Unifying task & Distributed Annex

Unifying task & Distributed Annex

[mamboking]

One of the things I really like about Erlang is that whether you're
dealing with local tasks or distributed ones, the syntax is basically
the same.  I think unifying Tasks & the Distributed Annex would be
great.  I would like to see everything just go the way of Tasks, local
and distributed.  Maybe the distributed Tasks could be limited to only
using rendezvous.  That way you could avoid having to deal with shared
state.

Just a thought.  Anyone else have any ideas about this?


Kevin Hostelley

Re: Unifying task & Distributed Annex

[Jeffrey R. Carter]

mamboking wrote:
> One of the things I really like about Erlang is that whether you're
> dealing with local tasks or distributed ones, the syntax is basically
> the same.  I think unifying Tasks & the Distributed Annex would be
> great.  I would like to see everything just go the way of Tasks, local
> and distributed.  Maybe the distributed Tasks could be limited to only
> using rendezvous.  That way you could avoid having to deal with shared
> state.
> 
> Just a thought.  Anyone else have any ideas about this?

Well, sure. In 1994. I thought the way to address distribution would be to have 
a way to assign tasks to the different machines, with appropriate restrictions. 
I also thought that formalizing the concept of passive tasks present in several 
Ada-83 compilers would be better than inventing a new construct with a new 
reserved word and hard-to-enforce restrictions.

-- 
Jeff Carter
"You tiny-brained wipers of other people's bottoms!"
Monty Python & the Holy Grail
18

Re: Unifying task & Distributed Annex

[tmoran]

> I think unifying Tasks & the Distributed Annex would be great.

They are different things.  You could write a normal program and easily
change it to a distributed program without ever using the pragma
Asynchronous.  That would let different CPUs, with different disks
containing different data files, run different parts of your
single-tasking program.  Thus a database read-only server partition
would likely not have any asynchronous procedures.  Using asynchronous
procedures in a distributed system is a low-level approach to
multi-tasking, foregoing all the help of Ada "task" handling.
For instance, the effect of
  select
    Operator_Call("Put out the fire");
  or
    delay 60.0;
    Call_Fire_Brigade;
  end select;
would need to be explicitly programmed if Operator_Call was an
asynchronous procedure in a remote partition, instead of an entry
in a task.