Partial Evaluation[Neil D. Jones, Carsten K. Gomard, and Peter Sestoft, Partial Evaluation and Automatic Program Generation, Prentice-Hall, 1993; ACM Computing Surveys, vol. 28, no. 3 (Sept. 1996), pp. 480-503.]

Partial evaluation specializes a function with respect to arguments that have known values. Given a program P(x, y) where the values of variables x are constant, a specializing function mix transforms P(x, y) → P_x(y) such that P(x, y) = P_x(y) for all inputs y. P_x(y) may be shorter and faster than P(x, y) . We call x static data and y dynamic data.

Partial evaluation involves:

• precomputing constant expressions involving x,
• propagating constant values,
• unfolding or specializing recursive calls,
• reducing symbolic expressions such as x * 1, x * 0, x + 0, (if true S₁ S₂).

A good rule of thumb is that an interpreted program takes ten times as long to execute as the equivalent compiled program. Partial evaluation removes interpretation by increasing the binding between a program and its execution environment.