------- Comment #7 from fang at csl dot cornell dot edu 2007-03-08 23:41
-------
Subject: Re: Overloaded operator delete[] doesn't get called
> Above, in operator new[], If WTF is false, returning NULL, I reproduce the
> same error (missing call to class operator delete []). If WTF is true
> (returning pool), the right operator delete [] is called.
>
> // output with WTF false
> class new[] operator
>
> // output with WTF true
> class new[] operator
> class delete[] operator
>
> Tested on: powerpc-apple-darwin8-g++-4.0.1
The same result ALSO happens if I de-inline the class operator new/delete
[], and compile their definitions in a different translation unit:
------------->8 snip 8<-------------------
// de-inlined class definitions
#include <iostream>
#include "one_array_only.h"
using std::cout;
bool
one_array_only::allocated = false;
one_array_only
one_array_only::pool[256];
void*
one_array_only::operator new[] (std::size_t size) throw() {
cout << "class new[] operator\n";
#if WTF
return pool; // does not trigger bug
#else
return NULL; // triggers bug: no operator delete[] called
#endif
}
void
one_array_only::operator delete[] (void *p) throw() {
cout << "class delete[] operator\n";
}
------------->8 snip 8<-------------------
The main translation unit is still calling one_array_only::operator
delete[](void*) (from an asm dump).
This suggests that one_array_only::operator delete[] is somehow being
mis-compiled.
Fang
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29164
