(j3.2006) (SC22WG5.4589) [ukfortran] Corrigendum 1 ballot

[Malcolm Cohen]

Van Snyder wrote:
>[f08/0023 8.1.6.7 178:8-9] Delete the hyphen in "pointer-assigned".

The editor, and the standard as published, disagrees with this suggestion; that 
is, "pointer-assigned" is used as a verb in several places in F2008, and I do 
not think this is problematic.

>C1284a is a bit larger hammer than necessary, but liberalizing it would
>be messy, so I don't object to it or recommend a change to the
>corrigendum.
>
>At some time we might want to sharpen the constraint to allow
>deallocating a polymorphic allocatable object that is not a dummy
>argument or a subobject of a dummy argument, not in common or a
>subobject of an object in common, and not accessed by host or use
>association or a subobject of such an object.

All these extra conditions have little to do with the problem, it is already 
prohibited to deallocate a pointer in common or accessed by host/use association 
etc.

The problem is that you don't know the dynamic type, so you don't know that it 
is safe to deallocate in a pure procedure.  For example after 
ALLOCATE(localpoly,SOURCE=remotepoly).  So you'd have to also have the condition 
"and could not have been allocated to have a dynamic type that has an impure 
final subroutine", at the very least.  That strikes me as being even less 
satisfactory than C1284a.  We don't want the compilers to have to be proving 
theorems about this.

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