Re: [xml-dev] functional databases noodling ...

[Michael Kay <mike@saxonica.com>]

On 14/02/2012 08:57, James Fuller wrote:
> every year at XML Prague, Michael Kay either via his talks or
> questions makes a single comment that spawns a few hundred hours of
> research on my part ... this year was his minor mention of 'functional
> databases' ... which sent me scrambling.
The functional database system that impressed me the most was IRIS:

http://www.hpl.hp.com/techreports/90/HPL-90-108.pdf

The data model for IRIS consists of objects, functions, and types. What 
you think of as attributes and relationships in other models are 
represented as functions (a function [=method] of an object that yields 
either an atomic value or another object). "Stored" and "derived" data 
are modelled symmetrically as functions (extensional functions or 
intensional functions). Types are objects, and relationships among types 
as well as relationships between types and instances are modelled in the 
same way as other relationships, i.e. as functions. A particularly nice 
feature is that objects can have multiple types and can change their 
type - so an employee can become a pensioner with no change of identity. 
Functional composition gives you effects similar to relational views, 
but in a much cleaner way. A query can be considered an inverse 
function: if there's a function from person to name, then it has an 
inverse which is a function from name to person, which gives you the 
basis of a query.
>
> a few thoughts immediately came to mind:
>
>       * how valid is it to say 'xquery is a functional
> programming/query language' ? I think 3.0 makes this statement true
Yes, I think in 3.0 we have all the ingredients of a functional 
programming language.
>
>       * is it possible that we could rename xml databases (who have
> xquery) into 'functional database' especially if they employ XQuery
> 3.0
If you look at the IRIS paper quoted, then no; because a key theme of 
functional databases was to take relational normalization a step further 
(creating something very like RDF triple stores) and reduce all 
information to a single canonical form by analyzing functional 
dependencies. An XML database will never achieve that because XML offers 
so many different ways of saying the same thing.
>
>       * does anyone on the list has experience with pure functional
> databases in existence
>
>
They'll need to have grey hair, because the fashions moved on. But I saw 
some great work done in the 1980s with functional database in areas like 
criminal intelligence gathering - the kind of areas where people are now 
trying to use RDF, for exactly the same reasons.

Michael Kay
Saxonica