A typesystems library with exaggerated claims https://github.com/Lokidottir/typerbole
|Latest on Hackage:||0.0.0.5|
This package is not currently in any snapshots. If you're interested in using it, we recommend adding it to Stackage Nightly. Doing so will make builds more reliable, and allow stackage.org to host generated Haddocks.
Parameterized typesystems, lambda cube typeclasses, and typechecking interfaces.
Like how datatypes such as
List a (
Tree a etc. in haskell have a parameter for a contained type, this library is based on the idea that a datatype that represents expressions can have a parameter for a typesystem.
An Example: The Lambda Calculus
As an example, we can put together a datatype that represents the syntax for the Lambda Calculus:
data LambdaTerm c v t = Variable v -- a variable bound by a lambda abstraction | Constant c -- a constant defined outside of the term | Apply (LambdaTerm c v t) (LambdaTerm c v t) -- an application of one term to another | Lambda (v, t) (LambdaTerm c v t) -- A lambda abstraction
This datatype has 3 parameters. The first two parameters represent constants and variables respectively, what’s important is the final parameter
t which is the parameter for the typesystem being used.
We can use the typesystem
Compiler.Typesystem.SimplyTyped as the typesystem to make this a simply-typed lambda calculus, or we could slot in
Hask, to change the typesystem associcated with with it.
Sadly there’s no magic that builds typecheckers for these (yet). Instead, using the language extensions
FlexibleInstances and the
Typecheckable typeclass from
Control.Typecheckable we write a typechecker for each of these occurences.
instance (...) => Typecheckable (LambdaTerm c v) (SimplyTyped m) where ... instance (...) => Typecheckable (LambdaTerm c v) (SystemF m p) where ... instance (...) => Typecheckable (LambdaTerm c v) (SystemFOmega m p k) where ... -- and so on.
Or we can just ignore it all and turn it into an untyped lambda calculus:
type UntypedLambdaTerm c v = LambdaTerm c v ()
The Lambda Cube
The lambda cube describes the properties of a number of typesystems, an overview can be found here. It is the basis for the library’s classification of typesystems, a typeclass hierarchy where each axis is represented by a typeclass whose methods and associated types are indicitive of the properties of the axis.
Supported lambda-cube axies
- [x] Simply-typed lambda calculus
- [x] Polymorphic lambda calculus
- [x] Higher-order lambda calculus
- [x] Dependently-typed lambda calculus (dubiously, not got a implemented typesystem to back it up)
- [ ] Give
Calculi.Lambda.Cube.Polymorphic.Unificationbetter documentation (incl. diagrams for graph-related functions/anything that’ll benefit).
- [ ] Finish the
Inferableinstances for the typesystems in
- [ ] Put together a working travis file.
- [ ] Implement a Calculus of Constructions typesystem.
- [ ] Document the type expression psudocode
- [ ] Design a typeclass for typesystems with constraints (
Num a => ...,
a ~ Tetc).
- [ ] Provide a default way of evaluating lambda expressions.
- [ ] Make the quasiquoters use the lambda cube typeclasses instead of specific typesystem implementations.
- [ ] Subhask-style automated test writing.
- [ ] Explore homotopy type theory
- [ ] Remove all extensions that aren’t light syntactic sugar from
default-extensionsand declare them explicitly in the modules they’re used.
- [ ] Listen to
- [ ] Move
Control.Typecheckableto it’s own package.
- [ ] Elaborate on the
Typechecktype. Maybe make it a typeclass.
Papers, Sites and Books read during development
Introduction to generalized type systems, Dr Henk Barendregt (Journal of Functional Programming, April 1991)
A Modern Perspective on Type Theory (x)
A proof of correctness for the Hindley-Milner type inference algorithm, Dr Jeff Vaughan (x)
Compositional Type Checking for Hindley-Milner Type Systems with Ad-hoc Polymorphism, Dr. Gergő Érdi (x)
Many wikipedia pages on type theory.