GMP compile failure

[Marc Glisse]

On Fri, 28 Dec 2007, Tim Hockin wrote:

> This (simplified) C++ snippet does not compile.  But it doesn't give
> me the possibly-expected undefined operator error:
>
>                mpz_class foo(123);
>                mpz_class bar = foo & 0x107;
[...]
> /usr/include/gmpxx.h:3413: error: invalid conversion from 'const long
> int' to 'const __mpz_struct*'
> /usr/include/gmpxx.h:3413: error:   initializing argument 3 of 'static
> void __gmp_binary_and::eval(__mpz_struct*, const __mpz_struct*, const
> __mpz_struct*)'

Whereas just "foo & 0x107;" compiles fine.

> I can make it build by explicitly casting the int to mpz_class, but
> that's kind of a pain.  I am converting a codebase to use GMP, and
> this is causing lots of errors.

You can also modify struct __gmp_binary_and in gmpxx.h to (untested):

struct __gmp_binary_and
{
   static void eval(mpz_ptr z, mpz_srcptr w, mpz_srcptr v)
   { mpz_and(z, w, v); }
   static void eval(mpz_ptr z, mpz_srcptr w, unsigned long int i)
   {
       mpz_t temp;
       mpz_init_set_ui(temp,i);
       mpz_and(z, w, temp);
       mpz_clear(temp);
   }
   static void eval(mpz_ptr z, unsigned long int i, mpz_srcptr w)
   {
       mpz_t temp;
       mpz_init_set_ui(temp,i);
       mpz_and(z, temp, w);
       mpz_clear(temp);
   }
};

But if you want to distribute your code, that is not a solution.

> Is there any answer to this, other than "cast it"?

I am not familiar enough with the code to know the right answer.

> why such a cryptic error?

Getting non-cryptic error messages in C++ is... hard. Here the code goes 
through many inline functions to try and avoid any useless copy and the 
type problem is detected only in the middle.

-- 
Marc Glisse