Lower VLA representation to pointers

[Joseph S. Myers]

On Wed, 11 Aug 2004, Jason Merrill wrote:

> Hmm, what happens if xyzzy is in a loop?  Previously we would properly
> release its memory when it went out of scope.  If we're converting to
> alloca, I suspect that's broken.

The following (execution) test works; the memory is deallocated on leaving 
the block:

void *volatile p;

int
main (void)
{
  int n;
  for (n = 0; n < 1000000; n++) {
    int x[n % 1000 + 1];
    x[0] = 1;
    x[n % 1000] = 2;
    p = x;
  }
  return 0;
}

The following test fails, but also fails in all prior releases that 
support the syntax (3.0.x on), I don't know about old 3.5 CVS:

void *volatile p;

int
main (void)
{
  int n = 0;
 lab:;
  int x[n % 1000 + 1];
  x[0] = 1;
  x[n % 1000] = 2;
  p = x;
  n++;
  if (n < 1000000)
    goto lab;
  return 0;
}

Note that the VLA goes out of scope and should be deallocated when you 
jump to a point before the declaration, unlike non-variable-length objects 
with automatic storage duration which have a constant address as long as 
you stay within their block, on which jumping to a point before the 
declaration has no special effect although the execution of the 
declaration initializes the value of the object or makes it indeterminate 
according to whether there is an initializer.

(C99 6.2.4#5-6 and footnote 27.)

-- 
Joseph S. Myers               http://www.srcf.ucam.org/~jsm28/gcc/
    jsm@polyomino.org.uk (personal mail)
    jsm28@gcc.gnu.org (Bugzilla assignments and CCs)