frap/Polymorphism.v
2017-02-09 12:54:27 -05:00

416 lines
11 KiB
Coq

(** Formal Reasoning About Programs <http://adam.chlipala.net/frap/>
* Supplementary Coq material: polymorphism and generic data structures
* Author: Adam Chlipala
* License: https://creativecommons.org/licenses/by-nc-nd/4.0/ *)
Require Import Frap.
Set Implicit Arguments.
(* This command sets up automatic inference of tedious arguments. *)
(* One of the recurring ingredients in effective machine formalization is data
* structures, and, as in general programming, we want to be able to define data
* structures that are *generic* in the kinds of data they contain. Coq
* supports such things in roughly the same way as Haskell and OCaml, via
* *parametric polymorphism*, which is closely related to the idea of *generics*
* in languages like Java. *)
(* Our first example: the [option] type family. While Java and friends force
* all sorts of different types to include the special value [null], in Coq we
* request that option explicitly by wrapping a type in [option]. Specifically,
* any value of type [option A], for some type [A], is either [None] (sort of
* like [null]) or [Some v] for a [v] of type [A]. *)
Inductive option (A : Set) : Set :=
| None
| Some (v : A).
Arguments None [A].
(* This command asks Coq to *infer* the [A] type for each specific use of
* [None]. *)
(* Here are a few example terms using [option]. *)
Example no_number : option nat := None.
Example a_number : option nat := Some 42.
Example no_number_squared : option (option nat) := None.
Example no_number_squared_inside : option (option nat) := Some None.
Example a_number_squared : option (option nat) := Some (Some 42).
(* Pattern matching is the key ingredient for working with inductive definitions
* of all sorts. Here are some examples matching on [option]s. *)
Definition increment_optional (no : option nat) : option nat :=
match no with
| None => None
| Some n => Some (n + 1)
end.
(* Here we use type [A * B] of *pairs*, inhabited by values [(a, b)], with
* [a : A] and [b : B]. *)
Definition add_optional (po : option (nat * nat)) : option nat :=
match po with
| None => None
| Some (n, m) => Some (n + m)
end.
(** * Lists *)
(* For functional programming (as in Coq), the king of all generic data
* structures is the *list*, which you explored a bit in the first problem set.
* Let's recap that type definition. *)
Inductive list (A : Set) : Set :=
| nil
| cons (hd : A) (tl : list A).
Arguments nil [A].
(* [nil] is the empty list, while [cons], standing for "construct," extends a
* list of length [n] into one of length [n+1]. *)
(* Here are some simple lists. *)
Example nats0 : list nat := nil.
Example nats1 : list nat := cons 1 nil.
Example nats2 : list nat := cons 1 (cons 2 nil).
(* Coq features a wonderful notation system, to help us write more concise and
* readable code after introducing new syntactic forms. We will not give a
* systematic presentation of the notation system, but we will show many
* examples, from which it is possible to infer generality by scientific
* induction. And, of course, the interested reader can always check the
* notations chapter of the Coq reference manual. *)
(* First, our examples can get more readable with an infix operator for [cons]. *)
Infix "::" := cons.
Example nats1' : list nat := 1 :: nil.
Example nats2' : list nat := 1 :: 2 :: nil.
(* Getting even more fancy, we declare a notation for list literals. *)
Notation "[ ]" := nil.
Notation "[ x1 ; .. ; xN ]" := (cons x1 (.. (cons xN nil) ..)).
Example nats0'' : list nat := [].
Example nats1'' : list nat := [1].
Example nats2'' : list nat := [1; 2].
Example nats3'' : list nat := [1; 2; 3].
(* Here are some classic recursive functions that operate over lists.
* First, here is how to compute the length of a list. Recall that we put
* *implicit* function arguments in curly braces, asking Coq to infer them at
* call sites. *)
Fixpoint length {A} (ls : list A) : nat :=
match ls with
| nil => 0
| _ :: ls' => 1 + length ls'
end.
(* The first problem set involved an exercise with list append and reverse
* operations. To avoid spoiling the proofs there about those functions, we
* will give their definitions here without proving the classic theorems from
* the problem set. *)
Fixpoint app {A} (ls1 ls2 : list A) : list A :=
match ls1 with
| nil => ls2
| x :: ls1' => x :: app ls1' ls2
end.
Infix "++" := app.
Fixpoint rev {A} (ls : list A) : list A :=
match ls with
| nil => nil
| x :: ls' => rev ls' ++ [x]
end.
Theorem length_app : forall A (ls1 ls2 : list A),
length (ls1 ++ ls2) = length ls1 + length ls2.
Admitted.
(* One of the classic gotchas in functional-programming class is how slow this
* naive [rev] is. Each [app] operation requires linear time, so running
* linearly many [app]s brings us to quadratic time for [rev]. Using a helper
* function, we can bring [rev] to its optimal linear time. *)
Fixpoint rev_append {A} (ls acc : list A) : list A :=
match ls with
| nil => acc
| x :: ls' => rev_append ls' (x :: acc)
end.
(* This function [rev_append] takes an extra *accumulator* argument, in which we
* gradually build up the original input in reversed order. The base case just
* returns the accumulator. Now reversal just needs to do a [rev_append] with
* an empty initial accumulator. *)
Definition rev' {A} (ls : list A) : list A :=
rev_append ls [].
(* A few test cases can help convince us that this seems to work. *)
Compute rev [1; 2; 3; 4].
Compute rev' [1; 2; 3; 4].
Compute rev ["hi"; "bye"; "sky"].
Compute rev' ["hi"; "bye"; "sky"].
(* OK, great. Now it seems worth investing in a correctness proof. We'll
* discover it interactively in class, but here's a worked-out final
* answer, with the several lemmas that we discover are useful. *)
(* List concatenation is associative. *)
Lemma app_assoc : forall A (ls1 ls2 ls3 : list A),
(ls1 ++ ls2) ++ ls3 = ls1 ++ (ls2 ++ ls3).
Proof.
induct ls1; simplify.
equality.
rewrite IHls1.
equality.
Qed.
(* The natural correctness condition for [rev_append]: it does what it says on
* the package, combining reversal with appending! *)
Lemma rev_append_ok : forall A (ls acc : list A),
rev_append ls acc = rev ls ++ acc.
Proof.
induct ls; simplify.
equality.
rewrite IHls.
simplify.
rewrite app_assoc.
simplify.
equality.
Qed.
(* Concatenating the empty list has no effect. *)
Lemma app_nil : forall A (ls : list A),
ls ++ [] = ls.
Proof.
induct ls; simplify.
equality.
equality.
Qed.
(* Now we can prove equivalence of [rev'] and [rev], with no new induction. *)
Theorem rev'_ok : forall A (ls : list A),
rev' ls = rev ls.
Proof.
simplify.
unfold rev'.
rewrite rev_append_ok.
apply app_nil.
Qed.
(** ** Zipping and unzipping *)
(* Another classic pair of list operations is zipping and unzipping.
* These functions convert between pairs of lists and lists of pairs. *)
Fixpoint zip {A1 A2} (ls1 : list A1) (ls2 : list A2) : list (A1 * A2) :=
match ls1, ls2 with
| x1 :: ls1', x2 :: ls2' => (x1, x2) :: zip ls1' ls2'
| _, _ => []
end.
(* Note how, when passed two lengths of different lists, [zip] drops the
* mismatched suffix of the longer list. *)
(* An explicit [Set] annotation is needed here, for obscure type-inference
* reasons. *)
Fixpoint unzip {A1 A2 : Set} (ls : list (A1 * A2)) : list A1 * list A2 :=
match ls with
| [] => ([], [])
| (x1, x2) :: ls' =>
let (ls1, ls2) := unzip ls' in
(x1 :: ls1, x2 :: ls2)
end.
(* A few common-sense properties hold of these definitions. *)
Theorem length_zip : forall A1 A2 (ls1 : list A1) (ls2 : list A2),
length (zip ls1 ls2) = min (length ls1) (length ls2).
Proof.
induct ls1; simplify.
linear_arithmetic.
cases ls2; simplify.
linear_arithmetic.
rewrite IHls1.
linear_arithmetic.
Qed.
(* We write [fst] and [snd] for the first and second projection operators on
* pairs, respectively. *)
Theorem length_unzip1 : forall (A1 A2 : Set) (ls : list (A1 * A2)),
length (fst (unzip ls)) = length ls.
Proof.
induct ls; simplify.
equality.
cases hd.
(* Note that [cases] allows us to pull apart a pair into its two pieces. *)
cases (unzip ls).
simplify.
equality.
Qed.
Theorem length_unzip2 : forall (A1 A2 : Set) (ls : list (A1 * A2)),
length (snd (unzip ls)) = length ls.
Proof.
induct ls; simplify.
equality.
cases hd.
cases (unzip ls).
simplify.
equality.
Qed.
Theorem zip_unzip : forall (A1 A2 : Set) (ls : list (A1 * A2)),
(let (ls1, ls2) := unzip ls in zip ls1 ls2) = ls.
Proof.
induct ls; simplify.
equality.
cases hd.
cases (unzip ls).
simplify.
equality.
Qed.
(* There are also interesting interactions with [app] and [rev]. *)
Theorem unzip_app : forall (A1 A2 : Set) (x y : list (A1 * A2)),
unzip (x ++ y)
= (let (x1, x2) := unzip x in
let (y1, y2) := unzip y in
(x1 ++ y1, x2 ++ y2)).
Proof.
induct x; simplify.
cases (unzip y).
equality.
cases hd.
cases (unzip x).
simplify.
rewrite IHx.
cases (unzip y).
equality.
Qed.
Theorem unzip_rev : forall (A1 A2 : Set) (ls : list (A1 * A2)),
unzip (rev ls) = (let (ls1, ls2) := unzip ls in
(rev ls1, rev ls2)).
Proof.
induct ls; simplify.
equality.
cases hd.
cases (unzip ls).
simplify.
rewrite unzip_app.
rewrite IHls.
simplify.
equality.
Qed.
(** * Binary trees *)
(* Another classic datatype is binary trees, which we can define like so. *)
Inductive tree (A : Set) : Set :=
| Leaf
| Node (l : tree A) (d : A) (r : tree A).
Arguments Leaf [A].
Example tr1 : tree nat := Node (Node Leaf 7 Leaf) 8 (Node Leaf 9 (Node Leaf 10 Leaf)).
(* There is a natural notion of size of a tree. *)
Fixpoint size {A} (t : tree A) : nat :=
match t with
| Leaf => 0
| Node l _ r => 1 + size l + size r
end.
(* There is also a natural sense of reversing a tree, flipping it around its
* vertical axis. *)
Fixpoint reverse {A} (t : tree A) : tree A :=
match t with
| Leaf => Leaf
| Node l d r => Node (reverse r) d (reverse l)
end.
(* There is a natural relationship between the two. *)
Theorem size_reverse : forall A (t : tree A),
size (reverse t) = size t.
Proof.
induct t; simplify.
equality.
linear_arithmetic.
Qed.
(* Another classic tree operation is flattening into lists. *)
Fixpoint flatten {A} (t : tree A) : list A :=
match t with
| Leaf => []
| Node l d r => flatten l ++ d :: flatten r
end.
(* Note here that operators [++] and [::] are right-associative. *)
Theorem length_flatten : forall A (t : tree A),
length (flatten t) = size t.
Proof.
induct t; simplify.
equality.
rewrite length_app.
simplify.
linear_arithmetic.
Qed.
Lemma rev_app : forall A (ls1 ls2 : list A),
rev (ls1 ++ ls2) = rev ls2 ++ rev ls1.
Proof.
induct ls1; simplify.
rewrite app_nil.
equality.
rewrite IHls1.
apply app_assoc.
Qed.
Theorem rev_flatten : forall A (t : tree A),
rev (flatten t) = flatten (reverse t).
Proof.
induct t; simplify.
equality.
rewrite rev_app.
simplify.
rewrite app_assoc.
simplify.
equality.
Qed.