Re: Abortable Timed Action

[Bob Duff <bobduff@theworld.com>]

Simon Wright <simon@pushface.org> writes:

> GNAT certainly used to have issues in this area. Indeed, you could abort
> the task and then deallocate it, and end with a memory leak (the task
> control block wasn't actually freed); the cure was to wait until
> 'Terminated.
>
> I believe that this is no longer a problem.

When you call Unchecked_Deallocation on an object containing tasks that
are not yet terminated, they keep running.  GNAT used to have a memory
leak as you say.  I think it was fixed perhaps 2 years ago.  So with
the latest GNAT, when you call Unchecked_Deallocation, the memory is
freed for tasks that are already terminated.  Not-yet-terminated tasks
keep running, and the memory gets freed when they terminate.

It is a language design flaw that you can't block a task until some
other task(s) are terminated -- other than at scope exit, which is
usually way too late.  You can busy wait:

    abort T;
    -- Now T is aborting itself, running finalization actions and whatnot.
    while not T'Terminated loop
        delay X;
    end loop;
    -- Here, we know that T is terminated.

But X is guaranteed to be too long or too short -- most likely both.  ;-)

- Bob