[Bug middle-end/50079] [4.7 Regression] FAIL: g++.dg/init/copy7.C execution test

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50079

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING
   Target Milestone|---                         |4.7.0
      Known to fail|                            |3.4.0

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-08-15 
10:40:58 UTC ---
As I said, please re-check after

2011-08-12  Richard Guenther  <rguent...@suse.de>

        * call.c (build_over_call): Instead of memcpy use an
        assignment of two MEM_REFs.

Please also verify if the C testcase

extern void abort();
extern void *
memcpy(void *dest, void *src, __SIZE_TYPE__ n)
{
  if (dest == src)
    abort();
  else
    {
      __SIZE_TYPE__ i;
      for (i = 0; i < n; i++)
        ((char *)dest)[i] = ((const char*)src)[i];
    }
}

struct A
{
  double d[10];
};

struct B
{
  struct A a;
  char bc;
};

struct B b;

void f(struct B *a1, struct B* a2)
{
  *a1 = *a2;
}

int main()
{
  f(&b,&b);
}

fails the same way (it would then even before the patch exposing the C++
testsuite failure).

I believe the testcase is broken, in that GCCs behavior of using memcpy
for block-moves dates back to date 1.

C testcase that also abort()s on x86_64 since at least 3.4.0:

extern void abort();
extern void *
memcpy(void *dest, void *src, __SIZE_TYPE__ n)
{
  if (dest == src)
    abort();
  else
    {
      __SIZE_TYPE__ i;
      for (i = 0; i < n; i++)
        ((char *)dest)[i] = ((const char*)src)[i];
    }
}

struct A
{
  double d[1024];
};

struct B
{
  struct A a;
  char bc;
} __attribute__((packed));

struct B b;

void f(struct B *a1, struct B* a2)
{
  *a1 = *a2;
}

int main()
{
  f(&b,&b);
}