Right now, what's the most important for the success and adoption of D?

[Daniel Keep]

Patrick Kreft wrote:
> Am 29.09.2007, 16:03 Uhr, schrieb Daniel Keep
> I know so long exist an instance of MyClass so her resource will exist
> too, or the GC is useless cuz he make man class useless.

Yes.

> Or i can not use that:
>  class MyClass
>  {
>      OtherClass a;
>      this() {
>         a = new OtherClass;
>      }
> 
>      void doit()
>      {
>     a.call() // scasm: opps a was deleted from GC Error eh?
>      }

I really hope that "scasm" isn't meant to be "sarcasm".  That line is
fine, but...

>      ~this() {
>         delete a;
>      }

This is wrong for exactly the reasons I stated before.

The problem is that you *cannot guarantee what order objects are
destroyed it*.  This means that by the time "delete a" is called, "a"
may have already been destroyed.

To put it in more concrete terms, the GC is free to destroy objects
(assuming the instance of MyClass is called "this") either:

1. "this" followed by "a", which means ~this works just fine, *or*
2. "a" followed by "this", which will probably cause a segfault.

You *cannot* tell which of those two will happen.

In general, you *must not* attempt to delete, read or write fields or
call member functions of objects inside a destructor.

As downs said, it sucks.

>  }
> 
> Look an http://www.digitalmars.com/d/garbage.html
> the GC use Reference counting that mean so long a is not noll, it will
> never delete.

Um, *no*.  The D garbage collector does *not* use reference counting; I
have no idea where you got that from.  The D garbage collector uses a
conservative mark-and-sweep algorithm, which means when a collection
occurs, it scans memory starting at the roots (which are the call stack
and global data).

	-- Daniel