------- Comment #1 from fang at csl dot cornell dot edu 2007-04-03 19:45
-------
Funny you mention this, I've done exactly this in my own utility headers:
create forward-declaration-only headers of STL containers and algorithms.
"STL/allocator_fwd.h":
namespace std {
template <class T>
class allocator;
}
"STL/list_fwd.h"
#include "STL/allocator_fwd.h"
namespace std {
template <class T, class A = allocator<T> >
class list;
}
etc...
I've even made typedef templates in some cases:
namespace std {
template <class T>
struct default_vector {
typedef vector<T, allocator<T> > type;
}
}
Another possible advantage of doing this would be to facilitate separate
template compilation of STL containers and their methods *without* relying on
the "extern template" extension. (Has worked for me.) I've never sought to
push this idea, as it might create a minor headache for co-developers who might
suddenly find that including a header that only uses a forward-declared
template class suddenly needs to include the full-definition header on the side
to be able to use it. e.g.
#include <STL/list_fwd.h>
template <class T>
void
some_func(const std::list<T>& l) {
...
}
Uses of the header that don't instantiate some_func will compile fine, whereas
instantiations will require the rest of <list>. I don't necessarily find this
a bad idea, but it can be a source of problems.
FWIW, I use autoconf to detect and conditionally include headers that we're not
supposed to include directly, such as <bits/stringfwd.h>.
#if HAVE_BITS_STRINGFWD_H
#include <bits/stringfwd.h>
#endif
Compilation speedups have been noticeable.
--
fang at csl dot cornell dot edu changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |fang at csl dot cornell dot
| |edu
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31464
