Duplication in source code is a widespread phenomenon that increases
program size and complexity, and makes program maintenance more
difficult. A solution to this problem is to detect clones (instances
of copied code) and to eliminate them. Elimination works by extracting
the cloned code into a separate new procedure, and replacing each
clone by a call to this procedure. Several automatic approaches to
detecting clones have been reported in the literature. In this paper
we address the issue of automatically extracting a previously detected
group of clones into a separate procedure.  We present an algorithm
that can extract "difficult" groups of clones, and a study that shows
that difficult clone groups arise frequently in practice, and that
our algorithm handles them well.