module 1Lab.Inductive where

open import 1Lab.Reflection
open import 1Lab.Equiv
open import 1Lab.Type hiding (case_of_ ; case_return_of_)

{-
Automation for applying induction principles
============================================

When working with mixed truncations/products/quotients, the relevant
induction principles are often repeatedly applied in close succession,
like in the snippet below.

  □-rec! λ { (a , b , w) → □-rec! (λ { (c , d , x) →
      inc (c , d , inc (a , x , w)) }) b }

Applying these induction principles is entirely mechanical: like other
mechanical processes, it's annoying to do by hand, generates ugly code,
and causes a deep sense of existential uselessness. This module
implements automation for automatically applying these eliminators "as
far as possible", using overlapping instances.

The Inductive class
-------------------

The core of the implementation for rec!/elim! is the Inductive class,
which is a slight misnomer. To support *nested* types with a single
constructor, the Inductive class is applied to a section of the motive,
not just base type. That is, if we want to simplify

  □ (□ A × □ (□ A)) → B
  ^~~~~~~~~~~~~~~~~ domain

we don't look for an instance of `Inductive domain`, we look for an
instance of `Inductive (domain → B)`. The relevant instances then
manipulate the entire function type at once, much like Extensional.

Writing rules
-------------

Like Extensional, instances of 'Inductive' should be maximally lazy, so
that the structural rules can fire as often as possible. That is,
instead of writing

  ⦃ _ : ∀ {x} → Inductive (P (inc x)) ⦄ → Inductive (∀ x → P x)

we write

  ⦃ _ : Inductive (∀ x → P (inc x)) ⦄ → Inductive (∀ x → P x)
-}

record Inductive {} (A : Type ) ℓm : Type (  lsuc ℓm) where
  field
    methods : Type ℓm
    from    : methods  A

open Inductive

private variable
   ℓ' ℓ'' ℓm : Level
  A B C : Type 
  P Q R : A  Type 

instance
  -- The basic structural rules allow us to stop recurring (we can
  -- produce an A given an A) and to skip an argument, introducing a
  -- function type into the methods:

  Inductive-default : Inductive A (level-of A)
  Inductive-default .methods = _
  Inductive-default .from x  = x

  Inductive-Π
    :  _ : {x : A}  Inductive (P x) ℓm 
     Inductive (∀ x  P x) (level-of A  ℓm)
  Inductive-Π  r  .methods  =  x  r {x} .methods
  Inductive-Π  r  .from f x = r .from (f x)

  {-# INCOHERENT Inductive-default #-}
  {-# OVERLAPPABLE Inductive-Π #-}

  -- Next, we have a rule for uncurrying pairs. This lets us handle types like
  --   □ (□ A × □ B) → C
  --
  -- by factoring through
  --
  --   □ (□ A × □ B) → C
  --   □ A × □ B → C
  --   □ A → □ B → C
  --   □ B → C

  Inductive-Σ
    :  {A : Type } {B : A  Type ℓ'} {C : (x : A)  B x  Type ℓ''}
      _ : Inductive ((x : A) (y : B x)  C x y) ℓm 
     Inductive ((x : Σ A B)  C (x .fst) (x .snd)) ℓm
  Inductive-Σ  r  .methods        = r .methods
  Inductive-Σ  r  .from f (x , y) = r .from f x y

  Inductive-Lift
    : {B : Lift  A  Type ℓ'}
      _ : Inductive (∀ x  B (lift x)) ℓm 
     Inductive (∀ x  B x) ℓm
  Inductive-Lift  i  = record { from = λ f (lift x)  i .from f x }

  -- However, we don't uncurry equivalences.

  Inductive-≃ : {C : A  B  Type ℓ''}  Inductive (∀ x  C x) _
  Inductive-≃ = Inductive-default

  {-# OVERLAPPING Inductive-≃ #-}

-- There are two distinct prefix entry points: rec! and elim!. The
-- difference is just in name, to provide a modicum of documentation. In
-- addition, they have an explicit function type, so that the refine
-- command will always introduce a question mark.

rec! :  r : Inductive (A  B) ℓm   r .methods  A  B
rec!  r  = r .from

elim! :  r : Inductive (∀ x  P x) ℓm   r .methods   x  P x
elim!  r  = r .from

-- We also define versions of case_of_ and case_return_of_, shadowing
-- those from 1Lab.Type, which insert a rec!/elim!.

case_of_ :  { ℓ'} {A : Type } {B : Type ℓ'}  r : Inductive (A  B) ℓm   A  r .methods  B
case x of f = rec! f x

case_return_of_ :  { ℓ'} {A : Type } (x : A) (B : A  Type ℓ')  r : Inductive (∀ x  B x) ℓm  (f : r .methods)  B x
case x return P of f = elim! f x

{-# INLINE case_of_        #-}
{-# INLINE case_return_of_ #-}