Abstract
The narrowing driven partial evaluation (NPE) [Vid96, AFV98, AV02]
is a technique for the specialization of functional and functional logic
programs. The original technique follows the online approach of partial
evaluators, i.e., it performs the propagation of static values and the termination
controls during the proper specialization process. On the other
hand, the offline schemes usually proceed in two distinct phases: the
binding time analysis, which propagates the static parameter values and
determines the control strategy to be employed by adding annotations
into the subject program, then the proper specialization phase, just obeys
the introduced annotations. Offline partial evaluators are usually more
efficient--but less precise--than online partial evaluators.

The aim of this thesis is to improve the approach to offline narrowing
driven partial evaluation of [Ram07]. In particular, the annotation procedure
is enhanced and a more effective partial evaluator is developed in
order to get better specialized and faster programs. As applications, we
carry out the specialization of interpreters which, in agreement with the
first Futamura projection [Fut71], is considered like a kind of program
compilation. This application represents the first approach to the compilation
by partial evaluation of functional logic programs. Additionally,
we consider the specialization of a domain specific language whose goal
is to generate programs for computerized numerical control of machines
(CNC).

From a theoretical point of view, the basic offline technique for higher
order functional programs is extended through an approach which is
known as defunctionalization. Furthermore, a polivariant transformation
that improves the accuracy of specialization process is applied without
to implement a real polivariant scheme.

The resulting scheme has a good balance between efficiency and precision,
thus improving previous proposals and providing a good starting
point for the development of effective specialization tools for functional
logical programs.