The magic set technique is a standard technique
for query evaluation in deductive databases,
and its variants are also used in modern commercial database systems like~DB2.
Numerous improvements of the basic technique have been proposed.
However, each of these optimizations makes the transformation
more complicated,
and combining them in a single system is at least difficult.

In this paper, a new transformation is introduced,
which is based on partial evaluation
of a bottom-up meta-interpreter for SLD-resolution.
In spite of its simplicity,
this technique gives us a whole bunch of optimizations for free:
For instance,
it contains a tail recursion optimization,
it transforms non-recursive into non-recursive programs,
it can pass arbitary conditions on the parameters to called predicates,
and it saves the join necessary to get subquery results back into the
calling context.
In this way,
it helps to integrate many of the previous efforts.

The usefulness of these optimizations is illustrated
with example programs querying the World Wide Web.