Tree-walking (was: index-of() behavior)

[Florent Georges]

Michael Kay wrote:

> To get close to emulating apply-templates, I think you need
> a function M that maps node-names to functions T, so you pass
> M to your tree-walking function W, and W when it reaches a node
> N calls M to obtain the function T corresponding to the node-name
> of N, and then calls T. (M=mode, T=template).

  Very interesting.  Except that I would define M as mapping a node (not a node name) to functions, to emulate match patterns.  But I guess a general, user-defined function to map from nodes to functions would be less than optimal.  I guess I am too much XSLT-oriented, but a set of pairs of match patterns (as defined by XSLT) and function objects sounds reasonable to me:

    M1 := (("pattern1", T1),
           ("pattern2", T2),
           ("pattern3", T3),
           ...)

    M2 := (("pattern-a", Ta),
           ("pattern-b", Tb),
           ("pattern-c", Tc),
           ...)

  One obvious advantage of what you describe is that W is not fixed, and could be defined by the user or by a library, and be reused, instead of having to write it again and again as in XSLT.

  I thought a few weeks ago that it would be interesting to expose patterns as first-class objects, even in XSLT.  It seems this would be an interesting application.

  Regards,

-- 
Florent Georges
http://www.fgeorges.org/