Re: [relax-ng-comment] RELAX NG Type Lattices, draft 0

[John Cowan <cowan@mercury.ccil.org>]

MURATA Makoto (FAMILY Given) scripsit:

> I often make the same mistake: definitions without presenting the key idea 
> and motivation in advance.  What is your key idea?

That it's important to have a way to define a lattice of RNG types,
so that people can declare information about them which can be checked
mechanically for consistency.

> Given two regular tree languages A and B, it is possible to determine if A 
> is a subset of B.  In the case of RELAX NG, we further have interleave, text, data, value,
> list, and attribute.  Handling of attribute is done by Hosoya: he gave one neat
> algorithm (see Hosoya and Murata presented at PLAN-X 2002).  

Very good.

> Interleave
> has been studied also: although I am aware of some algorithms, no
> practical ones are publicly available.

I was not aware of this difficulty, but it doesn't astonish me.

> Handling of data, value, and list
> is just difficult.  How can we overcome these problems?

I defined conformance in such a way as to avoid this: data (which
subsumes value) patterns are not required to be handled correctly,
provided only that the lattice validator doesn't claim the lattice
is valid when it isn't.

-- 
"Kill Gorgûn!  Kill orc-folk!           John Cowan
No other words please Wild Men.         jcowan@reutershealth.com
Drive away bad air and darkness         http://www.reutershealth.com
wit h bright iron!"  --Ghân-buri-Ghân   http://www.ccil.org/~cowan