AutoMore Automation


Require Export Imp.

Up to now, we've continued to use a quite restricted set of Coq's tactic facilities. In this chapter, we'll learn more about two very powerful features of Coq's tactic language: proof search via the auto and eauto tactics, and automated forward reasoning via the Ltac hypothesis matching machinery. Using these features together with Ltac's scripting facilities will enable us to make our proofs startlingly short! Used properly, they can also make proofs more maintainable and robust in the face of incremental changes to underlying definitions.
There's a third major source of automation we haven't fully studied yet, namely built-in decision procedures for specific kinds of problems: omega is one example, but there are others. This topic will be defered for a while longer.
Our motivating example will be this proof, repeated with just a few small changes from Imp. We will try to simplify this proof in several stages.

Ltac inv H := inversion H; subst; clear H.

Theorem ceval_deterministic: c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2.
  Case "E_Skip". reflexivity.
  Case "E_Ass". reflexivity.
  Case "E_Seq".
    assert (st' = st'0) as EQ1.
      SCase "Proof of assertion". apply IHE1_1; assumption.
    subst st'0.
    apply IHE1_2. assumption.
  Case "E_IfTrue".
    SCase "b evaluates to true".
      apply IHE1. assumption.
    SCase "b evaluates to false (contradiction)".
      rewrite H in H5. inversion H5.
  Case "E_IfFalse".
    SCase "b evaluates to true (contradiction)".
      rewrite H in H5. inversion H5.
    SCase "b evaluates to false".
      apply IHE1. assumption.
  Case "E_WhileEnd".
    SCase "b evaluates to false".
      reflexivity.
    SCase "b evaluates to true (contradiction)".
      rewrite H in H2. inversion H2.
  Case "E_WhileLoop".
    SCase "b evaluates to false (contradiction)".
      rewrite H in H4. inversion H4.
    SCase "b evaluates to true".
      assert (st' = st'0) as EQ1.
        SSCase "Proof of assertion". apply IHE1_1; assumption.
      subst st'0.
      apply IHE1_2. assumption. Qed.

The auto and eauto tactics

Thus far, we have (nearly) always written proof scripts that apply relevant hypothoses or lemmas by name. In particular, when a chain of hypothesis applications is needed, we have specified them explicitly. (The only exceptions introduced so far are using assumption to find a matching unqualified hypothesis or (e)constructor to find a matching constructor.)

Example auto_example_1 : (P Q R: Prop), (P Q) (Q R) P R.
Proof.
  intros P Q R H1 H2 H3.
  apply H2. apply H1. assumption.
Qed.

The auto tactic frees us from this drudgery by searching for a sequence of applications that will prove the goal

Example auto_example_1' : (P Q R: Prop), (P Q) (Q R) P R.
Proof.
  intros P Q R H1 H2 H3.
  auto.
Qed.

The auto tactic solves goals that are solvable by any combination of
  • intros,
  • apply (with a local hypothesis, by default).
The eauto tactic works just like auto, except that it uses eapply instead of apply.
A more complicated example:

Example auto_example_2 : P Q R S T U : Prop,
  (P Q)
  (P R)
  (T R)
  (S T U)
  ((PQ) (PS))
  T
  P
  U.
Proof. auto. Qed.

Search can take an arbitrarily long time, so there are limits to how far auto will search by default

Example auto_example_3 : (P Q R S T U: Prop),
                           (P Q) (Q R) (R S)
                           (S T) (T U) P U.
Proof.
  auto. (* When it cannot solve the goal, does nothing! *)
  auto 6. (* Optional argument says how deep to search (default depth is 5) *)
Qed.

auto and eauto consider the hypotheses in the current context together with a hint database of other lemmas and constructors
Some that we've already seen are installed by default.
  • e.g., eq_refl, conj, or_introl, and or_intror

Example auto_example_4 : P Q R : Prop,
  Q
  (Q R)
  P (Q R).
Proof.
  auto. Qed.

If we want to see which facts auto is using, we can use info_auto instead.

Example auto_example_5: 2 = 2.
Proof.
  info_auto. (* subsumes reflexivity because eq_refl is in hint database *)
Qed.

We can extend the hint database just for the purposes of one application of auto or eauto by writing auto using ....

Lemma le_antisym : n m: nat, (nm mn) n = m.
Proof. intros. omega. Qed.

Example auto_example_6 : n m p : nat,
  (np (nm mn))
  np
  n = m.
Proof.
  intros.
  auto. (* does nothing: auto doesn't destruct hypotheses! *)
  auto using le_antisym.
Qed.

Managing the hint database...
  • Hint Resolve T.
    Add theorem or constructor T to the global DB
  • Hint Constructors c.
    Add all constructors of c to the global DB
  • Hint Unfold d.
    Automatically expand defined symbol d during auto

Hint Resolve le_antisym.

Example auto_example_6' : n m p : nat,
  (np (nm mn))
  np
  n = m.
Proof.
  intros.
  auto. (* picks up hint from database *)
Qed.

Definition is_fortytwo x := x = 42.

Example auto_example_7: x, (x ≤ 42 42 ≤ x) is_fortytwo x.
Proof.
  auto. (* does nothing *)
Abort.

Hint Unfold is_fortytwo.

Example auto_example_7' : x, (x ≤ 42 42 ≤ x) is_fortytwo x.
Proof.
  info_auto.
Qed.

Hint Constructors ceval.

Definition st12 := update (update empty_state X 1) Y 2.
Definition st21 := update (update empty_state X 2) Y 1.

Example auto_example_8 : s',
  (IFB (BLe (AId X) (AId Y))
    THEN (Z ::= AMinus (AId Y) (AId X))
    ELSE (Y ::= APlus (AId X) (AId Z))
  FI) / st21 s'.
Proof.
  eexists. info_auto.
Qed.

Example auto_example_8' : s',
  (IFB (BLe (AId X) (AId Y))
    THEN (Z ::= AMinus (AId Y) (AId X))
    ELSE (Y ::= APlus (AId X) (AId Z))
  FI) / st12 s'.
Proof.
  eexists. info_auto.
Qed.

Now let's take a pass over ceval_deterministic using auto to simplify the proof script. We see that all simple sequences of hypothesis applications and all uses of reflexivity can be replaced by auto, which we add to the default tactic to be applied to each case.

Theorem ceval_deterministic': c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; auto.
  Case "E_Seq".
    assert (st' = st'0) as EQ1.
      SCase "Proof of assertion". auto.
    subst st'0.
    auto.
  Case "E_IfTrue".
    SCase "b evaluates to false (contradiction)".
      rewrite H in H5. inversion H5.
  Case "E_IfFalse".
    SCase "b evaluates to true (contradiction)".
      rewrite H in H5. inversion H5.
  Case "E_WhileEnd".
    SCase "b evaluates to true (contradiction)".
      rewrite H in H2. inversion H2.
  Case "E_WhileLoop".
    SCase "b evaluates to false (contradiction)".
      rewrite H in H4. inversion H4.
    SCase "b evaluates to true".
      assert (st' = st'0) as EQ1.
        SSCase "Proof of assertion". auto.
      subst st'0.
      auto. Qed.

Searching Hypotheses

The proof has become simpler, but there is still an annoying amount of repetition. Let's start by tackling the contradiction cases. Each of them occurs in a situation where we have both
H1: beval st b = false
and
H2: beval st b = true
as hypotheses. The contradiction is evident, but demonstrating it is a little complicated: we have to locate the two hypotheses H1 and H2 and do a rewrite following by an inversion. We'd like to automate this process.
Note: In fact, Coq has a built-in tactic congruence that will do the job. But we'll ignore the existence of this tactic for now, in order to demonstrate how to build forward search tactics by hand.
As a first step, we can abstract out the piece of script in question by writing a small amount of paramerized Ltac.

Ltac rwinv H1 H2 := rewrite H1 in H2; inv H2.

Theorem ceval_deterministic'': c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; auto.
  Case "E_Seq".
    assert (st' = st'0) as EQ1.
      SCase "Proof of assertion". auto.
    subst st'0.
    auto.
  Case "E_IfTrue".
    SCase "b evaluates to false (contradiction)".
      rwinv H H5.
  Case "E_IfFalse".
    SCase "b evaluates to true (contradiction)".
      rwinv H H5.
  Case "E_WhileEnd".
    SCase "b evaluates to true (contradiction)".
      rwinv H H2.
  Case "E_WhileLoop".
    SCase "b evaluates to false (contradiction)".
      rwinv H H4.
    SCase "b evaluates to true".
      assert (st' = st'0) as EQ1.
        SSCase "Proof of assertion". auto.
      subst st'0.
      auto. Qed.

But this is not much better. We really want Coq to discover the relevant hypotheses for us. We can do this by using the match goal with ... end facility of Ltac.

Ltac find_rwinv :=
  match goal with
    H1: ?E = true, H2: ?E = false _rwinv H1 H2
  end.

In words, this match goal looks for two (distinct) hypotheses that have the form of equalities with the same arbitrary expression E on the left and conflicting boolean values on the right; if such hypotheses are found, it binds H1 and H2 to their names, and applies the tactic after the .
Adding this tactic to our default string handles all the contradiction cases.

Theorem ceval_deterministic''': c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; try find_rwinv; auto.
  Case "E_Seq".
    assert (st' = st'0) as EQ1.
      SCase "Proof of assertion". auto.
    subst st'0.
    auto.
  Case "E_WhileLoop".
    SCase "b evaluates to true".
      assert (st' = st'0) as EQ1.
        SSCase "Proof of assertion". auto.
      subst st'0.
      auto. Qed.

Finally, let's see about the remaining cases. Each of them involves applying a conditional hypothesis to extract an equality. Currently we have phrased these as assertions, so that we have to predict what the resulting equality will be (although we can then use auto to prove it.) An alternative is to pick the relevant hypotheses to use, and then rewrite with them, as follows:

Theorem ceval_deterministic'''': c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; try find_rwinv; auto.
  Case "E_Seq".
    rewrite (IHE1_1 st'0 H1) in ×. auto.
  Case "E_WhileLoop".
    SCase "b evaluates to true".
      rewrite (IHE1_1 st'0 H3) in ×. auto. Qed.

Now we can automate the task of finding the relevant hypotheses to rewrite with.

Ltac find_eqn :=
  match goal with
    H1: x, ?P x ?L = ?R, H2: ?P ?X _
         rewrite (H1 X H2) in ×
  end.

But there are several pairs of hypotheses that have the right general form, and it seems tricky to pick out the ones we actually need. A key trick is to realize that we can try them all! Here's how this works:
  • rewrite will fail given a trivial equation of the form X = X.
  • each execution of match goal will keep trying to find a valid pair of hypotheses until the tactic on the RHS of the match succeeds; if there are no such pairs, it fails.
  • we can wrap the whole thing in a repeat which will keep doing useful rewrites until only trivial ones are left.


Theorem ceval_deterministic''''': c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; try find_rwinv; repeat find_eqn; auto.
  Qed.

The big pay-off in this approach is that our proof script should be robust in the face of modest changes to our language. For example, we can add a REPEAT command to the language. (This was an exercise in Hoare.v.)

Module Repeat.

Inductive com : Type :=
  | CSkip : com
  | CAsgn : id aexp com
  | CSeq : com com com
  | CIf : bexp com com com
  | CWhile : bexp com com
  | CRepeat : com bexp com.

REPEAT behaves like WHILE, except that the loop guard is checked after each execution of the body, with the loop repeating as long as the guard stays false. Because of this, the body will always execute at least once.

Tactic Notation "com_cases" tactic(first) ident(c) :=
  first;
  [ Case_aux c "SKIP" | Case_aux c "::=" | Case_aux c ";"
  | Case_aux c "IFB" | Case_aux c "WHILE"
  | Case_aux c "CRepeat" ].

Notation "'SKIP'" :=
  CSkip.
Notation "c1 ; c2" :=
  (CSeq c1 c2) (at level 80, right associativity).
Notation "X '::=' a" :=
  (CAsgn X a) (at level 60).
Notation "'WHILE' b 'DO' c 'END'" :=
  (CWhile b c) (at level 80, right associativity).
Notation "'IFB' e1 'THEN' e2 'ELSE' e3 'FI'" :=
  (CIf e1 e2 e3) (at level 80, right associativity).
Notation "'REPEAT' e1 'UNTIL' b2 'END'" :=
  (CRepeat e1 b2) (at level 80, right associativity).

Inductive ceval : state com state Prop :=
  | E_Skip : st,
      ceval st SKIP st
  | E_Ass : st a1 n X,
      aeval st a1 = n
      ceval st (X ::= a1) (update st X n)
  | E_Seq : c1 c2 st st' st'',
      ceval st c1 st'
      ceval st' c2 st''
      ceval st (c1 ; c2) st''
  | E_IfTrue : st st' b1 c1 c2,
      beval st b1 = true
      ceval st c1 st'
      ceval st (IFB b1 THEN c1 ELSE c2 FI) st'
  | E_IfFalse : st st' b1 c1 c2,
      beval st b1 = false
      ceval st c2 st'
      ceval st (IFB b1 THEN c1 ELSE c2 FI) st'
  | E_WhileEnd : b1 st c1,
      beval st b1 = false
      ceval st (WHILE b1 DO c1 END) st
  | E_WhileLoop : st st' st'' b1 c1,
      beval st b1 = true
      ceval st c1 st'
      ceval st' (WHILE b1 DO c1 END) st''
      ceval st (WHILE b1 DO c1 END) st''
  | E_RepeatEnd : st st' b1 c1,
      ceval st c1 st'
      beval st' b1 = true
      ceval st (CRepeat c1 b1) st'
  | E_RepeatLoop : st st' st'' b1 c1,
      ceval st c1 st'
      beval st' b1 = false
      ceval st' (CRepeat c1 b1) st''
      ceval st (CRepeat c1 b1) st''
.

Tactic Notation "ceval_cases" tactic(first) ident(c) :=
  first;
  [ Case_aux c "E_Skip" | Case_aux c "E_Ass"
  | Case_aux c "E_Seq"
  | Case_aux c "E_IfTrue" | Case_aux c "E_IfFalse"
  | Case_aux c "E_WhileEnd" | Case_aux c "E_WhileLoop"
  | Case_aux c "E_RepeatEnd" | Case_aux c "E_RepeatLoop"
].

Notation "c1 '/' st '' st'" := (ceval st c1 st')
                                 (at level 40, st at level 39).

Theorem ceval_deterministic: c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; try find_rwinv; repeat find_eqn; auto.
  Case "E_RepeatEnd".
    SCase "b evaluates to false (contradiction)".
       find_rwinv.
       (* oops: why didn't find_rwinv solve this for us already? 
          answer: we did things in the wrong order. *)

  case "E_RepeatLoop".
     SCase "b evaluates to true (contradiction)".
        find_rwinv.
Qed.

Theorem ceval_deterministic': c st st1 st2,
     c / st st1
     c / st st2
     st1 = st2.
Proof.
  intros c st st1 st2 E1 E2;
  generalize dependent st2;
  ceval_cases (induction E1) Case;
           intros st2 E2; inv E2; repeat find_eqn; try find_rwinv; auto.
Qed.

End Repeat.

These examples just give a flavor of what "hyper-automation" can do...
The details of using match goal are tricky, and debugging is not pleasant at all. But it is well worth adding at least simple uses to your proofs to avoid tedium and "future proof" your scripts.