Re: [boost] Problems with the May 9th version of function

From: Douglas Gregor (gregod_at_[hidden])
Date: 2001-05-13 10:11:48

On Saturday 12 May 2001 08:48 pm, you wrote:
> I've tried out the latest (May 9th) version of the archive for
> function.hpp. Here's what I did when I tried to compile it.
>
> 1.  I had to comment out the "template" qualifications when a template that
> came from within a class was being referenced.  I use Metrowerks
> CodeWarrior Pro 5.3, which is a generation behind the latest (6.1).  I
> don't know if CWP6.1 has the same problem.  If it doesn't, should we bother
> coming up with a workaround?

None of the other CodeWarrior folks have mentioned such a problem, so it 
seems specific to 5.3. I'm not sure if we should support older versions of a 
compiler, given the large number of hacks and #ifdef's already required.

> 2.  A new problem emerged due to the "nil_t" type.  An object of this type
> can be converted to any (data) pointer, via a conversion operator that is a

T can be bound to a function type, so it works for both data and function 
pointers.

> member template.  CWP5.3 cannot handle such operators, even though it can
> handle non-template-based operators and other member template functions.  I
> don't think even CWP6.1 fixes it, so we need a workaround strategy.  This
> could include a BOOST_NO_CONVERSION_OPERATOR_TEMPLATES #define in
> <boost/config.hpp>.

In truth, function doesn't even use that conversion, so commenting it out 
wouldn't hard function. However, nil is intended as a general replacement for 
the usage of NULL and 0, so it needs to be able to convert to any pointer 
type. Unfortunately, I can't think of any possible workaround, because we 
can't go back and add an operator= for pointers :(.

If this problem is still around for more compilers, we have a real problem 
:). The only workaround for a compile not supporting/using the templated 
conversion operator is the evil:

#define nil 0
typedef const int nil_t;

This won't always work, unfortunately :(

I hope CW6.1 doesn't have this problem...

        Doug