Cedille.ced

module cedille.

{- Quant.       Abstract.    Appl.
   ---------    ---------    -----
   Π x: 𝒌. 𝒌    λ X: 𝒌. U    T ·U
   Π x: T. 𝒌    λ x: T. U    T  e
   ∀ X: 𝒌. U    Λ X: 𝒌. e    t ·U
   ∀ x: T. U    Λ x: T. e    t -e
   Π x: T. U    λ x: T. e    t  e -}

-- Declare type with kind
Unit ◂ ★
= ∀ X: ★. X ➔ X.

-- Declare term with type
unit : Unit
= Λ X. λ x. x.

-- Local definitions
true : ∀ X: ★. X ➔ X ➔ X
= Λ X. λ x.
  -- Erased definition
  {Y: ★ = X} - λ y: Y. x.
false : ∀ X: ★. X ➔ X ➔ X
= Λ X. λ x.
  -- Non-erased definition
  -- χ T - t asserts t has type T
  [id = χ (X ➔ X) - λ y. y] - id.

-- Trivial equality proof
refl : ∀ X: ★. ∀ x: X. {x ≃ x}
= Λ X. Λ x. β.

-- Rewriting via an equality
-- Given a goal type T and an equality p: {e ≃ e'},
-- ρ p - u will rewrite T containing e to T' containing e'
-- in place of e, proving T' with u.
-- The term e of a trivial reflexive proof {e ≃ e}
-- can be specified using β<e>.
sym : ∀ X: ★. ∀ x: X. ∀ y: X. {x ≃ y} ➔ {y ≃ x}
= Λ X. Λ x. Λ y. λ p. ρ p - β<y>.

-- Equality with explicit erasure term
-- The erasure of an equality proof β{|t|} is t;
-- since the erasure of equalities could be anything,
-- the induction hypothesis must cover all erasures.
-- Note the missing type argument of d that was inferred.
J : ∀ X: ★. ∀ x: X. ∀ P: (Π y: X. {x ≃ y} ➔ ★).
    Π d: ∀ T: ★. ∀ t: T. P x β{|t|}.
    ∀ y: X. ∀ p: {x ≃ y}. P y p
= Λ X. Λ x. Λ P. λ d. Λ y. Λ p. ρ (sym -x -y p) - d -p.

-- Subst via J with automatic motive inference
-- In the below where the θ term is, the desired type is P x ➔ P y.
-- We wish to use J to prove this, and J requires a motive P'.
-- θ<y p> then abstracts the desired type over these variables,
-- creating the motive (λ y: X. λ p: {x ≃ y}. P x ➔ P y) (p is unused),
-- then applies the motive to (J -x).
-- Without θ, this would be written as J -x ·P' (Λ T. Λ t. λ px. px) -y -p,
-- which would be a lot more tedious to write with P' explicitly stated,
-- especially since the abstracted types X and {x ≃ y}
-- syntactically need to be written out in full.
subst : ∀ X: ★. ∀ x: X. ∀ y: X. ∀ P: X ➔ ★. ∀ p: {x ≃ y}. P x ➔ P y
= Λ X. Λ x. Λ y. Λ P. Λ p. θ<y p> (J -x) (Λ T. Λ t. λ px. px) -y -p.

-- Casting via an equality
-- Given terms e: T, e', and equality p: {e ≃ e'},
-- φ p - e {|e'|} erases to e' but has type T.
Omega : {unit ≃ λ x. x x} ➾ Unit
= Λ p. [omega = φ p - unit {|λ x. x x|}] - omega omega.

-- Ex falso quodlibet
-- Given an equality p: {e ≃ e'} of Böhm-separable terms,
-- i.e. terms with distinct closed head-normal forms,
-- δ - p can be assigned any type (although it erases to λ x. x).
-- Then φ can be used to construct an X that erases to any term;
-- in the below example it erases to p, just for fun.
absurd : {true ≃ false} ➔ ∀ X: ★. X
= λ p. [void: ∀ X: ★. X = δ - p] -
Λ X. φ (void ·{void ≃ p}) - (void ·X) {|p|}.

-- Declaring a datatype
-- Parameters are elided in all recursive references.
data Acc (X: ★) (R: X ➔ X ➔ ★) : X ➔ ★ =
| acc : ∀ x: X. (∀ y: X. R y x ➔ Acc y) ➔ Acc x.

-- Case destruction with motive
-- Given a construction c, σ c @P { | C a… ➔ e } deconstructs c
-- into a constructor C and arguments a…, producing e
-- (where a… may be types ·X or implicits -x, but not parameters)
-- with optional explicit motive P abstracting over indices and target.
-- Supposing inductive I and P = λ i…. λ x. P', in Coq this would be
-- case c as x in I _… i… return P' of | C a… ➔ e end.
accessible : ∀ X: ★. ∀ R: X ➔ X ➔ ★.
             ∀ x: X. ∀ y: X. R y x ➔ Acc ·X ·R x ➔ Acc ·X ·R y
= Λ X. Λ R. Λ x. Λ y. λ r. λ accx.
σ accx @(λ x': X. λ accx': Acc ·X ·R x'. {x' ≃ x} ➔ Acc ·X ·R y) {
| acc -x accy ➔ λ p. accy -y (ρ p - r)
} β<x>.
-- Note that destructing something with an index doesn't unify
-- the index outside of the destruction and inside of it,
-- so using the convoy pattern may be necessary.

-- Well-founded induction
-- If all X are accessible and the predicate P holding for all y R x
-- means that it holds for x, then P holds for all x.
-- Given a construction c, μ f. c @P { | C a… ➔ e } is a fixed point
-- recurring on c, again with optional explicit motive P.
-- Supposing inductive I, in Coq this would be an applied fixpoint
-- (fix f (x: I) : P x := case x of | C a… ➔ e end) c
-- (omitting inductive parameters and indices).
-- Notably, the type of f abstracts over the indices as well.
wfind : ∀ X: ★. ∀ R: X ➔ X ➔ ★. (∀ x: X. Acc ·X ·R x) ➔
        ∀ P: X ➔ ★. (∀ x: X. (∀ y: X. R y x ➔ P y) ➔ P x) ➔
        ∀ x: X. P x
= Λ X. Λ R. λ accessible. Λ P. λ ih. Λ x.
μ wfind. (accessible -x) @(λ x: X. λ accx: Acc ·X ·R x. P x) {
| acc -x accy ➔ ih -x (Λ y. λ r. wfind -y (accy -y r))
}.