Doing it better 1999-02-08 h
###

From Reuben.

> > Typing allows you to separate a unity and add efficiency, e.g. my
> > primitive ports/threads can be specialised into objects, processes,
> > channels &c. via typing.
>
> Ah, typing of code! I'll think about it.

I wasn't thinking about typing of code, I was thinking of typing of ports,
which are objects, but the typing constraints would certainly have
partially to be enforced by rewriting the scheduler, so it is partly
typing of code, yes.

> > I'm just going to use malloc. Later the heap can be written into the
> > system.
>
> You are deceived by a high-level language already! malloc must use more
> than 1K of code...?

That's shared libraries. Tau is going, as I've said before, to be
parasitic, so a small footprint means reusing what's already there. In any
case, a "native" heap manager would be a similar size. I'm talking about
the size of the tau binary, and the memory it'll use on top of what's
already there.

> > Yes. But again, my model of communication is much simpler than yours. It
> > can be specialised and made more efficient later.
>
> You don't have one, last I heared.

Yes I do. You can send messages to ports. What this means is up to the
scheduler. The serial scheduler views it as a subroutine call (basically,
threaded code) whereas the parallel scheduler will do primitive thread
switching. If you want to implement more complex communication then fine,
you do it by using ports which forward/duplicate/(de-)multiplex messages
&c.

> The root of our disagreement may be that I consider an external subroutine
> call to be a communication primitive.

What I said above seems to agree.

> are you going to do about my kind of communication primitives?

I've told you now, I think. I have a single notion of sending a message to
an address, which is interpreted in the context of a particular scheduler.

> I like the idea too. But more abstract bases are easier to reason about,
> not more basic ones.

Well, mine's more abstract too.

> How's the design going, anyway. Presumably you're not spending too much
> time on it...?

No. But I think it's not too far off coding time...all my notes are on one
side of A4 at the moment, which is a good sign, and I don't think I
currently have many problems to resolve, so if I copy it out again and it
still makes sense I'll start coding.

--
http://www.cl.cam.ac.uk/users/rrt1001/ | maxim, n.  wisdom for fools