The Forum for Discussion about The Third Manifesto and Related Matters

Please or Register to create posts and topics.

Heading inference, not type inference

Quote from Dave Voorhis on September 22, 2020, 2:22 pm
Quote from dandl on September 22, 2020, 1:20 pm
Quote from Dave Voorhis on September 22, 2020, 8:53 am
Quote from dandl on September 22, 2020, 5:52 am
Quote from johnwcowan on September 22, 2020, 4:09 am
Quote from dandl on September 22, 2020, 3:48 am

TTM is a proposal for a type system, around which a programming language is to be built. In my proposal, the RA is supported by a single relation type and a single tuple type.

Okay, so your proposal is a different type system from TTM: by all means.  (In which case, why are we discussing it here?  But no matter, I'm not a stickler for on-topic.)

My proposal is for a modest change to TTM, to achieve exactly the same end goals but with a type system that is more compatible with existing languages.

At, if I recall correctly, the expense of weakening type-safety.

Also, defining a heading as a set of strings, etc., is a particular implementation idea only appropriate to certain (rather limited) implementation target languages.

It shouldn't be a change to the model, which is sound.

As far as I can see, the weaknesses you have in mind are in the implementation based on common existing languages, not in the model. And yes, it would be better to refer to the heading as a 'set of attribute names' as per TTM, rather than specifically a set of strings. The proposed model is sound.

I mean the TTM model's heading is sound. It needs no changes.

And I mean that my proposed variant model is sound, with the added advantage that it can be implemented within existing type systems and with only minor changes to existing language specs and compilers.

Andl - A New Database Language - andl.org