Re: Ada 0Y plans for garbage collection?

[Dmitry A. Kazakov <mailbox@dmitry-kazakov.de>]

On Thu, 11 Sep 2003 21:21:21 GMT, "Robert I. Eachus"
<rieachus@attbi.com> wrote:

>>>I still think the way it was done in Modula-3 gives you the best of
>>>both worlds. If you want garbage collection, you allocate from a
>>>different pool. If you don't want it you don't use it, and there is no
>>>overhead.
>> 
>> Well, you can have this allready at least with GNAT. See
>> http://www.ada.krischik.com for details.
>
>Exactly.  And if you read the documentation there, you eventually 
>conclude that a garbage collected pool for Ada implementations may seem 
>like a good thing, but it creates more issues than it solves.  It can be 
>done, and it has been done.  But in practice building reference counts 
>into types which need garbage collection gives you much better control 
>over finalization.

Absolutely!

(I saw no single real-life example, where automatic GC were a solution
of choice)

>And yes, I know that there are general graph structures where reference 
>counts don't work.  I also know that in practice most such general 
>graphs can be constructed with structural pointers and general pointers 
>so that the structural pointers define a tree and nodes not in the tree 
>should be reclaimed. You could say that doing that is a lot of extra work.

Actually, it is not so much work. I did it many times. 

---------
But reference counting could be made much more easy and attractive if
smart pointers have more light-weighted implementation than controlled
types in generic packages. There are too many problems with that:

1. Controlled types are slow.

2. They have by-reference semantics, which is a source of nasty
pitfalls.

3. They are not transparent as true access types are. So you have to
define a plethora of proxy methods and take a great care to maintain
them when the pointed type gets changed. It is awful and unsafe.

4. Derived types are not supported. That is, there is no way to make a
smart pointer to a derived type class a member of the class of smart
pointers pointing to the base type class. One need to support two
parallel type hierarchies which is absolutely impossible with
generics.

---
Regards,
Dmitry Kazakov
www.dmitry-kazakov.de