Re: Type safety on wikipedia

["Randy Brukardt" <randy@rrsoftware.com>]

"Simon Wright" <simon@pushface.org> wrote in message
news:m23bj9h35k.fsf@grendel.local...
...
> One point I had forgotten is that deallocation of a controlled object
> will finalize it -- LRM 13.11.2(9) -- so looking at it through a
> dangling pointer will be a Bad Thing even if GC has ensured that the
> mamory hasn't been re-used for something else.

Which reminds me of a significant issue: Ada 95 doesn't allow early (or
late) finalization of limited controlled objects. (Whether it allows early
finalization of nonlimited controlled objects is open to debate.) That means
that GC cannot collect objects containing limited controlled components
"early". But since a program written with GC in mind won't use
Unchecked_Deallocation, and the access types will likely be library level,
"early" and "anytime before completion of main" are the same. Since I think
most objects in a program ought to be controlled, that's a significant
problem.

We talked about this a few times, but never came near a consensus on how to
fix this, so Ada 2005 inherits the issue intact. I suppose if anyone was
seriously trying to use GC in their implementation, the priority level would
go up...

                                    Randy.