case should handle hash collision

code now passes test.generative tests. Test run shows case to be faster than cond (and test code could be improved to more directly compare speed).

This is a fairly large patch – please advise if we need to find a way to break into into smaller chunks, or if there are other perf tests you would like to see.

Updated patch to fix above.

Explanation of bug:

When multiple test constants have a hash collision, the respective thens are combined with a condp, and the colliding hash value is used as the test constant. Since the combined then performs its own post-switch validation, the normal (non-colliding) behaviour of checking that the original test value equals the tested expression must be skipped.

The bug occurred because the skip-check set of case ints was not getting the necessary shift-mask applied to it.

The bug was not caught earlier because there was incomplete test coverage, namely, for the case of both hash collision and shift-mask application.

I am seeing failure for the following input:

Sorry for the screwy formatting in JIRA.

FYI: patch applies cleanly & tests pass against master as of 4/21 (2011)

Please Test patch