In 6th NWPT, page 269
ALF is based on Martin-Löf's type theory, which is a logical framework in the sense that the language can be extended with new definitions. The language is a functional language with dependent function types, which can be extended with data types, functions and constants as new definitions, in the spirit of traditional functional languages. The dependent function type gives us a richer type system, in which propositions and specifications of programs can be represented. Thus, the relation can denote a program of type or a proof of proposition . This relation is decidable and the type (proof) checking algorithm is the core of the proof editor ALF. However, the objective of ALF is to interactively build proof terms or programs, in a flexible and safe way. For this purpose we have extended the language with the notion of placeholders (meta variables) which denotes sub-terms intended to be filled in. Hence, an incomplete proof term is a term containing placeholders, and proof refinement corresponds to successively replacing a placeholder by a (possibly incomplete) sub-term.
The type checking algorithm is extended to handle incomplete terms which gives as result a unification problem, i.e. a set of equations and typing restrictions of the placeholders occurring in the incomplete term. This set corresponds to the requirements which must be satisfied in future refinements. The unification problem can not always be solved, due to higher order placeholders, and unsolved equations are left as constraints restricting future instantiations of the placeholders. Correctness of a refinement is established by type checking the refined (incomplete) proof term. Thus, proof refinement is reduced to type checking incomplete proof terms.
Chalmers Technical University, Inst. for Computer Science, Göteborg, Sweden. E-mail: firstname.lastname@example.org.
Available as PostScript,