lean2/src/builtin/cast.lean

80 lines
3.4 KiB
Text
Raw Normal View History

-- "Type casting" library.
-- Heterogeneous substitution
axiom hsubst {A B : TypeU} {a : A} {b : B} (P : ∀ T : TypeU, T → Bool) : P A a → a == b → P B b
universe M >= 1
universe U >= M + 1
definition TypeM := (Type M)
-- Type equality axiom, if two values are equal, then their types are equal
theorem type_eq {A B : TypeM} {a : A} {b : B} (H : a == b) : A == B
:= hsubst (λ (T : TypeU) (x : T), A == T) (refl A) H
-- Heterogenous symmetry
theorem hsymm {A B : TypeU} {a : A} {b : B} (H : a == b) : b == a
:= hsubst (λ (T : TypeU) (x : T), x == a) (refl a) H
-- Heterogenous transitivity
theorem htrans {A B C : TypeU} {a : A} {b : B} {c : C} (H1 : a == b) (H2 : b == c) : a == c
:= hsubst (λ (T : TypeU) (x : T), a == x) H1 H2
-- The cast operator allows us to cast an element of type A
-- into B if we provide a proof that types A and B are equal.
variable cast {A B : TypeU} : A == B → A → B
-- The CastEq axiom states that for any cast of x is equal to x.
axiom cast_eq {A B : TypeU} (H : A == B) (x : A) : x == cast H x
-- The CastApp axiom "propagates" the cast over application
axiom cast_app {A A' : TypeU} {B : A → TypeU} {B' : A' → TypeU}
(H1 : (∀ x, B x) == (∀ x, B' x)) (H2 : A == A')
(f : ∀ x, B x) (x : A) :
cast H1 f (cast H2 x) == f x
-- Heterogeneous congruence
theorem hcongr
{A A' : TypeM} {B : A → TypeM} {B' : A' → TypeM}
{f : ∀ x, B x} {g : ∀ x, B' x} {a : A} {b : A'}
(H1 : f == g)
(H2 : a == b)
: f a == g b
:= let L1 : A == A' := type_eq H2,
L2 : A' == A := symm L1,
b' : A := cast L2 b,
L3 : b == b' := cast_eq L2 b,
L4 : a == b' := htrans H2 L3,
L5 : f a == f b' := congr2 f L4,
S1 : (∀ x, B' x) == (∀ x, B x) := symm (type_eq H1),
g' : (∀ x, B x) := cast S1 g,
L6 : g == g' := cast_eq S1 g,
L7 : f == g' := htrans H1 L6,
L8 : f b' == g' b' := congr1 b' L7,
L9 : f a == g' b' := htrans L5 L8,
L10 : g' b' == g b := cast_app S1 L2 g b
in htrans L9 L10
exit -- Stop here, the following axiom is not sound
-- The following axiom is unsound when we treat Pi and
-- forall as the "same thing". The main problem is the
-- rule that says (Pi x : A, B) has type Bool if B has type Bool instead of
-- max(typeof(A), typeof(B))
--
-- Here is the problematic axiom
-- If two (dependent) function spaces are equal, then their domains are equal.
axiom dominj {A A' : (Type U)} {B : A → (Type U)} {B' : A' → (Type U)}
(H : (∀ x : A, B x) == (∀ x : A', B' x)) :
A == A'
-- Here is a derivation of false using the dominj axiom
theorem unsat : false :=
let L1 : (∀ x : true, true) := (λ x : true, trivial),
L2 : (∀ x : false, true) := (λ x : false, trivial),
-- When we keep Pi/forall as different things, the following two steps can't be used
L3 : (∀ x : true, true) = true := eqt_intro L1,
L4 : (∀ x : false, true) = true := eqt_intro L2,
L5 : (∀ x : true, true) = (∀ x : false, true) := trans L3 (symm L4),
L6 : true = false := dominj L5
in L6