[Testcases,committed] [Was Re: C++ PATCH: PR 11852]

[Andrew Pinski]

On Sunday, Sep 7, 2003, at 16:46 US/Pacific, Mark Mitchell wrote:

>
> This PR demonstrated the presence of totally bogus code in
> initializer_constant_valid_p; for a CONSTRUCTOR, only the last element
> of the CONSTRUCTOR was being checked.
>
> Bootstrapped and tested on i686-pc-linux-gnu, applied on the mainline
> and on the branch.
>
> --
> Mark Mitchell
> CodeSourcery, LLC
> mark@codesourcery.com
>
> 2003-09-07  Mark Mitchell  <mark@codesourcery.com>
>
> 	PR c++/11852
> 	* varasm.c (initializer_constant_valid_p): Correct logic for
> 	CONSTRUCTORs.
>
> 2003-09-07  Mark Mitchell  <mark@codesourcery.com>
>
> 	PR c++/11852
> 	* g++.dg/init/struct1.C: New test.
>

This patch also fixes PR middle-end/11665 so I committed the testcases 
for them on both the mainline and the 3.3 branch:
ChangeLog:
2003-09-07  Andrew Pinski  <pinskia@physics.uc.edu>

         PR middle-end/11665
         * gcc.c-torture/compile/20030907-1.c: New test.
         * g++.dg/init/array11.C: New test.

gcc.c-torture/compile/20030907-1.c, g++.dg/init/array11.C
(yes they are both the same, but it should run though both front-ends):
/* PR 11665
    Orgin: jwhite@cse.unl.edu
    The problem was in initializer_constant_valid_p,
    "for a CONSTRUCTOR, only the last element
    of the CONSTRUCTOR was being checked"
    (from the email of the patch which fixed this).
    This used to ICE because GCC thought gdt_table was a
    constant value when it is not.  */

int x;
struct gdt
{
unsigned a,b,c,d,e,f;
};
void f()
{
struct gdt gdt_table[2]=
{
     {
                 0,
                 ( (((unsigned)(&x))<<(24))&(-1<<(8)) ),
     },
};
}