Re: [PATCH] Fix EQUIVALENCE vs. SAVE (PR fortran/18518)

[Brooks Moses <bmoses at stanford dot edu>]

Steve Kargl wrote:
> I'm beginning to dislike Chapter 14 of F95 (or Chapter 16 of F2003).
>
> 14.7.6    Events that cause variables to become undefined
> Variables become undefined as follows:
>       (1)   When a variable of a given type becomes defined, all
>             associated variables of different type become undefined. ...
>
> subroutine t(y)
>
>    real x
>    integer i
>    save x,i
>    equivalence(x,i)
>
>    interface
>       subroutine foo(i, x, y)
>          integer, intent(in) :: i
>          real, intent(in) :: x
>          real, intent(inout) :: y
>       end subroutine foo
>    end interface
>    interface
>       subroutine bar(i, x, y)
>          integer, intent(in) :: i
>          real, intent(in) :: x
>          real, intent(inout) :: y
>       end subroutine bar
>    end interface
>
>    if (y < 0.) then
>       i = 0
>       !
>       ! x is undefined?
>       !
>       call foo(i,x,y)
>    else
>       x = 1.
>       !
>       ! i is undefined?
>       !
>       call bar(i,x,y)
>    end if
> end
>
> Is this conforming?  That is, can an undefined variable
> be used as an INTENT(IN) function argument.  If it is
> conforming, does it make sense? 

I can't find any reason why it wouldn't be standard-conforming, though 
proving negatives is difficult, and I can't see any place where the 
standard states it positively.

What it does state (in section 12.5.2.1) is that non-INTENT(IN) dummy 
arguments can be referenced if the associated actual argument is defined 
-- which implies the possiblity that it could be undefined.  And in 
looking at all the mentions INTENT(IN), I see nothing specific about 
requiring definition.

Thus, I would claim that this is conforming.

It also does, I believe, make sense -- so long as foo only internally 
references i, not x, and bar only internally references x, not i.  (In 
which case, I'm not sure why you'd be passing them as arguments in any 
real-life situation, but that's irrelevant to the question.)

In particular, I don't see why it is in principle any different from 
this, with regards to undefined dummy arguments:

program a
  real :: b, c
  c = 2
  d = f(b,c)
contains
  function f(x,y)
    real, intent(in) :: x,y
    f = y**2
  end function
end

- Brooks