Re: completeness of the relational lattice

On Jun 22, 12:42 pm, Jan Hidders <hidd..._at_gmail.com> wrote:
> On 22 jun, 20:36, Marshall <marshall.spi..._at_gmail.com> wrote:

[lots of agreement snipped]

> > In simplified terms, you're proposing that we allow the
> > construction of relations with a specific set of attributes
> > and a specific body (namely: empty.) I'm proposing that
> > we only allow the construction of relations with a specific
> > set of attributes. Hence my proposal introduces fewer
> > concepts, therefore it is simpler, and we should by default
> > choose the simplest alternative that gets us what we
> > need.
>
> Let me see if I understand. So you propose a constructm, say [X] with
> X a set of attributes, with the semantics that it returns an arbitrary
> relation with header S.

Ack, no!

> That makes your algebra non-deterministic!

No no no! Abort, abort!

> Do you realize how much problems non-determinism are going
> to introduce in the proofs?

Being someone who has written a lot of concurrent code, I am on a first name basis with nondeterminism. This is not to say that we are friendly with each other; merely that we are well acquainted. No, I am extremely sensitive to issues of nondeterminism, and nothing in the algebra ought to be nondeterministic. In my programming language, I am careful to ensure this is so, and in fact have carefully charted all the possible holes where nondeterminism may leak in. A modest amount of *controlled* nondeterminism in multithreaded code and around I/O is acceptable; in the core algebra, no no no.

Oops, gotta go, more later.

Marshall Received on Fri Jun 22 2007 - 21:59:36 CEST