Differences of Implementations

[Murata Makoto <mura034@attglobal.net>]

Somebody asked differences between RELAX Core V1 implementations and 
TREX implementations.

RELAX Core V1 has three implementations of verifiers.  One (VBRELAX) 
is similar to PyTREX; it backtracks.  The other two implementations 
(RELAX Verifier for C++ and RELAX Verifier for Java) are based on 
non-deterministic bottom-up tree automata; for each element, multiple 
states (or non-terminals or labels) are assigned.

In my understanding, JamesC's implementation of TREX also uses 
non-deterministic bottom-up tree automata.  However, James constructs 
tree automata lazily.  Consequently, his implementation is extremely 
fast when your schema is large but your instance is small.

On the other hand, some constructs of TREX make it hard to actually create tree 
automata in advance.  I am concerned, since creation of tree automata 
is required for type inference (note: XML Query people plan to incorporate 
type inference into their query language).  I am hoping that introduction of 
reasonable restrictions will make it possible to construct tree automata 
in advance.

Cheers,

Makoto