-ff2c question: Complex type in procedure calls (related to PR 30887 / %VAL())

[Tobias Schlüter]

Brooks Moses wrote:
> Tobias SchlÃŒter wrote:
>> Tobias Burnus wrote:
>>> In the test case, I have:
>>> external c_to_c8
>>> complex(8) u8, v8, w8, c_to_c8
>>
>> No, you have not :-)  You wrote complex*8, which gets the following 
>> special treatment in gfc_match_old_kind_spec:
>>
>>    /* Massage the kind numbers for complex types.  */
>>    if (ts->type == BT_COMPLEX)
>>      {
>>        if (ts->kind % 2)
>>     {
>>       gfc_error ("Old-style type declaration %s*%d not supported at %C",
>>              gfc_basic_typename (ts->type), original_kind);
>>       return MATCH_ERROR;
>>     }
>>        ts->kind /= 2;
>>      }
> 
> Wow, that's an ugly syntax hack.  We don't have that one documented, do we?
> 
> Honestly, I think that's counterintuitive enough that it ought to get a 
> warning at anything less than -std=legacy, besides being documented, but 
> maybe that's just me.

It's completely intuitive :-) a variable defined in the legacy "type*n 
x" syntax, is a variable of type "type" with a storage size of "n" 
bytes.  People using complex*n instead of complex(n) deserve to be 
punished.  This includes Tobias B.  Just kidding.

I agree that this should be documented.  I'm not so sure about a 
warning.  The syntax is widely used.

- Tobi