open import 1Lab.Prelude

open import Data.List.Properties
open import Data.List.Base
open import Data.Dec.Base
open import Data.Fin.Base
open import Data.Sum.Base
open import Data.Id.Base
open import Data.Bool

open import Meta.Idiom

module Data.List.Membership where

private variable
  ℓ ℓ' : Level
  A B : Type ℓ
  P Q : A → Type ℓ'
  x y : A
  xs ys : List A

Membership in lists🔗

We can inductively define a notion of membership between elements $a:A$ and lists $as:\mathrm{List}(A)\text{;}$ we write $a{\in }_{l}as\text{.}$ However, note that this notion of membership is not a proposition! The type $a{\in }_{l}as$ has at least as many inhabitants as there are occurrences of $a$ in $as\text{;}$ and if the type $A$ is not a set, then each proof of $a\in as$ can be acted on by a loop on $a$ to give a new proof.

data _∈ₗ_ {ℓ} {A : Type ℓ} (x : A) : List A → Type ℓ where
  here  : ∀ {x'} (p : x ≡ x') → x ∈ₗ (x' ∷ xs)
  there : (p : x ∈ₗ xs)       → x ∈ₗ (y ∷ xs)

instance
  Membership-List : ∀ {ℓ} {A : Type ℓ} → Membership A (List A) ℓ
  Membership-List = record { _∈_ = _∈ₗ_ }

There is a more (homotopically) straightforward characterisation of membership in lists: the fibres of the lookup function xs ! i. These are given by an index $i:[\mathrm{length}(xs)]\text{,}$ living in the standard finite set with as many elements as the list has positions, together with a proof that $xs!i=x\text{.}$

The inverse equivalences between these notions is defined below; the proof that they are are inverses is a straightforward induction in both cases, so it’s omitted for space.

element→!-fibre : ∀ {x : A} {xs} → x ∈ xs → fibre (xs !_) x
element→!-fibre (here p) = fzero , sym p
element→!-fibre (there prf) with element→!-fibre prf
... | ix , p = fsuc ix , p

!-fibre→element : ∀ {x : A} {xs} → fibre (xs !_) x → x ∈ xs
!-fibre→element {A = A} {x = x} = λ (ix , p) → go ix p module !-fibre→element where
  go : ∀ {xs} (ix : Fin (length xs)) → xs ! ix ≡ x → x ∈ xs
  go ix _  with fin-view ix
  go {xs = x ∷ xs} _ p | zero     = here  (sym p)
  go {xs = x ∷ xs} _ p | (suc ix) = there (go ix p)

The equivalence between these definitions explains why $a{\in }_{l}as$ can be so complicated. First, it’s at least a set, since it stores the index. Second, it stores a path, which can be arbitrarily complicated depending on the type $A\text{.}$

!-fibre→element→fibre : ∀ {x : A} {xs} (f : fibre (xs !_) x) → element→!-fibre (!-fibre→element f) ≡ f
!-fibre→element→fibre {A = A} {x = x} (ix , p) = go ix p where
  go : ∀ {xs} (ix : Fin (length xs)) (p : xs ! ix ≡ x) → element→!-fibre (!-fibre→element.go {xs = xs} ix p) ≡ (ix , p)
  go ix p with fin-view ix
  go {xs = x ∷ xs} _ p | zero = refl
  go {xs = x ∷ xs} _ p | suc ix = Σ-pathp (ap fsuc (ap fst p')) (ap snd p')
    where p' = go {xs = xs} ix p

element→!-fibre→element
  : {x : A} {xs : List A} (p : x ∈ xs) → p ≡ !-fibre→element (element→!-fibre p)
element→!-fibre→element (here p)  = refl
element→!-fibre→element (there p) = ap there (element→!-fibre→element p)

element≃!-fibre : ∀ {x : A} {xs} → (x ∈ₗ xs) ≃ fibre (xs !_) x
element≃!-fibre .fst = element→!-fibre
element≃!-fibre .snd = is-iso→is-equiv λ where
  .is-iso.inv  p → !-fibre→element p
  .is-iso.rinv p → !-fibre→element→fibre p
  .is-iso.linv p → sym (element→!-fibre→element p)

Despite the potential complexity of $a{\in }_{l}as\text{,}$ we do note that if $A$ is a set, then all that matters is the index; If $A$ is moreover discrete, then $a{\in }_{l}as$ is decidable.

elem? : ⦃ _ : Discrete A ⦄ (x : A) (xs : List A) → Dec (x ∈ xs)
elem? x [] = no λ ()
elem? x (y ∷ xs) with x ≡ᵢ? y
... | yes reflᵢ = yes (here refl)
... | no ¬p with elem? x xs
... | yes p = yes (there p)
... | no ¬q = no λ { (here p) → ¬p (Id≃path.from p) ; (there q) → ¬q q }

instance
  Dec-∈ₗ : ⦃ _ : Discrete A ⦄ {x : A} {xs : List A} → Dec (x ∈ xs)
  Dec-∈ₗ {x = x} {xs} = elem? x xs

Removing duplicates🔗

Still working with a discrete type, we can define a function nub which removes duplicates from a list. It is constructed by inductively deciding whether or not to keep the head of the list, discarding those which already appear further down. This has terrible the terrible time complexity $O(n^2)\text{,}$ but it works for an arbitrary discrete type, which is the best possible generality.

nub-cons : (x : A) (xs : List A) → Dec (x ∈ xs) → List A
nub-cons x xs (yes _) = xs
nub-cons x xs (no _)  = x ∷ xs

nub : ⦃ _ : Discrete A ⦄ → List A → List A
nub []       = []
nub (x ∷ xs) = nub-cons x (nub xs) (elem? x (nub xs))

The function nub is characterised by the following two facts. First, membership in nub is a proposition — each element appears at most once. Second, membership is (logically) preserved when nubbing a list — note that the function mapping old indices to new indices is not injective, since all occurrences of an element will be mapped to the same (first) occurrence in the deduplicated list.

member-nub-is-prop
  : ∀ ⦃ _ : Discrete A ⦄ {x : A} (xs : List A) → is-prop (x ∈ nub xs)
member→member-nub
  : ∀ ⦃ _ : Discrete A ⦄ {x : A} {xs : List A} → x ∈ xs → x ∈ nub xs

member-nub-is-prop (x ∷ xs) p1 p2 with elem? x (nub xs) | p1 | p2
... | yes p | p1 | p2 = member-nub-is-prop xs p1 p2
... | no ¬p | here  p1 | here  p2 = ap _∈ₗ_.here (Discrete→is-set auto _ _ p1 p2)
... | no ¬p | here  p1 | there p2 = absurd (¬p (subst (_∈ nub xs) p1 p2))
... | no ¬p | there p1 | here  p2 = absurd (¬p (subst (_∈ nub xs) p2 p1))
... | no ¬p | there p1 | there p2 = ap there (member-nub-is-prop xs p1 p2)

member→member-nub {xs = x ∷ xs} (here p) with elem? x (nub xs)
... | yes x∈nub = subst (_∈ nub xs) (sym p) x∈nub
... | no ¬x∈nub = here p
member→member-nub {xs = x ∷ xs} (there α) with elem? x (nub xs)
... | yes x∈nub = member→member-nub α
... | no ¬x∈nub = there (member→member-nub α)

!-tabulate : ∀ {n} (f : Fin n → A) i → tabulate f ! i ≡ f (subst Fin (length-tabulate f) i)
!-tabulate _ ix with fin-view ix
!-tabulate {n = suc n} f _ | zero  = refl
!-tabulate {n = suc n} f _ | suc i = !-tabulate (f ∘ fsuc) i

!-tabulate-fibre : ∀ {n} (f : Fin n → A) x → fibre (tabulate f !_) x ≃ fibre f x
!-tabulate-fibre f x = Σ-ap (path→equiv (ap Fin (length-tabulate f))) λ i →
  path→equiv (ap (_≡ x) (!-tabulate f i))

member-tabulate : ∀ {n} (f : Fin n → A) x → (x ∈ tabulate f) ≃ fibre f x
member-tabulate f x = element≃!-fibre ∙e !-tabulate-fibre f x

map-member
  : ∀ {A : Type ℓ} {B : Type ℓ'} (f : A → B) {x : A} {xs : List A}
  → x ∈ xs → f x ∈ map f xs
map-member f (here p)  = here (ap f p)
map-member f (there x) = there (map-member f x)

++-memberₗ : x ∈ₗ xs → x ∈ₗ (xs ++ ys)
++-memberₗ (here p)  = here p
++-memberₗ (there p) = there (++-memberₗ p)

++-memberᵣ : x ∈ₗ ys → x ∈ₗ (xs ++ ys)
++-memberᵣ {xs = []}     p = p
++-memberᵣ {xs = x ∷ xs} p = there (++-memberᵣ p)

any-one-of
  : ∀ {ℓ} {A : Type ℓ}
  → (f : A → Bool) (x : A) (xs : List A)
  → x ∈ xs → f x ≡ true
  → any-of f xs ≡ true
any-one-of f x (y ∷ xs) (here x=y) x-true =
  ap₂ or (subst (λ e → f e ≡ true) x=y x-true) refl
any-one-of f x (y ∷ xs) (there x∈xs) x-true =
  ap₂ or refl (any-one-of f x xs x∈xs x-true) ∙ or-truer _