping signed types as well.
> > > >
> > > > gcc/testsuite:
> > > > 	PR sanitizer/68418
> > > > 	* gcc.dg/ubsan/c99-wrapv-shift-1.c,
> > > > 	gcc.dg/ubsan/c99-wrapv-shift-2.c: New testcases.
> > Doesn't this change how pointer types are handled?
>
> Why would pointer types be shifted at all (at the ubsan level,
> which is basically the AST)?
It's not really a question of why, it's a change in behaviour. 
Previously this code would emit instrumentation objects of pointer type 
if pointers are signed on the target.  After your change it will not, in 
fact, it may trigger a checking failure.

So you'd have to argue that we don't care about sanitization of these 
operations on pointers and verify that we don't trigger a checking 
failure.  I'm really not the best judge of whether or not we want to 
instrument pointer shifts -- they're not terribly useful in general, but 
I'm always [un]pleasantly surprised at what people actually do.


Jeff