Preprocessor keyword #echo is missing

[Martin Sebor]

Zack Weinberg wrote:
> Martin Sebor <sebor@roguewave.com> writes:
> 
> 
>>A #echo at the compiler level rather than at the preprocessor level
>>would, IMHO, be extremely useful in the context of metaprogramming.
>>At the very least, it would provide a framework for the proposed
>>implementation of the static assert feature:
>>
>>http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/papers/2002/n1381.htm
> 
> 
> That is an #error analog, not #echo. I do consider it a useful
> feature, but if I were implementing it I'd just take the spec straight
> from N1381 (with appropriate modifications for C) -- don't see the
> need for a static_echo().

Yes, static_assert is equivalent to a #error. I was suggesting
a more ambitious facility, one that would subsume static_assert,
i.e., one that could (but wouldn't be limited to) halt compilation
based on a compile-time predicate, but that could also print
text messages during compilation.

E.g., something along these (half baked) lines

     template <int N>
     struct factorial {
         enum { value = N * factorial<N - 1> };
     };

     template <>
     struct factorial<0> { enum { value = 1 }; };

     template <int N>
     struct print_int {
         #echo "the value is %d", N
     };

     template struct print_int<factorial<7>::value>;

> 
> I wonder why the string literal is included there, when it isn't for
> the runtime assert().

To issue more meaningful and better readable diagnostics
that would be possible otherwise. The libstdc++ concept
checking facility is a good example of how hard to read
the diagnostics can be and how much effort it takes to
make them at least somewhat comprehensible.

Regards
Martin