Possible future extensions to auto-completion mechanism

We have found that when we compose⁴ predicates from primitive predicates it is possible, at compile-time, to perform an analysis of the rules which will never be triggered. These rules can then be flagged and simply ignored in further compositions involving this predicate. Such an optimisation will remove a lot of redundant code and make the mechanism more efficient. For example in the decision tree example in Section 5, 50% of the auto-completion rules of the un-flagged predicate are never triggered.

We hope to improve upon the algorithm mentioned in Section 4.5 by doing a dependency analysis at compile-time and use that information while iterating over the list of predicates.

There are many instances of HTML forms where a set of values can be predicted for a field. This set can be used to predict other field values (possibly, a set of values). Clearly, this requires a more complex mechanism(than one suggested in this report) and might require additional specifications but certainly it will involve extending the current work. We hope to investigate that in the future.