(j3.2006) Did we intend to prevent this?

[Malcolm Cohen]

I wrote:
> That would be extremely painful - we don't want to be doing arithmetic on
> relocatable variable addresses at compile time.

Van claimed:
>What arithmetic?  Fortran doesn't offer syntax for that.

Sure it does.  TRANSFER it to an integer array, then you can play all kinds of 
fun and games!

And after
   INTEGER,PARAMETER :: ia(*) = TRANSFER(dt,[1]) + 1
you can do
   CALL sub(ia)
   ...
   SUBROUTINE sub(idum)
   INTEGER idum(:)
   TYPE(t) dt2
   dt2 = TRANSFER(ia,dt) ! Guaranteed to give you back your original value.

...
>I'm not advocating that variable addresses, or initial data targets'
>values, be full-fledged constants.
>
>Since one can have an initial data target in a variable declaration,
>viz.
>
>  character(len=42), target, save :: A
>  character(len=:), pointer :: B => A
>
>and in a type definition
>
>  type :: t
>    character(len=:), pointer :: P => A
>  end type t
>
>it just seemed inconsistent that one cannot appear as part of a
>structure constructor that is required to be a constant expression, viz.
>
>  type(t), parameter :: Z = t(a)
>
>(or at least two compilers told me this is not allowed).

That's very sensible of them.

>  Except Bob has
>found that according to 7.1.12p1(3)(b), t(a) is a constant expression.

Yes, so we will need an interp to fix the standard.  This one does not look so 
easy.

Cheers,
-- 
................................Malcolm Cohen, Nihon NAG, Tokyo.