If we restrict our concern to validation, non-determinism is not a problem at all even if automata are used internally. The validator only have to maintain a SET of states. (This is what RELAX Core validators and XDuce do.) In the case of Jing, it maintains a CHOICE of expressions to be matched.
However, if you consider data binding, non-determinism does cause a problem. This is because we cannot uniquely determine which part in the schema matches the current element from time to time. Should we standardize non-determinism conditions so that data binding becomes easier? I have been against this idea, since different data-binding engines require different types of non-determinism. Moreover, datatype ambiguities as shown in Eric's book is hard to detect. Cheers, -- MURATA Makoto (FAMILY Given) <[email protected]>
