Re: valarray_copy implementation question

[Christian BRUEL <christian dot bruel at st dot com>]

OK, as expected the attached reduced case reports a read error with 
valgrind :

==10504== Invalid read of size 1
==10504==    at 0x400D3E: void 
std::__valarray_copy<int>(std::_Array<int>, unsigned long, 
std::_Array<int>, std::_Array<bool>) (in /home/bruel/tmp/a.out)
==10504==    by 0x400ADB: 
std::mask_array<int>::operator=(std::valarray<int> const&) const (in 
/home/bruel/tmp/a.out)
==10504==    by 0x4007E0: main (in /home/bruel/tmp/a.out)
==10504==  Address 0x4e2a091 is 0 bytes after a block of size 1 alloc'd

compiled with g++ -O2 x86_64-unknown-linux-gnu 3.3.3

and goes into an infinite loop on SH4 with a gcc 4.4

Cheers

Christian

Paolo Carlini wrote:
> Gabriel Dos Reis wrote:
> > I did not consider the interpretation put forward by Christian, although I can
> > see his point.  I suspect I have to  double-check what Fortran
> > does in this case and consider whether it actually is a LWG issue.
> >   
> Thanks Gaby for your help.
> > Do you know what other implementations do?
> >   
> I can check another one. If you can pass me a precise testcase, then I
> can give you the data point.
>
> Paolo.
// Test __valarray_copy.  ISO/IEC 14882:1998, 26.3.8.2 mask_array assignment.

// { dg-do run }

#include <valarray>

int main(void)
{
  int n = 1;

  std::valarray<int>  dest (n);   // 0
  std::valarray<bool> b (n);      // false
  std::valarray<int>  src (1, n); // 1

  dest[b] = src; // 0

  if (dest.size () != n)
    return 1;

  return 0;
}