(j3.2006) Finalization ordering question

[Aleksandar Donev]

On Wednesday 24 October 2007 12:15, Jim Xia wrote:

> Another problem is you're dealing with a non-contiguious array here in
> calling the parent's finalizer.
You are just pointing out that this whole "higher rank finalizer" business is 
ill-designed...I couldn't agree more. As I said, I believe FINAL routines 
should have been treated more closely with type-bound procedures. That would 
have made the dummy polymorphic, so it would effectively always be 
non-contiguous (since you never know how large the dynamic type is anyway). 
The existing design requires the opposite, that the (single) dummy be 
non-polymorphic...

BUT, that said, the only thing that does make sense is for the same thing to 
be done with the parent that is done with the object itself. They are one and 
the same thing but viewed through a different lense. This is a basic OOP 
principle.

Does anyone have an example where non-elemental finalizers are in some way 
better than elemental ones? There is an example in C.1.8 but there is nothing 
there that requires rank-1 stuff. From what Malcolm said it seems there was 
some issue about there not being impure elementals at the time. But why does 
there have to be an elemental procedure at all? Isn't any scalar procedure 
enough---just call it element-by-element.

Aleks