Re: [Valgrind-developers] Changing the definition of HReg

[Greg P. <gp...@ap...>]

On Feb 4, 2013, at 9:45 AM, Tom Hughes <to...@co...> wrote:
> On 04/02/13 17:25, Patrick J. LoPresti wrote:
>> Never access an object via the wrong pointer type.  The only exception
>> is "char *", which can alias anything by definition.  (You can also
>> use a union to access the same memory as different types.  I believe
>> this takes you into the realm of implementation-defined behavior.
>> That is also bad, in my opinion, but it is not the disaster of
>> undefined behavior.)
> 
> The C standard says you can only read a union through the member it was 
> last written through. My guess would be that doing otherwise is invokes 
> undefined behaviour, but you could be right that is is "just" 
> implementation defined.

C99 fixed this. Type punning through a union is now officially supported. (The result depends on the implementation-defined bit representation of the objects of course, but I expect valgrind already requires the implementation to be two's-complement integers and IEEE float.)

valgrind probably performs formally-undefined aliasing and type punning everywhere anyway, which is why it builds with -fno-strict-aliasing.



-- 
Greg Parker     gp...@ap...     Runtime Wrangler