Re: Refactoring the Type System

On 11/12/2010 07:07 PM, Jeff Davis wrote:
> On Fri, 2010-11-12 at 08:34 -0800, David Fetter wrote:
>> Folks,
>>
>> For the past couple of years, I've been hearing from the PostGIS
>> people among others that our type system just isn't flexible enough
>> for their needs.  It's really starting to show its age, or possibly
>> design compromises that seemed reasonable a decade or more ago, but
>> are less so now.
>>
>> To that end, I've put up a page on the wiki that includes a list of
>> issues to be addressed.  It's intended to be changed, possibly
>> completely.
>>
>> http://wiki.postgresql.org/wiki/Refactor_Type_System
>>
>> What might the next version of the type system look like?
> This problems (as stated) strikes me as pretty overwhelming.
>
> 1. As far as I can tell, we have the best type system of any SQL DBMS.
> 2. Getting a type system right is a hard problem by itself, and there
> isn't any obvious consensus (although I think there is some agreement on
> some issues).
> 3. Type systems are more challenging for a DBMS because you need to
> account for things like storage, indexing, and optimization in ways that
> programming languages don't (consider: when comparing an int4 and an
> int8, you may want to coerce based on what indexes you have available).
> 4. SQL standard issues. In particular, I think that any modern type
> system will run into pretty severe problems with NULLs in one way or
> another. I think we'd have to pay very close attention to the standard
> when designing a new type system, because I suspect that retrofitting
> the standard onto a system we invent independently would be a disaster.
> 5. Backwards compatibility issues.
>
> I think the best we'll do is be able to hack on some of the things that
> we actively want and have clear use cases for, such as type interfaces.
> We might have to give up on some of the more ambitious ideas that
> involve propagating interesting information through the type inference
> system; or having any real type that wasn't declared with CREATE TYPE.
> Consider that right now we bundle the element type information along
> with the array _value_.

Yeah, composites too, IIRC. It's a bit sad. But those are really just 
warts show the difficulties we face in implementing types. I'm still 
waiting for some seriously different yet possible thing we could do. 
(And I agree we do have about the best type system around).

cheers

andrew