[llvm-commits] [llvm-gcc-4.2] r110043 - /llvm-gcc-4.2/trunk/gcc/objc/objc-act.c

[Stuart Hastings]

On Aug 2, 2010, at 10:12 AM, Stuart Hastings wrote:

> Author: stuart
> Date: Mon Aug  2 12:12:51 2010
> New Revision: 110043
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=110043&view=rev
> Log:
> Issue error instead of ICE.  Radar 8246180.
> 
> Modified:
>    llvm-gcc-4.2/trunk/gcc/objc/objc-act.c
> 
> Modified: llvm-gcc-4.2/trunk/gcc/objc/objc-act.c
> URL: http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/gcc/objc/objc-act.c?rev=110043&r1=110042&r2=110043&view=diff
> ==============================================================================
> --- llvm-gcc-4.2/trunk/gcc/objc/objc-act.c (original)
> +++ llvm-gcc-4.2/trunk/gcc/objc/objc-act.c Mon Aug  2 12:12:51 2010
> @@ -2003,7 +2003,17 @@
> objc_build_compound_setter_call (tree receiver, tree prop_ident, tree rhs)
> {
>   tree temp, bind, comma_exp;
> -  if (TREE_SIDE_EFFECTS (rhs))
> +  /* LLVM LOCAL begin 8246180 */
> +#ifdef OBJCPLUS
> +  if (TYPE_NEEDS_CONSTRUCTING (TREE_TYPE(rhs)))
> +    error("setting a C++ non-POD object value is not implemented - assign the value to a temporary and use the temporary.");
> +#endif
> +  if (TREE_SIDE_EFFECTS (rhs)
> +#ifdef OBJCPLUS
> +      || TYPE_NEEDS_CONSTRUCTING (TREE_TYPE(rhs))
> +#endif
> +  /* LLVM LOCAL end 8246180 */
> +      )
>     {
>       /* To allow for correct property assignment semantics
>          and in accordance with C99 rules we generate: type temp;
> @@ -2021,10 +2031,7 @@
>       {
> 	tree type = TREE_TYPE (rhs);
> 	if (TYPE_NEEDS_CONSTRUCTING (type))
> -          {
> -	    comma_exp = temp;
> -	    error("setting a C++ non-POD object value is not implemented - assign the value to a temporary and use the temporary.");
> -	  }
> +          comma_exp = temp;
> 	else
>       	  comma_exp = build_modify_expr (temp, NOP_EXPR, rhs);
>       }


I have a test case for this, but I'm not sure how to mark it for the test suite.  The correct behavior for llvm-gcc is to issue the diagnostic (above), but clang should compile it without error (it does).

Is there a way for a test case to XFAIL itself only for a particular front-end (e.g. llvm-gcc versus clang) ?

Thanks,

stuart