Last active
November 5, 2024 09:52
-
-
Save sgraf812/b9c10d8386a5da7ffe014e9f1dd9bc83 to your computer and use it in GitHub Desktop.
Denotational call-by-name semantics for untyped lambda calculus in Guarded Cubical Agda
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
{-# OPTIONS --cubical --guarded #-} | |
module DenSem where | |
open import Cubical.Foundations.Prelude hiding (_[_↦_]) | |
open import Cubical.Foundations.Transport | |
open import Cubical.Data.Nat | |
open import Cubical.Data.Maybe | |
open import Cubical.Relation.Nullary.Base | |
-- Vendored Guarded Prelude (trusted code, best skipped on first read): | |
module Prims where | |
primitive | |
primLockUniv : Set₁ | |
open Prims renaming (primLockUniv to LockU) public | |
private | |
variable | |
l : Level | |
A B : Set l | |
postulate | |
Tick : LockU | |
▹_ : ∀ {l} → Set l → Set l | |
▹ A = (@tick x : Tick) -> A | |
▸_ : ∀ {l} → ▹ Set l → Set l | |
▸ A = (@tick x : Tick) → A x | |
next : A → ▹ A | |
next x _ = x | |
postulate | |
dfix : ∀ {l} {A : Set l} → (▹ A → A) → ▹ A | |
pfix : ∀ {l} {A : Set l} (f : ▹ A → A) → dfix f ≡ next (f (dfix f)) | |
fix : ∀ {l} {A : Set l} → (▹ A → A) → A | |
fix f = f (dfix f) | |
-- End of trusted code | |
Name = ℕ -- simpler to compare than string | |
decEq-ℕ : (x y : ℕ) → Dec (x ≡ y) | |
decEq-ℕ zero zero = yes refl | |
decEq-ℕ zero (suc y) = no znots | |
decEq-ℕ (suc y) zero = no snotz | |
decEq-ℕ (suc x) (suc y) with decEq-ℕ x y | |
... | yes p = yes (cong suc p) | |
... | no np = no (λ p → np (injSuc p)) | |
instance | |
decEq-ℕ-imp : {x y : ℕ} → Dec (x ≡ y) | |
decEq-ℕ-imp {x} {y} = decEq-ℕ x y | |
data Exp : Set where | |
var : Name → Exp | |
lam : Name → Exp → Exp | |
app : Exp → Exp → Exp | |
-- partial function, serving as a finite Map: | |
_⇀_ : ∀ {ℓ} → Set ℓ → Set ℓ → Set ℓ | |
A ⇀ B = A → Maybe B | |
infix 1 _⇀_ | |
empty-pfun : ∀{A B : Set} → A ⇀ B | |
empty-pfun _ = nothing | |
_[_↦_] : ∀ {A B : Set} {{dec : {x y : A} → Dec (x ≡ y)}} | |
→ (A ⇀ B) → A → B → (A ⇀ B) | |
_[_↦_] {{dec}} ρ x b y with dec {x} {y} | |
... | yes _ = just b | |
... | no _ = ρ y | |
-- Finally the semantic domain model | |
data T (A : Set) : Set where -- Delay monad | |
ret : A → T A | |
step : ▹ (T A) → T A | |
_>>=-T_ : ∀ {A} {B} → T A → (A → T B) → T B | |
T.ret a >>=-T k = k a | |
T.step τ >>=-T k = T.step (λ α → τ α >>=-T k) | |
data ValueF (V⁻ : ▹ Set) : Set where | |
fun : (▸ (λ α → T (V⁻ α)) → T (ValueF V⁻)) → ValueF V⁻ | |
stuck : ValueF V⁻ | |
Value = fix ValueF | |
D = T Value | |
≡-D : ▸ (λ α → T (dfix ValueF α)) ≡ ▹ D | |
≡-D i = (@tick α : Tick) → T (pfix ValueF i α) | |
-- yeah, this is what it takes unless Agda adds more support | |
-- for negative guarded recursion in data types so that we can | |
-- just write (▹ (T ValueF)) in the type of ValueF.fun | |
apply : Value → D → D | |
apply (ValueF.fun f) a = f (transport⁻ ≡-D (next a)) | |
apply _ _ = T.ret ValueF.stuck | |
-- the meat: | |
D⟦_⟧_ : Exp → (Name ⇀ D) → D | |
D⟦ Exp.var x ⟧ ρ with ρ x | |
... | nothing = T.ret ValueF.stuck | |
... | just d = d | |
D⟦ Exp.lam x e ⟧ ρ = T.ret (ValueF.fun (λ d → D⟦ e ⟧(ρ [ x ↦ T.step (transport ≡-D d) ]))) | |
D⟦ Exp.app e₁ e₂ ⟧ ρ = (D⟦ e₁ ⟧ ρ) >>=-T (λ v → apply v (D⟦ e₂ ⟧ ρ)) | |
idid = Exp.app (Exp.lam 0 (Exp.var 0)) (Exp.lam 0 (Exp.var 0)) | |
omega = Exp.app (Exp.lam 0 (Exp.app (Exp.var 0) (Exp.var 0))) (Exp.lam 0 (Exp.app (Exp.var 0) (Exp.var 0))) | |
-- Press Ctrl-C Ctrl-N and input either res_idid or res_omega | |
-- to see the "normalised" expression | |
-- (apparently `transp v` is not considered a value, so this | |
-- works for omega as well) | |
res_idid = D⟦ idid ⟧ empty-pfun | |
res_omega = D⟦ omega ⟧ empty-pfun |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Yes, exactly. I guess one could write pattern synonyms to get the desired API, but it would be great if Agda would just accept this definition to begin with