Re: Alternative algorithm to restrict use of data/string in patterns.

[James Clark <jjc@jclark.com>]

This looks good.  A couple of minor comments:

> To capture the tri-state "element"/"text"/"anyString", I'd like to
> introduce "fpdsType".
>
> One important operation over fpdsType is "OR" operation. That is,
> fpdsType||fpdsType -> fpdsType.
>
> This operation is commutative: fpds1|fpds2 == fpds2|fpds1.
>
> Definition of OR of fpdsType
> ----------------------------
> X|Y == Y|X
> X|X == X
>
> "text"|X              == "text"
> "anyString"|"element" == "anyString"
>
> If pattern p can start with transition "element"(X) and "text"(Y), then
> what matters is "text"(X|Y). Or operation is used to extract what
> matters.

If we consider the three states to be ordered:

element < anyString < text

Then the "OR" operation is just max. Right?

> So to correctly calculate what kind of transitions can be possible,
> _calcFPDS has to receive fpdsType that succeeds pattern p.

Would another way to do it be to have calcFPDS just take s single argument
and for group to do

if (p is <group>p1 p2</group>) {
   if (isNullable(p1))
      return calcFPDS(p1) || calcFPDS(p2);
   else
      return calcFPDS(p1);
}

where isNullable(p) checks whether p can match the empty sequence (and any
atts)? (I'm not saying this is easier.  I just want to check my
understanding.)

James