Re: [boost] cast.hpp improvements anyone?

From: David Abrahams (abrahams_at_[hidden])
Date: 2000-07-13 22:16:53

From: "Mark Rodgers" <mark.rodgers_at_[hidden]>

> > These tricks work for GCC
> > 2.95.2, but I've been unable to get them to work with MSVC6.
> ...
> > I think there might be trouble with Borland, though.
>
> Is it MSVC or STLport?  I removed all the #ifdefs so that it
> was using your new method always and I found:

Well, it was MSVC/STLport and GCC 2.95.2/STLport

> 1. It worked fine with both Borland compilers I use, and did
>    indeed eliminate the warnings, although so does a simple
>    #pragma warn -8012

Oh, that's nice! I knew Borland was having some trouble with (partial)
specialization on non-type (bool) template paramters in operators.hpp, so I
suspected it wouldn't work.

> 2. MSVC didn't generate any warnings until I bumped the warning
>    level up to 4, and then your change worked fine and did
>    reduce the cast_test.cpp warnings from 55 down to 47.

Terriffic. I think the difference is that MSVC's native <limits> uses enums
to define static constants, and the compiler is able to use these at compile
time. The STLport uses out-of-line static const bools.
Boris, take note!

> 3. It wouldn't work with GCC 2.95.2 at all due to a lack of
>    <limits>

Try w/STLport. GCC is one of the worst cases, since there's no way to turn
on and off warnings dynamically.

> 4. When I added the following test
>
<snip>
>
>    It failed (didn't throw).  That's because I don't think the
>    last test is correct:
>
>     template <>
>     struct greater_than_type_max<false>
>     {
>         template <class X, class Y>
>         static bool check(X x, Y)
>             { return x >= 0 && static_cast<X>(static_cast<Y>(x)) == x; }
>     };
>
>    This seems to be checking that it didn't overflow rather than
>    that it would.  So the following test also failed

Oh, thanks! I'll reverse the test.

> I always think it is worthwhile to eliminate warnings by doing
> the right thing rather than simply suppressing them with a
> #pragma, so keep it up.

Thanks for the feedback,
Dave