tactics.v

(* Copyright (c) 2012-2015, Robbert Krebbers. *)
(* This file is distributed under the terms of the BSD license. *)
(** This file collects general purpose tactics that are used throughout
the development. *)
From Coq Require Import Omega.
From Coq Require Export Psatz.
From stdpp Require Export decidable.

Lemma f_equal_dep {A B} (f g : ∀ x : A, B x) x : f = g → f x = g x.
Proof. intros ->; reflexivity. Qed.
Lemma f_equal_help {A B} (f g : A → B) x y : f = g → x = y → f x = g y.
Proof. intros -> ->; reflexivity. Qed.
Ltac f_equal :=
  let rec go :=
    match goal with
    | _ => reflexivity
    | _ => apply f_equal_help; [go|try reflexivity]
    | |- ?f ?x = ?g ?x => apply (f_equal_dep f g); go
    end in
  try go.

(** We declare hint databases [f_equal], [congruence] and [lia] and containing
solely the tactic corresponding to its name. These hint database are useful in
to be combined in combination with other hint database. *)
Hint Extern 998 (_ = _) => f_equal : f_equal.
Hint Extern 999 => congruence : congruence.
Hint Extern 1000 => lia : lia.
Hint Extern 1000 => omega : omega.
Hint Extern 1001 => progress subst : subst. (** backtracking on this one will
be very bad, so use with care! *)

(** The tactic [intuition] expands to [intuition auto with *] by default. This
is rather efficient when having big hint databases, or expensive [Hint Extern]
declarations as the ones above. *)
Tactic Notation "intuition" := intuition auto.

(* [done] can get slow as it calls "trivial". [fast_done] can solve way less
   goals, but it will also always finish quickly. *)
Ltac fast_done :=
  solve [ reflexivity | eassumption | symmetry; eassumption ].
Tactic Notation "fast_by" tactic(tac) :=
  tac; fast_done.

(** A slightly modified version of Ssreflect's finishing tactic [done]. It
also performs [reflexivity] and uses symmetry of negated equalities. Compared
to Ssreflect's [done], it does not compute the goal's [hnf] so as to avoid
unfolding setoid equalities. Note that this tactic performs much better than
Coq's [easy] tactic as it does not perform [inversion]. *)
Ltac done :=
  trivial; intros; solve
  [ repeat first
    [ fast_done
    | solve [trivial]
    | solve [symmetry; trivial]
    | discriminate
    | contradiction
    | solve [apply not_symmetry; trivial]
    | split ]
  | match goal with H : ¬_ |- _ => solve [case H; trivial] end ].
Tactic Notation "by" tactic(tac) :=
  tac; done.

(** Aliases for trans and etrans that are easier to type *)
Tactic Notation "trans" constr(A) := transitivity A.
Tactic Notation "etrans" := etransitivity.

(** Tactics for splitting conjunctions:

- [split_and] : split the goal if is syntactically of the shape [_ ∧ _]
- [split_ands?] : split the goal repeatedly (perhaps zero times) while it is
  of the shape [_ ∧ _].
- [split_ands!] : works similarly, but at least one split should succeed. In
  order to do so, it will head normalize the goal first to possibly expose a
  conjunction.

Note that [split_and] differs from [split] by only splitting conjunctions. The
[split] tactic splits any inductive with one constructor. *)
Tactic Notation "split_and" :=
  match goal with
  | |- _ ∧ _ => split
  | |- Is_true (_ && _) => apply andb_True; split
  end.
Tactic Notation "split_and" "?" := repeat split_and.
Tactic Notation "split_and" "!" := hnf; split_and; split_and?.

Tactic Notation "destruct_and" "?" :=
  repeat match goal with
  | H : False |- _ => destruct H
  | H : _ ∧ _ |- _ => destruct H
  | H : Is_true (bool_decide _) |- _ => apply (bool_decide_unpack _) in H
  | H : Is_true (_ && _) |- _ => apply andb_True in H; destruct H
  end.
Tactic Notation "destruct_and" "!" := progress (destruct_and?).

(** The tactic [case_match] destructs an arbitrary match in the conclusion or
assumptions, and generates a corresponding equality. This tactic is best used
together with the [repeat] tactical. *)
Ltac case_match :=
  match goal with
  | H : context [ match ?x with _ => _ end ] |- _ => destruct x eqn:?
  | |- context [ match ?x with _ => _ end ] => destruct x eqn:?
  end.

(** The tactic [unless T by tac_fail] succeeds if [T] is not provable by
the tactic [tac_fail]. *)
Tactic Notation "unless" constr(T) "by" tactic3(tac_fail) :=
  first [assert T by tac_fail; fail 1 | idtac].

(** The tactic [repeat_on_hyps tac] repeatedly applies [tac] in unspecified
order on all hypotheses until it cannot be applied to any hypothesis anymore. *)
Tactic Notation "repeat_on_hyps" tactic3(tac) :=
  repeat match goal with H : _ |- _ => progress tac H end.

(** The tactic [clear dependent H1 ... Hn] clears the hypotheses [Hi] and
their dependencies. *)
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) :=
  clear dependent H1; clear dependent H2.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) :=
  clear dependent H1 H2; clear dependent H3.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4) :=
  clear dependent H1 H2 H3; clear dependent H4.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4)
  hyp(H5) := clear dependent H1 H2 H3 H4; clear dependent H5.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4) hyp(H5)
  hyp (H6) := clear dependent H1 H2 H3 H4 H5; clear dependent H6.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4) hyp(H5)
  hyp (H6) hyp(H7) := clear dependent H1 H2 H3 H4 H5 H6; clear dependent H7.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4) hyp(H5)
  hyp (H6) hyp(H7) hyp(H8) :=
  clear dependent H1 H2 H3 H4 H5 H6 H7; clear dependent H8.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4) hyp(H5)
  hyp (H6) hyp(H7) hyp(H8) hyp(H9) :=
  clear dependent H1 H2 H3 H4 H5 H6 H7 H8; clear dependent H9.
Tactic Notation "clear" "dependent" hyp(H1) hyp(H2) hyp(H3) hyp(H4) hyp(H5)
  hyp (H6) hyp(H7) hyp(H8) hyp(H9) hyp(H10) :=
  clear dependent H1 H2 H3 H4 H5 H6 H7 H8 H9; clear dependent H10.

(** The tactic [is_non_dependent H] determines whether the goal's conclusion or
hypotheses depend on [H]. *)
Tactic Notation "is_non_dependent" constr(H) :=
  match goal with
  | _ : context [ H ] |- _ => fail 1
  | |- context [ H ] => fail 1
  | _ => idtac
  end.

(** The tactic [var_eq x y] fails if [x] and [y] are unequal, and [var_neq]
does the converse. *)
Ltac var_eq x1 x2 := match x1 with x2 => idtac | _ => fail 1 end.
Ltac var_neq x1 x2 := match x1 with x2 => fail 1 | _ => idtac end.

(** Operational type class projections in recursive calls are not folded back
appropriately by [simpl]. The tactic [csimpl] uses the [fold_classes] tactics
to refold recursive calls of [fmap], [mbind], [omap] and [alter]. A
self-contained example explaining the problem can be found in the following
Coq-club message:

https://sympa.inria.fr/sympa/arc/coq-club/2012-10/msg00147.html *)
Ltac fold_classes :=
  repeat match goal with
  | |- appcontext [ ?F ] =>
    progress match type of F with
    | FMap _ =>
       change F with (@fmap _ F);
       repeat change (@fmap _ (@fmap _ F)) with (@fmap _ F)
    | MBind _ =>
       change F with (@mbind _ F);
       repeat change (@mbind _ (@mbind _ F)) with (@mbind _ F)
    | OMap _ =>
       change F with (@omap _ F);
       repeat change (@omap _ (@omap _ F)) with (@omap _ F)
    | Alter _ _ _ =>
       change F with (@alter _ _ _ F);
       repeat change (@alter _ _ _ (@alter _ _ _ F)) with (@alter _ _ _ F)
    end
  end.
Ltac fold_classes_hyps H :=
  repeat match type of H with
  | appcontext [ ?F ] =>
    progress match type of F with
    | FMap _ =>
       change F with (@fmap _ F) in H;
       repeat change (@fmap _ (@fmap _ F)) with (@fmap _ F) in H
    | MBind _ =>
       change F with (@mbind _ F) in H;
       repeat change (@mbind _ (@mbind _ F)) with (@mbind _ F) in H
    | OMap _ =>
       change F with (@omap _ F) in H;
       repeat change (@omap _ (@omap _ F)) with (@omap _ F) in H
    | Alter _ _ _ =>
       change F with (@alter _ _ _ F) in H;
       repeat change (@alter _ _ _ (@alter _ _ _ F)) with (@alter _ _ _ F) in H
    end
  end.
Tactic Notation "csimpl" "in" hyp(H) :=
  try (progress simpl in H; fold_classes_hyps H).
Tactic Notation "csimpl" := try (progress simpl; fold_classes).
Tactic Notation "csimpl" "in" "*" :=
  repeat_on_hyps (fun H => csimpl in H); csimpl.

(** The tactic [simplify_eq] repeatedly substitutes, discriminates,
and injects equalities, and tries to contradict impossible inequalities. *)
Tactic Notation "simplify_eq" := repeat
  match goal with
  | H : _ ≠ _ |- _ => by destruct H
  | H : _ = _ → False |- _ => by destruct H
  | H : ?x = _ |- _ => subst x
  | H : _ = ?x |- _ => subst x
  | H : _ = _ |- _ => discriminate H
  | H : _ ≡ _ |- _ => apply leibniz_equiv in H
  | H : ?f _ = ?f _ |- _ => apply (inj f) in H
  | H : ?f _ _ = ?f _ _ |- _ => apply (inj2 f) in H; destruct H
    (* before [injection] to circumvent bug #2939 in some situations *)
  | H : ?f _ = ?f _ |- _ => progress injection H as H
    (* first hyp will be named [H], subsequent hyps will be given fresh names *)
  | H : ?f _ _ = ?f _ _ |- _ => progress injection H as H
  | H : ?f _ _ _ = ?f _ _ _ |- _ => progress injection H as H
  | H : ?f _ _ _ _ = ?f _ _ _ _ |- _ => progress injection H as H
  | H : ?f _ _ _ _ _ = ?f _ _ _ _ _ |- _ => progress injection H as H
  | H : ?f _ _ _ _ _ _ = ?f _ _ _ _ _ _ |- _ => progress injection H as H
  | H : ?x = ?x |- _ => clear H
    (* unclear how to generalize the below *)
  | H1 : ?o = Some ?x, H2 : ?o = Some ?y |- _ =>
    assert (y = x) by congruence; clear H2
  | H1 : ?o = Some ?x, H2 : ?o = None |- _ => congruence
  | H : @existT ?A _ _ _ = existT _ _ |- _ =>
     apply (Eqdep_dec.inj_pair2_eq_dec _ (decide_rel (@eq A))) in H
  end.
Tactic Notation "simplify_eq" "/=" :=
  repeat (progress csimpl in * || simplify_eq).
Tactic Notation "f_equal" "/=" := csimpl in *; f_equal.

Ltac setoid_subst_aux R x :=
  match goal with
  | H : R x ?y |- _ =>
     is_var x;
     try match y with x _ => fail 2 end;
     repeat match goal with
     | |- context [ x ] => setoid_rewrite H
     | H' : context [ x ] |- _ =>
        try match H' with H => fail 2 end;
        setoid_rewrite H in H'
     end;
     clear x H
  end.
Ltac setoid_subst :=
  repeat match goal with
  | _ => progress simplify_eq/=
  | H : @equiv ?A ?e ?x _ |- _ => setoid_subst_aux (@equiv A e) x
  | H : @equiv ?A ?e _ ?x |- _ => symmetry in H; setoid_subst_aux (@equiv A e) x
  end.

(** f_equiv works on goals of the form [f _ = f _], for any relation and any
number of arguments. It looks for an appropriate [Proper] instance, and applies
it. The tactic is somewhat limited, since it cannot be used to backtrack on
the Proper instances that has been found. To that end, we try to ensure the
trivial instance in which the resulting goals have an [eq]. More generally,
when having [Proper (equiv ==> dist) f] and [Proper (dist ==> dist) f], it will
favor the second. *)
Ltac f_equiv :=
  match goal with
  | _ => reflexivity
  (* We support matches on both sides, *if* they concern the same variable, or
     variables in some relation. *)
  | |- ?R (match ?x with _ => _ end) (match ?x with _ => _ end) =>
    destruct x
  | H : ?R ?x ?y |- ?R2 (match ?x with _ => _ end) (match ?y with _ => _ end) =>
     destruct H
  (* First assume that the arguments need the same relation as the result *)
  | |- ?R (?f ?x) (?f _) => apply (_ : Proper (R ==> R) f)
  (* For the case in which R is polymorphic, or an operational type class,
  like equiv. *)
  | |- (?R _) (?f ?x) (?f _) => apply (_ : Proper (R _ ==> _) f)
  | |- (?R _ _) (?f ?x) (?f _) => apply (_ : Proper (R _ _ ==> _) f)
  | |- (?R _ _ _) (?f ?x) (?f _) => apply (_ : Proper (R _ _ _ ==> _) f)
  | |- (?R _) (?f ?x ?y) (?f _ _) => apply (_ : Proper (R _ ==> R _ ==> _) f)
  | |- (?R _ _) (?f ?x ?y) (?f _ _) => apply (_ : Proper (R _ _ ==> R _ _ ==> _) f)
  | |- (?R _ _ _) (?f ?x ?y) (?f _ _) => apply (_ : Proper (R _ _ _ ==> R _ _ _ ==> _) f)
  | |- (?R _ _ _ _) (?f ?x ?y) (?f _ _) => apply (_ : Proper (R _ _ _ _ ==> R _ _ _ _ ==> _) f)
  (* Next, try to infer the relation. Unfortunately, there is an instance
     of Proper for (eq ==> _), which will always be matched. *)
  (* TODO: Can we exclude that instance? *)
  (* TODO: If some of the arguments are the same, we could also
     query for "pointwise_relation"'s. But that leads to a combinatorial
     explosion about which arguments are and which are not the same. *)
  | |- ?R (?f ?x) (?f _) => apply (_ : Proper (_ ==> R) f)
  | |- ?R (?f ?x ?y) (?f _ _) => apply (_ : Proper (_ ==> _ ==> R) f)
   (* In case the function symbol differs, but the arguments are the same,
      maybe we have a pointwise_relation in our context. *)
  | H : pointwise_relation _ ?R ?f ?g |- ?R (?f ?x) (?g ?x) => apply H
  end.

(** auto_proper solves goals of the form "f _ = f _", for any relation and any
    number of arguments, by repeatedly apply f_equiv and handling trivial cases.
    If it cannot solve an equality, it will leave that to the user. *)
Ltac auto_proper :=
  (* Deal with "pointwise_relation" *)
  repeat lazymatch goal with
  | |- pointwise_relation _ _ _ _ => intros ?
  end;
  (* Normalize away equalities. *)
  simplify_eq;
  (* repeatedly apply congruence lemmas and use the equalities in the hypotheses. *)
  try (f_equiv; fast_done || auto_proper).

(** solve_proper solves goals of the form "Proper (R1 ==> R2)", for any
    number of relations. All the actual work is done by f_equiv;
    solve_proper just introduces the assumptions and unfolds the first
    head symbol. *)
Ltac solve_proper :=
  (* Introduce everything *)
  intros;
  repeat lazymatch goal with
  | |- Proper _ _ => intros ???
  | |- (_ ==> _)%signature _ _ => intros ???
  | |- pointwise_relation _ _ _ _ => intros ?
  | |- ?R ?f _ => try let f' := constr:(λ x, f x) in intros ?
  end; simpl;
  (* Unfold the head symbol, which is the one we are proving a new property about *)
  lazymatch goal with
  | |- ?R (?f _ _ _ _ _ _ _ _) (?f _ _ _ _ _ _ _ _) => unfold f
  | |- ?R (?f _ _ _ _ _ _ _) (?f _ _ _ _ _ _ _) => unfold f
  | |- ?R (?f _ _ _ _ _ _) (?f _ _ _ _ _ _) => unfold f
  | |- ?R (?f _ _ _ _ _) (?f _ _ _ _ _) => unfold f
  | |- ?R (?f _ _ _ _) (?f _ _ _ _) => unfold f
  | |- ?R (?f _ _ _) (?f _ _ _) => unfold f
  | |- ?R (?f _ _) (?f _ _) => unfold f
  | |- ?R (?f _) (?f _) => unfold f
  end;
  solve [ auto_proper ].

(** The tactic [intros_revert tac] introduces all foralls/arrows, performs tac,
and then reverts them. *)
Ltac intros_revert tac :=
  lazymatch goal with
  | |- ∀ _, _ => let H := fresh in intro H; intros_revert tac; revert H
  | |- _ => tac
  end.

(** Given a tactic [tac2] generating a list of terms, [iter tac1 tac2]
runs [tac x] for each element [x] until [tac x] succeeds. If it does not
suceed for any element of the generated list, the whole tactic wil fail. *)
Tactic Notation "iter" tactic(tac) tactic(l) :=
  let rec go l :=
  match l with ?x :: ?l => tac x || go l end in go l.

(** Given H : [A_1 → ... → A_n → B] (where each [A_i] is non-dependent), the
tactic [feed tac H tac_by] creates a subgoal for each [A_i] and calls [tac p]
with the generated proof [p] of [B]. *)
Tactic Notation "feed" tactic(tac) constr(H) :=
  let rec go H :=
  let T := type of H in
  lazymatch eval hnf in T with
  | ?T1 → ?T2 =>
    (* Use a separate counter for fresh names to make it more likely that
    the generated name is "fresh" with respect to those generated before
    calling the [feed] tactic. In particular, this hack makes sure that
    tactics like [let H' := fresh in feed (fun p => pose proof p as H') H] do
    not break. *)
    let HT1 := fresh "feed" in assert T1 as HT1;
      [| go (H HT1); clear HT1 ]
  | ?T1 => tac H
  end in go H.

(** The tactic [efeed tac H] is similar to [feed], but it also instantiates
dependent premises of [H] with evars. *)
Tactic Notation "efeed" constr(H) "using" tactic3(tac) "by" tactic3 (bytac) :=
  let rec go H :=
  let T := type of H in
  lazymatch eval hnf in T with
  | ?T1 → ?T2 =>
    let HT1 := fresh "feed" in assert T1 as HT1;
      [bytac | go (H HT1); clear HT1 ]
  | ?T1 → _ =>
    let e := fresh "feed" in evar (e:T1);
    let e' := eval unfold e in e in
    clear e; go (H e')
  | ?T1 => tac H
  end in go H.
Tactic Notation "efeed" constr(H) "using" tactic3(tac) :=
  efeed H using tac by idtac.

(** The following variants of [pose proof], [specialize], [inversion], and
[destruct], use the [feed] tactic before invoking the actual tactic. *)
Tactic Notation "feed" "pose" "proof" constr(H) "as" ident(H') :=
  feed (fun p => pose proof p as H') H.
Tactic Notation "feed" "pose" "proof" constr(H) :=
  feed (fun p => pose proof p) H.

Tactic Notation "efeed" "pose" "proof" constr(H) "as" ident(H') :=
  efeed H using (fun p => pose proof p as H').
Tactic Notation "efeed" "pose" "proof" constr(H) :=
  efeed H using (fun p => pose proof p).

Tactic Notation "feed" "specialize" hyp(H) :=
  feed (fun p => specialize p) H.
Tactic Notation "efeed" "specialize" hyp(H) :=
  efeed H using (fun p => specialize p).

Tactic Notation "feed" "inversion" constr(H) :=
  feed (fun p => let H':=fresh in pose proof p as H'; inversion H') H.
Tactic Notation "feed" "inversion" constr(H) "as" simple_intropattern(IP) :=
  feed (fun p => let H':=fresh in pose proof p as H'; inversion H' as IP) H.

Tactic Notation "feed" "destruct" constr(H) :=
  feed (fun p => let H':=fresh in pose proof p as H'; destruct H') H.
Tactic Notation "feed" "destruct" constr(H) "as" simple_intropattern(IP) :=
  feed (fun p => let H':=fresh in pose proof p as H'; destruct H' as IP) H.


(** The following tactic can be used to add support for patterns to tactic notation:
It will search for the first subterm of the goal matching [pat], and then call [tac]
with that subterm. *)
Ltac find_pat pat tac :=
  match goal with
  |- context [?x] =>
      unify pat x with typeclass_instances;
      tryif tac x then idtac else fail 2
  end.

(** Coq's [firstorder] tactic fails or loops on rather small goals already. In 
particular, on those generated by the tactic [unfold_elem_ofs] which is used
to solve propositions on collections. The [naive_solver] tactic implements an
ad-hoc and incomplete [firstorder]-like solver using Ltac's backtracking
mechanism. The tactic suffers from the following limitations:
- It might leave unresolved evars as Ltac provides no way to detect that.
- To avoid the tactic becoming too slow, we allow a universally quantified
  hypothesis to be instantiated only once during each search path.
- It does not perform backtracking on instantiation of universally quantified
  assumptions.

We use a counter to make the search breath first. Breath first search ensures
that a minimal number of hypotheses is instantiated, and thus reduced the
posibility that an evar remains unresolved.

Despite these limitations, it works much better than Coq's [firstorder] tactic
for the purposes of this development. This tactic either fails or proves the
goal. *)
Lemma forall_and_distr (A : Type) (P Q : A → Prop) :
  (∀ x, P x ∧ Q x) ↔ (∀ x, P x) ∧ (∀ x, Q x).
Proof. firstorder. Qed.

(** The tactic [no_new_unsolved_evars tac] executes [tac] and fails if it
creates any new evars. This trick is by Jonathan Leivent, see:
https://coq.inria.fr/bugs/show_bug.cgi?id=3872 *)

Ltac no_new_unsolved_evars tac := exact ltac:(tac).

Tactic Notation "naive_solver" tactic(tac) :=
  unfold iff, not in *;
  repeat match goal with
  | H : context [∀ _, _ ∧ _ ] |- _ =>
    repeat setoid_rewrite forall_and_distr in H; revert H
  end;
  let rec go n :=
  repeat match goal with
  (**i intros *)
  | |- ∀ _, _ => intro
  (**i simplification of assumptions *)
  | H : False |- _ => destruct H
  | H : _ ∧ _ |- _ => destruct H
  | H : ∃ _, _  |- _ => destruct H
  | H : ?P → ?Q, H2 : ?P |- _ => specialize (H H2)
  | H : Is_true (bool_decide _) |- _ => apply (bool_decide_unpack _) in H
  | H : Is_true (_ && _) |- _ => apply andb_True in H; destruct H
  (**i simplify and solve equalities *)
  | |- _ => progress simplify_eq/=
  (**i solve the goal *)
  | |- _ =>
    solve
    [ eassumption
    | symmetry; eassumption
    | apply not_symmetry; eassumption
    | reflexivity ]
  (**i operations that generate more subgoals *)
  | |- _ ∧ _ => split
  | |- Is_true (bool_decide _) => apply (bool_decide_pack _)
  | |- Is_true (_ && _) => apply andb_True; split
  | H : _ ∨ _ |- _ => destruct H
  (**i solve the goal using the user supplied tactic *)
  | |- _ => solve [tac]
  end;
  (**i use recursion to enable backtracking on the following clauses. *)
  match goal with
  (**i instantiation of the conclusion *)
  | |- ∃ x, _ => no_new_unsolved_evars ltac:(eexists; go n)
  | |- _ ∨ _ => first [left; go n | right; go n]
  | _ =>
    (**i instantiations of assumptions. *)
    lazymatch n with
    | S ?n' =>
      (**i we give priority to assumptions that fit on the conclusion. *)
      match goal with
      | H : _ → _ |- _ =>
        is_non_dependent H;
        no_new_unsolved_evars
          ltac:(first [eapply H | efeed pose proof H]; clear H; go n')
      end
    end
  end
  in iter (fun n' => go n') (eval compute in (seq 1 6)).
Tactic Notation "naive_solver" := naive_solver eauto.