Re: [PATCH] PR19100: truthvalue_conversion vs. TREE_OVERFLOW (take3)

[Mark Mitchell <mark at codesourcery dot com>]

Roger Sayle wrote:
> The following patch is my third attempt at resolving PR middle-end/19100
> which is a wrong code bug affecting both C and C++ on mainline.  I still
> believe that the correct fix is to clean-up truthvalue_conversion, however
> I also agree that reducing/eliminating TREE_OVERFLOW from GCC is a good
> thing, so the patch below humours Mark's request to investigate a fix in
> the middle-end.  [It did however require two fixes to int_fits_type_p :)]
>
> 	* fold-const.c (fold_convert_const): Don't set TREE_OVERFLOW when
> 	converting constants from one integer type to another.

Is this OK with the Ada/Java folks?  It certainly makes sense to me from 
a C/C++ perspective.

> 	* c-common.c (convert_and_check): Avoid testing TREE_OVERFLOW to
> 	detect conversion overflow, instead use int_fits_type_p.

I'll let Joseph comment on this bit, but it seems plausible to me.

Thank you for trying to hard to fix this bug.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304