The current implementation of case-sensitive matching performs an exact match on
the stemmed search token with the stemmed document token. It remains an
unintuitive property of this behavior that a case-sensitive search for "RUNNING"
will match documents containing "RUNNINg" but will not match documents
containing "rUNNING". However, after discussion, we decided that this seems
less harmful than the problems introduced by the spec'd behavior (as implemented
by Mark in this patch), and we were unable to come up with a solution that
provides sane semantics for all cases.
Our feeling is that there aren't very compelling use cases for case-sensitive
matching in combination with "stem-insensitive" matching, and so what we choose
here shouldn't matter all that much. As such, we don't currently plan to follow
up with any other behavior changes in this area.