Library Coq.Classes.Morphisms


Typeclass-based morphism definition and standard, minimal instances

Author: Matthieu Sozeau Institution: LRI, CNRS UMR 8623 - University Paris Sud

Require Import Coq.Program.Basics.
Require Import Coq.Program.Tactics.
Require Import Coq.Relations.Relation_Definitions.
Require Export Coq.Classes.RelationClasses.


Morphisms.

We now turn to the definition of Proper and declare standard instances. These will be used by the setoid_rewrite tactic later.
A morphism for a relation R is a proper element of the relation. The relation R will be instantiated by respectful and A by an arrow type for usual morphisms.
Section Proper.
  Let U := Type.
  Context {A B : U}.

  Class Proper (R : relation A) (m : A) : Prop :=
    proper_prf : R m m.

Every element in the carrier of a reflexive relation is a morphism for this relation. We use a proxy class for this case which is used internally to discharge reflexivity constraints. The Reflexive instance will almost always be used, but it won't apply in general to any kind of Proper (A -> B) _ _ goal, making proof-search much slower. A cleaner solution would be to be able to set different priorities in different hint bases and select a particular hint database for resolution of a type class constraint.

  Class ProperProxy (R : relation A) (m : A) : Prop :=
    proper_proxy : R m m.

  Lemma eq_proper_proxy (x : A) : ProperProxy (@eq A) x.

  Lemma reflexive_proper_proxy `(Reflexive A R) (x : A) : ProperProxy R x.

  Lemma proper_proper_proxy x `(Proper R x) : ProperProxy R x.

Respectful morphisms.
The fully dependent version, not used yet.

  Definition respectful_hetero
  (A B : Type)
  (C : A -> Type) (D : B -> Type)
  (R : A -> B -> Prop)
  (R' : forall (x : A) (y : B), C x -> D y -> Prop) :
    (forall x : A, C x) -> (forall x : B, D x) -> Prop :=
    fun f g => forall x y, R x y -> R' x y (f x) (g y).

The non-dependent version is an instance where we forget dependencies.

  Definition respectful (R : relation A) (R' : relation B) : relation (A -> B) :=
    Eval compute in @respectful_hetero A A (fun _ => B) (fun _ => B) R (fun _ _ => R').

End Proper.

We favor the use of Leibniz equality or a declared reflexive relation when resolving ProperProxy, otherwise, if the relation is given (not an evar), we fall back to Proper.
#[global]
Hint Extern 1 (ProperProxy _ _) =>
  class_apply @eq_proper_proxy || class_apply @reflexive_proper_proxy : typeclass_instances.

#[global]
Hint Extern 2 (ProperProxy ?R _) =>
  not_evar R; class_apply @proper_proper_proxy : typeclass_instances.

Notations reminiscent of the old syntax for declaring morphisms.
Delimit Scope signature_scope with signature.

Module ProperNotations.

  Notation " R ++> R' " := (@respectful _ _ (R%signature) (R'%signature))
    (right associativity, at level 55) : signature_scope.

  Notation " R ==> R' " := (@respectful _ _ (R%signature) (R'%signature))
    (right associativity, at level 55) : signature_scope.

  Notation " R --> R' " := (@respectful _ _ (flip (R%signature)) (R'%signature))
    (right associativity, at level 55) : signature_scope.

End ProperNotations.


Export ProperNotations.

Local Open Scope signature_scope.

solve_proper try to solve the goal Proper (?==> ... ==>?) f by repeated introductions and setoid rewrites. It should work fine when f is a combination of already known morphisms and quantifiers.

Ltac solve_respectful t :=
 match goal with
   | |- respectful _ _ _ _ =>
     let H := fresh "H" in
     intros ? ? H; solve_respectful ltac:(setoid_rewrite H; t)
   | _ => t; reflexivity
 end.

Ltac solve_proper := unfold Proper; solve_respectful ltac:(idtac).

f_equiv is a clone of f_equal that handles setoid equivalences. For example, if we know that f is a morphism for E1==>E2==>E, then the goal E (f x y) (f x' y') will be transformed by f_equiv into the subgoals E1 x x' and E2 y y'.

Ltac f_equiv :=
 match goal with
  | |- ?R (?f ?x) (?f' _) =>
    let T := type of x in
    let Rx := fresh "R" in
    evar (Rx : relation T);
    let H := fresh in
    assert (H : (Rx==>R)%signature f f');
    unfold Rx in *; clear Rx; [ f_equiv | apply H; clear H; try reflexivity ]
  | |- ?R ?f ?f' =>
    solve [change (Proper R f); eauto with typeclass_instances | reflexivity ]
  | _ => idtac
 end.

Section Relations.
  Let U := Type.
  Context {A B : U} (P : A -> U).

forall_def reifies the dependent product as a definition.

  Definition forall_def : Type := forall x : A, P x.

Dependent pointwise lifting of a relation on the range.

  Definition forall_relation
             (sig : forall a, relation (P a)) : relation (forall x, P x) :=
    fun f g => forall a, sig a (f a) (g a).

Non-dependent pointwise lifting
  Definition pointwise_relation (R : relation B) : relation (A -> B) :=
    fun f g => forall a, R (f a) (g a).

  Lemma pointwise_pointwise (R : relation B) :
    relation_equivalence (pointwise_relation R) (@eq A ==> R).

Subrelations induce a morphism on the identity.

  Global Instance subrelation_id_proper `(subrelation A RA RA') : Proper (RA ==> RA') id.

The subrelation property goes through products as usual.
And of course it is reflexive.

  Lemma subrelation_refl R : @subrelation A R R.

Proper is itself a covariant morphism for subrelation. We use an unconvertible premise to avoid looping.

  Lemma subrelation_proper `(mor : Proper A R' m)
        `(unc : Unconvertible (relation A) R R')
        `(sub : subrelation A R' R) : Proper R m.

  Global Instance proper_subrelation_proper :
    Proper (subrelation ++> eq ==> impl) (@Proper A).

  Global Instance pointwise_subrelation `(sub : subrelation B R R') :
    subrelation (pointwise_relation R) (pointwise_relation R') | 4.

For dependent function types.
  Lemma forall_subrelation (R S : forall x : A, relation (P x)) :
    (forall a, subrelation (R a) (S a)) -> subrelation (forall_relation R) (forall_relation S).
End Relations.

Typeclasses Opaque respectful pointwise_relation forall_relation.

#[global]
Hint Unfold Reflexive : core.
#[global]
Hint Unfold Symmetric : core.
#[global]
Hint Unfold Transitive : core.

Resolution with subrelation: favor decomposing products over applying reflexivity for unconstrained goals.
Ltac subrelation_tac T U :=
  (is_ground T ; is_ground U ; class_apply @subrelation_refl) ||
    class_apply @subrelation_respectful || class_apply @subrelation_refl.

#[global]
Hint Extern 3 (@subrelation _ ?T ?U) => subrelation_tac T U : typeclass_instances.

CoInductive apply_subrelation : Prop := do_subrelation.

Ltac proper_subrelation :=
  match goal with
    [ H : apply_subrelation |- _ ] => clear H ; class_apply @subrelation_proper
  end.

#[global]
Hint Extern 5 (@Proper _ ?H _) => proper_subrelation : typeclass_instances.

Essential subrelation instances for iff, impl and pointwise_relation.
We use an extern hint to help unification.

#[global]
Hint Extern 4 (subrelation (@forall_relation ?A ?B ?R) (@forall_relation _ _ ?S)) =>
  apply (@forall_subrelation A B R S) ; intro : typeclass_instances.

Section GenericInstances.
  Let U := Type.
  Context {A B C : U}.

We can build a PER on the Coq function space if we have PERs on the domain and codomain.

  Program Instance respectful_per `(PER A R, PER B R') : PER (R ==> R').


The complement of a relation conserves its proper elements.

  Program Definition complement_proper
          `(mR : Proper (A -> A -> Prop) (RA ==> RA ==> iff) R) :
    Proper (RA ==> RA ==> iff) (complement R) := _.


The flip too, actually the flip instance is a bit more general.

  Program Definition flip_proper
          `(mor : Proper (A -> B -> C) (RA ==> RB ==> RC) f) :
    Proper (RB ==> RA ==> RC) (flip f) := _.


Every Transitive relation gives rise to a binary morphism on impl, contravariant in the first argument, covariant in the second.



Proper declarations for partial applications.











Every Transitive relation induces a morphism by "pushing" an R x y on the left of an R x z proof to get an R y z goal.



Every Symmetric and Transitive relation gives rise to an equivariant morphism.
Coq functions are morphisms for Leibniz equality, applied only if really needed.

  Global Instance reflexive_eq_dom_reflexive `(Reflexive B R') :
    Reflexive (@Logic.eq A ==> R').

respectful is a morphism for relation equivalence.

  Global Instance respectful_morphism :
    Proper (relation_equivalence ++> relation_equivalence ++> relation_equivalence)
           (@respectful A B).

R is Reflexive, hence we can build the needed proof.
Treating flip: can't make them direct instances as we need at least a flip present in the goal.

  Lemma flip1 `(subrelation A R' R) : subrelation (flip (flip R')) R.

  Lemma flip2 `(subrelation A R R') : subrelation R (flip (flip R')).

That's if and only if

  Lemma eq_subrelation `(Reflexive A R) : subrelation (@eq A) R.

Once we have normalized, we will apply this instance to simplify the problem.

  Definition proper_flip_proper `(mor : Proper A R m) : Proper (flip R) m := mor.

Every reflexive relation gives rise to a morphism, only for immediately solving goals without variables.

  Lemma reflexive_proper `{Reflexive A R} (x : A) : Proper R x.

  Lemma proper_eq (x : A) : Proper (@eq A) x.

End GenericInstances.

Class PartialApplication.

CoInductive normalization_done : Prop := did_normalization.

Class Params {A : Type} (of : A) (arity : nat).

Ltac partial_application_tactic :=
  let rec do_partial_apps H m cont :=
    match m with
      | ?m' ?x => class_apply @Reflexive_partial_app_morphism ;
        [(do_partial_apps H m' ltac:(idtac))|clear H]
      | _ => cont
    end
  in
  let rec do_partial H ar m :=
    lazymatch ar with
      | 0%nat => do_partial_apps H m ltac:(fail 1)
      | S ?n' =>
        match m with
          ?m' ?x => do_partial H n' m'
        end
    end
  in
  let params m sk fk :=
    (let m' := fresh in head_of_constr m' m ;
     let n := fresh in evar (n:nat) ;
     let v := eval compute in n in clear n ;
      let H := fresh in
        assert(H:Params m' v) by (subst m'; once typeclasses eauto) ;
          let v' := eval compute in v in subst m';
            (sk H v' || fail 1))
    || fk
  in
  let on_morphism m cont :=
    params m ltac:(fun H n => do_partial H n m)
      ltac:(cont)
  in
  match goal with
    | [ _ : normalization_done |- _ ] => fail 1
    | [ _ : @Params _ _ _ |- _ ] => fail 1
    | [ |- @Proper ?T _ (?m ?x) ] =>
      match goal with
        | [ H : PartialApplication |- _ ] =>
          class_apply @Reflexive_partial_app_morphism; [|clear H]
        | _ => on_morphism (m x)
          ltac:(class_apply @Reflexive_partial_app_morphism)
      end
  end.

Bootstrap !!!

Instance proper_proper {A} : Proper (relation_equivalence ==> eq ==> iff) (@Proper A).

Ltac proper_reflexive :=
  match goal with
    | [ _ : normalization_done |- _ ] => fail 1
    | _ => class_apply proper_eq || class_apply @reflexive_proper
  end.

#[global]
Hint Extern 1 (subrelation (flip _) _) => class_apply @flip1 : typeclass_instances.
#[global]
Hint Extern 1 (subrelation _ (flip _)) => class_apply @flip2 : typeclass_instances.

#[global]
Hint Extern 1 (Proper _ (complement _)) => apply @complement_proper
  : typeclass_instances.
#[global]
Hint Extern 1 (Proper _ (flip _)) => apply @flip_proper
  : typeclass_instances.
#[global]
Hint Extern 2 (@Proper _ (flip _) _) => class_apply @proper_flip_proper
  : typeclass_instances.
#[global]
Hint Extern 4 (@Proper _ _ _) => partial_application_tactic
  : typeclass_instances.
#[global]
Hint Extern 7 (@Proper _ _ _) => proper_reflexive
  : typeclass_instances.

Special-purpose class to do normalization of signatures w.r.t. flip.

Section Normalize.
  Context (A : Type).

  Class Normalizes (m : relation A) (m' : relation A) : Prop :=
    normalizes : relation_equivalence m m'.

Current strategy: add flip everywhere and reduce using subrelation afterwards.

  Lemma proper_normalizes_proper `(Normalizes R0 R1, Proper A R1 m) : Proper R0 m.

  Lemma flip_atom R : Normalizes R (flip (flip R)).

End Normalize.

Lemma flip_arrow {A : Type} {B : Type}
      `(NA : Normalizes A R (flip R'''), NB : Normalizes B R' (flip R'')) :
  Normalizes (A -> B) (R ==> R') (flip (R''' ==> R'')%signature).

Ltac normalizes :=
  match goal with
    | [ |- Normalizes _ (respectful _ _) _ ] => class_apply @flip_arrow
    | _ => class_apply @flip_atom
  end.

Ltac proper_normalization :=
  match goal with
    | [ _ : normalization_done |- _ ] => fail 1
    | [ _ : apply_subrelation |- @Proper _ ?R _ ] =>
      let H := fresh "H" in
      set(H:=did_normalization) ; class_apply @proper_normalizes_proper
  end.

#[global]
Hint Extern 1 (Normalizes _ _ _) => normalizes : typeclass_instances.
#[global]
Hint Extern 6 (@Proper _ _ _) => proper_normalization
  : typeclass_instances.

When the relation on the domain is symmetric, we can flip the relation on the codomain. Same for binary functions.
When the relation on the domain is symmetric, a predicate is compatible with iff as soon as it is compatible with impl. Same with a binary relation.
A PartialOrder is compatible with its underlying equivalence.
From a PartialOrder to the corresponding StrictOrder: lt = le /\ ~eq. If the order is total, we could also say gt = ~le.
From a StrictOrder to the corresponding PartialOrder: le = lt \/ eq. If the order is total, we could also say ge = ~lt.

Lemma StrictOrder_PreOrder
 `(Equivalence A eqA, StrictOrder A R, Proper _ (eqA==>eqA==>iff) R) :
 PreOrder (relation_disjunction R eqA).

#[global]
Hint Extern 4 (PreOrder (relation_disjunction _ _)) =>
  class_apply StrictOrder_PreOrder : typeclass_instances.

Lemma StrictOrder_PartialOrder
  `(Equivalence A eqA, StrictOrder A R, Proper _ (eqA==>eqA==>iff) R) :
  PartialOrder eqA (relation_disjunction R eqA).

#[global]
Hint Extern 4 (StrictOrder (relation_conjunction _ _)) =>
  class_apply PartialOrder_StrictOrder : typeclass_instances.

#[global]
Hint Extern 4 (PartialOrder _ (relation_disjunction _ _)) =>
  class_apply StrictOrder_PartialOrder : typeclass_instances.