I'm not sure that the following really is a bug, but i definitly think so.

The code below does not compile.
------------------
int main()
{
int a = 2;
int[5] b;
int[] c = new int[b[a]];

return 0;
}
----------------
t.d(5): need size of rightmost array, not type [a]b


But if i insert some paranthesis around the b[a] argument i give for the allocation size of c, it compiles perfectly.
----------------
int main()
{
int a = 2;
int[5] b;
int[] c = new int[(b[a])];

return 0;
}



It seemse like the D-typechecker dosen't evaluate b[a] to an int in the first case, perhaps that's completly correct, but i can't see why...

/Robin

Hmm, it seems the ( ) solve many problems ( bugs? ) , like the opCall() for the Streams Mik was working on.  Just in time for 1.0 :D.

C

On Mon, 1 Mar 2004 18:21:17 +0000 (UTC), Robin Håkanson <Robin_member@pathlink.com> wrote:

> I'm not sure that the following really is a bug, but i definitly think so.
>
> The code below does not compile.
> ------------------
> int main()
> {
> int a = 2;
> int[5] b;
> int[] c = new int[b[a]];	
>
> return 0;
> }
> ----------------
> t.d(5): need size of rightmost array, not type [a]b
>
>
> But if i insert some paranthesis around the b[a] argument i give for the
> allocation size of c, it compiles perfectly.
> ----------------
> int main()
> {
> int a = 2;
> int[5] b;
> int[] c = new int[(b[a])];	
>
> return 0;
> }
>
>
>
> It seemse like the D-typechecker dosen't evaluate b[a] to an int in the first
> case, perhaps that's completly correct, but i can't see why...
>
> /Robin
>
>



-- 
Using M2, Opera's revolutionary e-mail client: http://www.opera.com/m2/

Definitely looks like a bug. In parse.c the parsing of "new" looks like
it confuses int[b[a]] with an associative array - possibly by thinking
it was parsing a declaration in parseBasicType2... I can't really tell.
You can replace "b[a]" with any expression like 0+b[a] or b[a]+0
or as you found (b[a]) and it'll get out the the jam.

-Ben

Can't the compiler know in this case that b (thus b[2]) is uninitialized, and be rejecting the "new int[really=0]" as undefined?  Then when parens wrap it the compiler knows it will have a value (type int)...  Well, the message is a bit strange, but is probably trying to say "type member of array of ints", whereas the () wrapper will force evaluation to some int value.  Convoluted?

<snip>
>> The code below does not compile.
>> ------------------
>> int main()
>> {
>> int a = 2;
>> int[5] b;
>> int[] c = new int[b[a]];  // ok if b[a] is wrapped as (b[a])
>
>>
>> return 0;
>> }
>> ----------------
>> t.d(5): need size of rightmost array, not type [a]b
>>
</snip>