From: Ed Brey (brey_at_[hidden])
Date: 2001-03-02 17:22:10
From: "Trevor Perrin" <tperrin_at_[hidden]>
> A couple questions about the workaround:
>
> - what does "deletion with std::nothrow" mean?
A case like "new (std::nothrow) int", where the expression evaluates to 0
instead of throwing if there is an error allocating the memory.
> - also, if I add only the non-array delete:
>
> inline void operator delete(const void* p) throw() {operator
> delete(const_cast<void*>(p));}
>
> but then do:
>
> const int* y = new int[10];
> delete[] y;
>
> the non-array version above still gets called (if I add the array
version
> as well then it gets called appropriately); I believe it's okay to delete
> arrays with the non-array version in VC (so the above I think works), but
it
> still seems incorrect function resolution.
I've verified by stepping through the code that the array and non-array
versions of operator delete end up calling the same code, so it happens to
work. Relying on this fact isn't a problem, since the buggy compiler puts
you into compiler-specific hack-land in the first place. Compiler
work-arounds are allowed to take advantage of non-portable implementation
details (of course, avoiding them if fine too, if possible). As I recall,
there is't a non-const operator delete[] to call, so when I implemented the
workaround for myself I had to rely on the detail that VC6 happens to
implement scalar and array delete the same.
> Anyways, if there's a procedure for submitting bugs/workarounds please
> point me at it (you mentioned a VC Known Bug page, which I'd be very
> interested in seeing!); thanks.
http://www.boost.org/more/microsoft_vcpp.html
Note that at this time, this page doesn't cover the problems and workarounds
regarding new and delete expressions/operators. (Also note that there are a
couple other new/delete expression related VC6 bugs that I skipped
mentioning, since they are more obscure and tend not to be triggered in
conjunction with boost libraries.)