Let test_stack_unification_interaction_with_delta A : (if negb _ then true else false) = if orb false (negb A) then true else false := eq_refl. (* Test patterns unification *) Lemma l1 : (forall P, (exists x:nat, P x) -> False) -> forall P, (exists x:nat, P x /\ P x) -> False. Proof. intros; apply (H _ H0). Qed. Lemma l2 : forall A:Set, forall Q:A->Set, (forall (P: forall x:A, Q x -> Prop), (exists x:A, exists y:Q x, P x y) -> False) -> forall (P: forall x:A, Q x -> Prop), (exists x:A, exists y:Q x, P x y /\ P x y) -> False. Proof. intros; apply (H _ H0). Qed. Lemma l3 : (forall P, ~(exists x:nat, P x)) -> forall P:nat->Prop, ~(exists x:nat, P x -> P x). Proof. intros; apply H. Qed. (* Feature introduced June 2011 *) Lemma l7 : forall x (P:nat->Prop), (forall f, P (f x)) -> P (x+x). Proof. intros x P H; apply H. Qed. (* Example submitted for Zenon *) Axiom zenon_noteq : forall T : Type, forall t : T, ((t <> t) -> False). Axiom zenon_notall : forall T : Type, forall P : T -> Prop, (forall z : T, (~(P z) -> False)) -> (~(forall x : T, (P x)) -> False). (* Must infer "P := fun x => x=x" in zenon_notall *) Check (fun _h1 => (zenon_notall nat _ (fun _T_0 => (fun _h2 => (zenon_noteq _ _T_0 _h2))) _h1)). (* Core of an example submitted by Ralph Matthes (BZ#849) It used to fail because of the K-variable x in the type of "sum_rec ..." which was not in the scope of the evar ?B. Solved by a head beta-reduction of the type "(fun _ : unit + unit => L unit) x" of "sum_rec ...". Shall we used more reduction when solving evars (in real_clean)?? Is there a risk of starting too long reductions? Note that the example originally came from a non re-typable pretty-printed term (the checked term is actually re-printed the same form it is checked). *) Set Implicit Arguments. Inductive L (A:Set) : Set := c : A -> L A. Parameter f: forall (A:Set)(B:Set), (A->B) -> L A -> L B. Parameter t: L (unit + unit). Check (f (fun x : unit + unit => sum_rec (fun _ : unit + unit => L unit) (fun y => c y) (fun y => c y) x) t). (* Test patterns unification in apply *) Require Import Arith. Parameter x y : nat. Parameter G:x=y->x=y->Prop. Parameter K:x<>y->x<>y->Prop. Lemma l4 : (forall f:x=y->Prop, forall g:x<>y->Prop, match eq_nat_dec x y with left a => f a | right a => g a end) -> match eq_nat_dec x y with left a => G a a | right a => K a a end. Proof. intros. apply H. Qed. (* Test unification modulo eta-expansion (if possible) *) (* In this example, two instances for ?P (argument of hypothesis H) can be inferred (one is by unifying the type [Q true] and [?P true] of the goal and type of [H]; the other is by unifying the argument of [f]); we need to unify both instances up to allowed eta-expansions of the instances (eta is allowed if the meta was applied to arguments) This used to fail before revision 9389 in trunk *) Lemma l5 : forall f : (forall P, P true), (forall P, f P = f P) -> forall Q, f (fun x => Q x) = f (fun x => Q x). Proof. intros. apply H. Qed. (* Feature deactivated in commit 14189 (see commit log) (* Test instantiation of evars by unification *) Goal (forall x, 0 + x = 0 -> True) -> True. intros; eapply H. rewrite <- plus_n_Sm. (* should refine ?x with S ?x' *) Abort. *) (* Check handling of identity equation between evars *) (* The example failed to pass until revision 10623 *) Lemma l6 : (forall y, (forall x, (forall z, y = 0 -> y + z = 0) -> y + x = 0) -> True) -> True. intros. eapply H. intros. apply H0. (* Check that equation ?n[H] = ?n[H] is correctly considered true *) reflexivity. Qed. (* Check treatment of metas erased by K-redexes at the time of turning them to evas *) Inductive nonemptyT (t : Type) : Prop := nonemptyT_intro : t -> nonemptyT t. Goal True. try case nonemptyT_intro. (* check that it fails w/o anomaly *) Abort. (* Test handling of return type and when it is decided to make the predicate dependent or not - see "bug" BZ#1851 *) Goal forall X (a:X) (f':nat -> X), (exists f : nat -> X, True). intros. exists (fun n => match n with O => a | S n' => f' n' end). constructor. Qed. (* Check use of types in unification (see Andrej Bauer's mail on coq-club, June 1 2009; it did not work in 8.2, probably started to work after Sozeau improved support for the use of types in unification) *) Goal (forall (A B : Set) (f : A -> B), (fun x => f x) = f) -> forall (A B C : Set) (g : (A -> B) -> C) (f : A -> B), g (fun x => f x) = g f. Proof. intros. rewrite H with (f:=f0). Abort. (* Three tests provided by Dan Grayson as part of a custom patch he made for a more powerful "destruct" for handling Voevodsky's Univalent Foundations. The test checks if second-order matching in tactic unification is able to guess by itself on which dependent terms to abstract so that the elimination predicate is well-typed *) Definition test1 (X : Type) (x : X) (fxe : forall x1 : X, identity x1 x1) : identity (fxe x) (fxe x). Proof. destruct (fxe x). apply identity_refl. Defined. (* a harder example *) Definition UU := Type . Inductive paths {T:Type}(t:T): T -> UU := idpath: paths t t. Inductive foo (X0:UU) (x0:X0) : forall (X:UU)(x:X), UU := newfoo : foo x0 x0. Definition idonfoo {X0:UU} {x0:X0} {X1:UU} {x1:X1} : foo x0 x1 -> foo x0 x1. Proof. intros t. exact t. Defined. Lemma test2 (T:UU) (t:T) (k : foo t t) : paths k (idonfoo k). Proof. destruct k. apply idpath. Defined. (* an example with two constructors *) Inductive foo' (X0:UU) (x0:X0) : forall (X:UU)(x:X), UU := | newfoo1 : foo' x0 x0 | newfoo2 : foo' x0 x0 . Definition idonfoo' {X0:UU} {x0:X0} {X1:UU} {x1:X1} : foo' x0 x1 -> foo' x0 x1. Proof. intros t. exact t. Defined. Lemma test3 (T:UU) (t:T) (k : foo' t t) : paths k (idonfoo' k). Proof. destruct k. apply idpath. apply idpath. Defined. (* An example where it is necessary to evar-normalize the instance of an evar to evaluate if it is a pattern *) Check let a := ?[P] in fun (H : forall y (P : nat -> Prop), y = 0 -> P y) x (p:x=0) => H ?[y] a p : x = 0. (* We have to solve "?P ?y[x] == x = 0" knowing from "p : (x=0) == (?y[x] = 0)" that "?y := x" *)