ralsei / system-f-redex

Church-style System F with definitions in Redex.

System F in Redex

This is an implementation of a Church-style (explicitly typed-annotated) System F (polymorphic) lambda calculus with definitions (let-expressions) and branching (booleans) in Redex. Included are:

System F

• Type synthesis, CBV small-step semantics
• Church encodings of numerals and some arithmetic

ANF-Restricted System F

• Type synthesis, CBV small-step semantics
• Uses Redex's in-hole contexts for continuations used during compilation
• Compiler from System F to ANF-restricted System F (defined with an ambient continuation)
  • The naïve version copies the continuation twice when compiling if expressions
  • The improved version is defined over System F's typing rules and avoids code copying

(ANF-Restricted) System F with Closures

• Type synthesis, CBV small-step semantics that evaluate closures during application
• Compiler from ANF-restricted System F to closure-converted System F (defined over System F-ANF's typing rules)

(ANF-Restricted) Hoisted System F (with Closures)

• Type synthesis, CBV big-step semantics (implemented as a judgement rather than a reduction-relation)
• Compiler from closure-converted System F to hoisted System F
• N.B. Each code block label is only visible in subsequent code blocks

System Fω (extends System F)

• Why? I don't know.
• Higher-kinded polymorphic lambda calculus with NO let-expressions
• Type synthesis, term evaluation, type evaluation

redex-chk, but Better

• Renames judgment to judgement
• Adds redex-judgement-equals-chk to test for equivalence between judgement output term and a provided term

Reduction Strategies

• This doesn't belong here

TODOs

• Consider adding fixpoints (this might be of interest for ACC... or not)
• Add inventory of metafunctions and how to run things to this README
• Fix redex-judgement-equals-chk macro so that when check-true fails, the highlight goes over the failed branch, not over the macro itself (redex-chk)
• Consider a heap allocation pass. This is primarily to concretize what a closure would look like at low level. I'm guessing the free types and terms would each be an array.