Re: Task discriminants

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

"Martin Dowie" <martin.dowie@btopenworld.com> wrote in message
news:c8luk4$29b$1@hercules.btinternet.com...
> "Randy Brukardt" <randy@rrsoftware.com> wrote in message
> news:YIadnaE4uYUXxjPd4p2dnA@megapath.net...
> > OTOH, it's fairly hard to imagine why an implementation would do
anything
> > other than use a loop to do it, which would of course give the right
> answer.
> > So practically, it probably would work right.
>
> Given this last statement, is it worth changing the RM to make this a
> Confirmation or Implementation Advice or some such?

Well, we'd need a compelling reason to make a change here, because it's
possible that some implementors do use this rule to some advantage. The only
time it matters is when you have side effects during elaboration, which is
generally discouraged anyway.

To solve the problem at hand, I'd suggest the technique that Robert Eachus
demonstrated. It doesn't depend on a majic function or order of evaluation.
Yes, it serializes the initialization of the tasks, but so does the
elaboration mechanism.

                      Randy.