annotate src/Subset.v @ 75:ec2c1830a7a1

Type-checking example, with discussion
author Adam Chlipala <adamc@hcoop.net>
date Fri, 03 Oct 2008 15:10:30 -0400
parents a21447f76aad
children 82a2189fa283
rev   line source
adamc@70 1 (* Copyright (c) 2008, Adam Chlipala
adamc@70 2 *
adamc@70 3 * This work is licensed under a
adamc@70 4 * Creative Commons Attribution-Noncommercial-No Derivative Works 3.0
adamc@70 5 * Unported License.
adamc@70 6 * The license text is available at:
adamc@70 7 * http://creativecommons.org/licenses/by-nc-nd/3.0/
adamc@70 8 *)
adamc@70 9
adamc@70 10 (* begin hide *)
adamc@70 11 Require Import List.
adamc@70 12
adamc@70 13 Require Import Tactics.
adamc@70 14
adamc@70 15 Set Implicit Arguments.
adamc@70 16 (* end hide *)
adamc@70 17
adamc@70 18
adamc@74 19 (** %\part{Programming with Dependent Types}
adamc@74 20
adamc@74 21 \chapter{Subset Types and Variations}% *)
adamc@70 22
adamc@70 23 (** So far, we have seen many examples of what we might call "classical program verification." We write programs, write their specifications, and then prove that the programs satisfy their specifications. The programs that we have written in Coq have been normal functional programs that we could just as well have written in Haskell or ML. In this chapter, we start investigating uses of %\textit{%#<i>#dependent types#</i>#%}% to integrate programming, specification, and proving into a single phase. *)
adamc@70 24
adamc@70 25
adamc@70 26 (** * Introducing Subset Types *)
adamc@70 27
adamc@70 28 (** Let us consider several ways of implementing the natural number predecessor function. We start by displaying the definition from the standard library: *)
adamc@70 29
adamc@70 30 Print pred.
adamc@70 31 (** [[
adamc@70 32
adamc@70 33 pred = fun n : nat => match n with
adamc@70 34 | 0 => 0
adamc@70 35 | S u => u
adamc@70 36 end
adamc@70 37 : nat -> nat
adamc@70 38 ]] *)
adamc@70 39
adamc@70 40 (** We can use a new command, [Extraction], to produce an OCaml version of this function. *)
adamc@70 41
adamc@70 42 Extraction pred.
adamc@70 43
adamc@70 44 (** %\begin{verbatim}
adamc@70 45 (** val pred : nat -> nat **)
adamc@70 46
adamc@70 47 let pred = function
adamc@70 48 | O -> O
adamc@70 49 | S u -> u
adamc@70 50 \end{verbatim}%
adamc@70 51
adamc@70 52 #<pre>
adamc@70 53 (** val pred : nat -> nat **)
adamc@70 54
adamc@70 55 let pred = function
adamc@70 56 | O -> O
adamc@70 57 | S u -> u
adamc@70 58 </pre># *)
adamc@70 59
adamc@70 60 (** Returning 0 as the predecessor of 0 can come across as somewhat of a hack. In some situations, we might like to be sure that we never try to take the predecessor of 0. We can enforce this by giving [pred] a stronger, dependent type. *)
adamc@70 61
adamc@70 62 Lemma zgtz : 0 > 0 -> False.
adamc@70 63 crush.
adamc@70 64 Qed.
adamc@70 65
adamc@70 66 Definition pred_strong1 (n : nat) : n > 0 -> nat :=
adamc@70 67 match n return (n > 0 -> nat) with
adamc@70 68 | O => fun pf : 0 > 0 => match zgtz pf with end
adamc@70 69 | S n' => fun _ => n'
adamc@70 70 end.
adamc@70 71
adamc@70 72 (** We expand the type of [pred] to include a %\textit{%#<i>#proof#</i>#%}% that its argument [n] is greater than 0. When [n] is 0, we use the proof to derive a contradiction, which we can use to build a value of any type via a vacuous pattern match. When [n] is a successor, we have no need for the proof and just return the answer. The proof argument can be said to have a %\textit{%#<i>#dependent#</i>#%}% type, because its type depends on the %\textit{%#<i>#value#</i>#%}% of the argument [n].
adamc@70 73
adamc@70 74 There are two aspects of the definition of [pred_strong1] that may be surprising. First, we took advantage of [Definition]'s syntactic sugar for defining function arguments in the case of [n], but we bound the proofs later with explicit [fun] expressions. Second, there is the [return] clause for the [match], which we saw briefly in Chapter 2. Let us see what happens if we write this function in the way that at first seems most natural. *)
adamc@70 75
adamc@70 76 (** [[
adamc@70 77 Definition pred_strong1' (n : nat) (pf : n > 0) : nat :=
adamc@70 78 match n with
adamc@70 79 | O => match zgtz pf with end
adamc@70 80 | S n' => n'
adamc@70 81 end.
adamc@70 82
adamc@70 83 [[
adamc@70 84 Error: In environment
adamc@70 85 n : nat
adamc@70 86 pf : n > 0
adamc@70 87 The term "pf" has type "n > 0" while it is expected to have type
adamc@70 88 "0 > 0"
adamc@70 89 ]]
adamc@70 90
adamc@70 91 The term [zgtz pf] fails to type-check. Somehow the type checker has failed to take into account information that follows from which [match] branch that term appears in. The problem is that, by default, [match] does not let us use such implied information. To get refined typing, we must always add special [match] annotations.
adamc@70 92
adamc@70 93 In this case, we must use a [return] annotation to declare the relationship between the %\textit{%#<i>#value#</i>#%}% of the [match] discriminee and the %\textit{%#<i>#type#</i>#%}% of the result. There is no annotation that lets us declare a relationship between the discriminee and the type of a variable that is already in scope; hence, we delay the binding of [pf], so that we can use the [return] annotation to express the needed relationship.
adamc@70 94
adamc@70 95 Why does Coq not infer this relationship for us? Certainly, it is not hard to imagine heuristics that would handle this particular case and many others. In general, however, the inference problem is undecidable. The known undecidable problem of %\textit{%#<i>#higher-order unification#</i>#%}% reduces to the [match] type inference problem. Over time, Coq is enhanced with more and more heuristics to get around this problem, but there must always exist [match]es whose types Coq cannot infer without annotations.
adamc@70 96
adamc@70 97 Let us now take a look at the OCaml code Coq generates for [pred_strong1]. *)
adamc@70 98
adamc@70 99 Extraction pred_strong1.
adamc@70 100
adamc@70 101 (** %\begin{verbatim}
adamc@70 102 (** val pred_strong1 : nat -> nat **)
adamc@70 103
adamc@70 104 let pred_strong1 = function
adamc@70 105 | O -> assert false (* absurd case *)
adamc@70 106 | S n' -> n'
adamc@70 107 \end{verbatim}%
adamc@70 108
adamc@70 109 #<pre>
adamc@70 110 (** val pred_strong1 : nat -> nat **)
adamc@70 111
adamc@70 112 let pred_strong1 = function
adamc@70 113 | O -> assert false (* absurd case *)
adamc@70 114 | S n' -> n'
adamc@70 115 </pre># *)
adamc@70 116
adamc@70 117 (** The proof argument has disappeared! We get exactly the OCaml code we would have written manually. This is our first demonstration of the main technically interesting feature of Coq program extraction: program components of type [Prop] are erased systematically.
adamc@70 118
adamc@70 119 We can reimplement our dependently-typed [pred] based on %\textit{%#<i>#subset types#</i>#%}%, defined in the standard library with the type family [sig]. *)
adamc@70 120
adamc@70 121 Print sig.
adamc@70 122 (** [[
adamc@70 123
adamc@70 124 Inductive sig (A : Type) (P : A -> Prop) : Type :=
adamc@70 125 exist : forall x : A, P x -> sig P
adamc@70 126 For sig: Argument A is implicit
adamc@70 127 For exist: Argument A is implicit
adamc@70 128 ]]
adamc@70 129
adamc@70 130 [sig] is a Curry-Howard twin of [ex], except that [sig] is in [Type], while [ex] is in [Prop]. That means that [sig] values can survive extraction, while [ex] proofs will always be erased. The actual details of extraction of [sig]s are more subtle, as we will see shortly.
adamc@70 131
adamc@70 132 We rewrite [pred_strong1], using some syntactic sugar for subset types. *)
adamc@70 133
adamc@70 134 Locate "{ _ : _ | _ }".
adamc@70 135 (** [[
adamc@70 136
adamc@70 137 Notation Scope
adamc@70 138 "{ x : A | P }" := sig (fun x : A => P)
adamc@70 139 : type_scope
adamc@70 140 (default interpretation)
adamc@70 141 ]] *)
adamc@70 142
adamc@70 143 Definition pred_strong2 (s : {n : nat | n > 0}) : nat :=
adamc@70 144 match s with
adamc@70 145 | exist O pf => match zgtz pf with end
adamc@70 146 | exist (S n') _ => n'
adamc@70 147 end.
adamc@70 148
adamc@70 149 Extraction pred_strong2.
adamc@70 150
adamc@70 151 (** %\begin{verbatim}
adamc@70 152 (** val pred_strong2 : nat -> nat **)
adamc@70 153
adamc@70 154 let pred_strong2 = function
adamc@70 155 | O -> assert false (* absurd case *)
adamc@70 156 | S n' -> n'
adamc@70 157 \end{verbatim}%
adamc@70 158
adamc@70 159 #<pre>
adamc@70 160 (** val pred_strong2 : nat -> nat **)
adamc@70 161
adamc@70 162 let pred_strong2 = function
adamc@70 163 | O -> assert false (* absurd case *)
adamc@70 164 | S n' -> n'
adamc@70 165 </pre>#
adamc@70 166
adamc@70 167 We arrive at the same OCaml code as was extracted from [pred_strong1], which may seem surprising at first. The reason is that a value of [sig] is a pair of two pieces, a value and a proof about it. Extraction erases the proof, which reduces the constructor [exist] of [sig] to taking just a single argument. An optimization eliminates uses of datatypes with single constructors taking single arguments, and we arrive back where we started.
adamc@70 168
adamc@70 169 We can continue on in the process of refining [pred]'s type. Let us change its result type to capture that the output is really the predecessor of the input. *)
adamc@70 170
adamc@70 171 Definition pred_strong3 (s : {n : nat | n > 0}) : {m : nat | proj1_sig s = S m} :=
adamc@70 172 match s return {m : nat | proj1_sig s = S m} with
adamc@70 173 | exist 0 pf => match zgtz pf with end
adamc@70 174 | exist (S n') _ => exist _ n' (refl_equal _)
adamc@70 175 end.
adamc@70 176
adamc@70 177 (** The function [proj1_sig] extracts the base value from a subset type. Besides the use of that function, the only other new thing is the use of the [exist] constructor to build a new [sig] value, and the details of how to do that follow from the output of our earlier [Print] command.
adamc@70 178
adamc@70 179 By now, the reader is probably ready to believe that the new [pred_strong] leads to the same OCaml code as we have seen several times so far, and Coq does not disappoint. *)
adamc@70 180
adamc@70 181 Extraction pred_strong3.
adamc@70 182
adamc@70 183 (** %\begin{verbatim}
adamc@70 184 (** val pred_strong3 : nat -> nat **)
adamc@70 185
adamc@70 186 let pred_strong3 = function
adamc@70 187 | O -> assert false (* absurd case *)
adamc@70 188 | S n' -> n'
adamc@70 189 \end{verbatim}%
adamc@70 190
adamc@70 191 #<pre>
adamc@70 192 (** val pred_strong3 : nat -> nat **)
adamc@70 193
adamc@70 194 let pred_strong3 = function
adamc@70 195 | O -> assert false (* absurd case *)
adamc@70 196 | S n' -> n'
adamc@70 197 </pre>#
adamc@70 198
adamc@70 199 We have managed to reach a type that is, in a formal sense, the most expressive possible for [pred]. Any other implementation of the same type must have the same input-output behavior. However, there is still room for improvement in making this kind of code easier to write. Here is a version that takes advantage of tactic-based theorem proving. We switch back to passing a separate proof argument instead of using a subset type for the function's input, because this leads to cleaner code. *)
adamc@70 200
adamc@70 201 Definition pred_strong4 (n : nat) : n > 0 -> {m : nat | n = S m}.
adamc@70 202 refine (fun n =>
adamc@70 203 match n return (n > 0 -> {m : nat | n = S m}) with
adamc@70 204 | O => fun _ => False_rec _ _
adamc@70 205 | S n' => fun _ => exist _ n' _
adamc@70 206 end).
adamc@70 207
adamc@70 208 (** We build [pred_strong4] using tactic-based proving, beginning with a [Definition] command that ends in a period before a definition is given. Such a command enters the interactive proving mode, with the type given for the new identifier as our proof goal. We do most of the work with the [refine] tactic, to which we pass a partial "proof" of the type we are trying to prove. There may be some pieces left to fill in, indicated by underscores. Any underscore that Coq cannot reconstruct with type inference is added as a proof subgoal. In this case, we have two subgoals:
adamc@70 209
adamc@70 210 [[
adamc@70 211
adamc@70 212 2 subgoals
adamc@70 213
adamc@70 214 n : nat
adamc@70 215 _ : 0 > 0
adamc@70 216 ============================
adamc@70 217 False
adamc@70 218 ]]
adamc@70 219
adamc@70 220 [[
adamc@70 221
adamc@70 222 subgoal 2 is:
adamc@70 223 S n' = S n'
adamc@70 224 ]]
adamc@70 225
adamc@70 226 We can see that the first subgoal comes from the second underscore passed to [False_rec], and the second subgoal comes from the second underscore passed to [exist]. In the first case, we see that, though we bound the proof variable with an underscore, it is still available in our proof context. It is hard to refer to underscore-named variables in manual proofs, but automation makes short work of them. Both subgoals are easy to discharge that way, so let us back up and ask to prove all subgoals automatically. *)
adamc@70 227
adamc@70 228 Undo.
adamc@70 229 refine (fun n =>
adamc@70 230 match n return (n > 0 -> {m : nat | n = S m}) with
adamc@70 231 | O => fun _ => False_rec _ _
adamc@70 232 | S n' => fun _ => exist _ n' _
adamc@70 233 end); crush.
adamc@70 234 Defined.
adamc@70 235
adamc@70 236 (** We end the "proof" with [Defined] instead of [Qed], so that the definition we constructed remains visible. This contrasts to the case of ending a proof with [Qed], where the details of the proof are hidden afterward. Let us see what our prooof script constructed. *)
adamc@70 237
adamc@70 238 Print pred_strong4.
adamc@70 239 (** [[
adamc@70 240
adamc@70 241 pred_strong4 =
adamc@70 242 fun n : nat =>
adamc@70 243 match n as n0 return (n0 > 0 -> {m : nat | n0 = S m}) with
adamc@70 244 | 0 =>
adamc@70 245 fun _ : 0 > 0 =>
adamc@70 246 False_rec {m : nat | 0 = S m}
adamc@70 247 (Bool.diff_false_true
adamc@70 248 (Bool.absurd_eq_true false
adamc@70 249 (Bool.diff_false_true
adamc@70 250 (Bool.absurd_eq_true false (pred_strong4_subproof n _)))))
adamc@70 251 | S n' =>
adamc@70 252 fun _ : S n' > 0 =>
adamc@70 253 exist (fun m : nat => S n' = S m) n' (refl_equal (S n'))
adamc@70 254 end
adamc@70 255 : forall n : nat, n > 0 -> {m : nat | n = S m}
adamc@70 256 ]]
adamc@70 257
adamc@70 258 We see the code we entered, with some proofs filled in. The first proof obligation, the second argument to [False_rec], is filled in with a nasty-looking proof term that we can be glad we did not enter by hand. The second proof obligation is a simple reflexivity proof.
adamc@70 259
adamc@70 260 We are almost done with the ideal implementation of dependent predecessor. We can use Coq's syntax extension facility to arrive at code with almost no complexity beyond a Haskell or ML program with a complete specification in a comment. *)
adamc@70 261
adamc@70 262 Notation "!" := (False_rec _ _).
adamc@70 263 Notation "[ e ]" := (exist _ e _).
adamc@70 264
adamc@70 265 Definition pred_strong5 (n : nat) : n > 0 -> {m : nat | n = S m}.
adamc@70 266 refine (fun n =>
adamc@70 267 match n return (n > 0 -> {m : nat | n = S m}) with
adamc@70 268 | O => fun _ => !
adamc@70 269 | S n' => fun _ => [n']
adamc@70 270 end); crush.
adamc@70 271 Defined.
adamc@71 272
adamc@71 273
adamc@71 274 (** * Decidable Proposition Types *)
adamc@71 275
adamc@71 276 (** There is another type in the standard library which captures the idea of program values that indicate which of two propositions is true. *)
adamc@71 277
adamc@71 278 Print sumbool.
adamc@71 279 (** [[
adamc@71 280
adamc@71 281 Inductive sumbool (A : Prop) (B : Prop) : Set :=
adamc@71 282 left : A -> {A} + {B} | right : B -> {A} + {B}
adamc@71 283 For left: Argument A is implicit
adamc@71 284 For right: Argument B is implicit
adamc@71 285 ]] *)
adamc@71 286
adamc@71 287 (** We can define some notations to make working with [sumbool] more convenient. *)
adamc@71 288
adamc@71 289 Notation "'Yes'" := (left _ _).
adamc@71 290 Notation "'No'" := (right _ _).
adamc@71 291 Notation "'Reduce' x" := (if x then Yes else No) (at level 50).
adamc@71 292
adamc@71 293 (** The [Reduce] notation is notable because it demonstrates how [if] is overloaded in Coq. The [if] form actually works when the test expression has any two-constructor inductive type. Moreover, in the [then] and [else] branches, the appropriate constructor arguments are bound. This is important when working with [sumbool]s, when we want to have the proof stored in the test expression available when proving the proof obligations generated in the appropriate branch.
adamc@71 294
adamc@71 295 Now we can write [eq_nat_dec], which compares two natural numbers, returning either a proof of their equality or a proof of their inequality. *)
adamc@71 296
adamc@71 297 Definition eq_nat_dec (n m : nat) : {n = m} + {n <> m}.
adamc@71 298 refine (fix f (n m : nat) {struct n} : {n = m} + {n <> m} :=
adamc@71 299 match n, m return {n = m} + {n <> m} with
adamc@71 300 | O, O => Yes
adamc@71 301 | S n', S m' => Reduce (f n' m')
adamc@71 302 | _, _ => No
adamc@71 303 end); congruence.
adamc@71 304 Defined.
adamc@71 305
adamc@71 306 (** Our definition extracts to reasonable OCaml code. *)
adamc@71 307
adamc@71 308 Extraction eq_nat_dec.
adamc@71 309
adamc@71 310 (** %\begin{verbatim}
adamc@71 311 (** val eq_nat_dec : nat -> nat -> sumbool **)
adamc@71 312
adamc@71 313 let rec eq_nat_dec n m =
adamc@71 314 match n with
adamc@71 315 | O -> (match m with
adamc@71 316 | O -> Left
adamc@71 317 | S n0 -> Right)
adamc@71 318 | S n' -> (match m with
adamc@71 319 | O -> Right
adamc@71 320 | S m' -> eq_nat_dec n' m')
adamc@71 321 \end{verbatim}%
adamc@71 322
adamc@71 323 #<pre>
adamc@71 324 (** val eq_nat_dec : nat -> nat -> sumbool **)
adamc@71 325
adamc@71 326 let rec eq_nat_dec n m =
adamc@71 327 match n with
adamc@71 328 | O -> (match m with
adamc@71 329 | O -> Left
adamc@71 330 | S n0 -> Right)
adamc@71 331 | S n' -> (match m with
adamc@71 332 | O -> Right
adamc@71 333 | S m' -> eq_nat_dec n' m')
adamc@71 334 </pre>#
adamc@71 335
adamc@71 336 Proving this kind of decidable equality result is so common that Coq comes with a tactic for automating it. *)
adamc@71 337
adamc@71 338 Definition eq_nat_dec' (n m : nat) : {n = m} + {n <> m}.
adamc@71 339 decide equality.
adamc@71 340 Defined.
adamc@71 341
adamc@71 342 (** Curious readers can verify that the [decide equality] version extracts to the same OCaml code as our more manual version does. That OCaml code had one undesirable property, which is that it uses %\texttt{%#<tt>#Left#</tt>#%}% and %\texttt{%#<tt>#Right#</tt>#%}% constructors instead of the boolean values built into OCaml. We can fix this, by using Coq's facility for mapping Coq inductive types to OCaml variant types. *)
adamc@71 343
adamc@71 344 Extract Inductive sumbool => "bool" ["true" "false"].
adamc@71 345 Extraction eq_nat_dec'.
adamc@71 346
adamc@71 347 (** %\begin{verbatim}
adamc@71 348 (** val eq_nat_dec' : nat -> nat -> bool **)
adamc@71 349
adamc@71 350 let rec eq_nat_dec' n m0 =
adamc@71 351 match n with
adamc@71 352 | O -> (match m0 with
adamc@71 353 | O -> true
adamc@71 354 | S n0 -> false)
adamc@71 355 | S n0 -> (match m0 with
adamc@71 356 | O -> false
adamc@71 357 | S n1 -> eq_nat_dec' n0 n1)
adamc@71 358 \end{verbatim}%
adamc@71 359
adamc@71 360 #<pre>
adamc@71 361 (** val eq_nat_dec' : nat -> nat -> bool **)
adamc@71 362
adamc@71 363 let rec eq_nat_dec' n m0 =
adamc@71 364 match n with
adamc@71 365 | O -> (match m0 with
adamc@71 366 | O -> true
adamc@71 367 | S n0 -> false)
adamc@71 368 | S n0 -> (match m0 with
adamc@71 369 | O -> false
adamc@71 370 | S n1 -> eq_nat_dec' n0 n1)
adamc@71 371 </pre># *)
adamc@72 372
adamc@72 373 (** %\smallskip%
adamc@72 374
adamc@72 375 We can build "smart" versions of the usual boolean operators and put them to good use in certified programming. For instance, here is a [sumbool] version of boolean "or." *)
adamc@72 376
adamc@72 377 Notation "x || y" := (if x then Yes else Reduce y) (at level 50).
adamc@72 378
adamc@72 379 (** Let us use it for building a function that decides list membership. We need to assume the existence of an equality decision procedure for the type of list elements. *)
adamc@72 380
adamc@72 381 Section In_dec.
adamc@72 382 Variable A : Set.
adamc@72 383 Variable A_eq_dec : forall x y : A, {x = y} + {x <> y}.
adamc@72 384
adamc@72 385 (** The final function is easy to write using the techniques we have developed so far. *)
adamc@72 386
adamc@72 387 Definition In_dec : forall (x : A) (ls : list A), {In x ls} + { ~In x ls}.
adamc@72 388 refine (fix f (x : A) (ls : list A) {struct ls}
adamc@72 389 : {In x ls} + { ~In x ls} :=
adamc@72 390 match ls return {In x ls} + { ~In x ls} with
adamc@72 391 | nil => No
adamc@72 392 | x' :: ls' => A_eq_dec x x' || f x ls'
adamc@72 393 end); crush.
adamc@72 394 Qed.
adamc@72 395 End In_dec.
adamc@72 396
adamc@72 397 (** [In_dec] has a reasonable extraction to OCaml. *)
adamc@72 398
adamc@72 399 Extraction In_dec.
adamc@72 400
adamc@72 401 (** %\begin{verbatim}
adamc@72 402 (** val in_dec : ('a1 -> 'a1 -> bool) -> 'a1 -> 'a1 list -> bool **)
adamc@72 403
adamc@72 404 let rec in_dec a_eq_dec x = function
adamc@72 405 | Nil -> false
adamc@72 406 | Cons (x', ls') ->
adamc@72 407 (match a_eq_dec x x' with
adamc@72 408 | true -> true
adamc@72 409 | false -> in_dec a_eq_dec x ls')
adamc@72 410 \end{verbatim}%
adamc@72 411
adamc@72 412 #<pre>
adamc@72 413 (** val in_dec : ('a1 -> 'a1 -> bool) -> 'a1 -> 'a1 list -> bool **)
adamc@72 414
adamc@72 415 let rec in_dec a_eq_dec x = function
adamc@72 416 | Nil -> false
adamc@72 417 | Cons (x', ls') ->
adamc@72 418 (match a_eq_dec x x' with
adamc@72 419 | true -> true
adamc@72 420 | false -> in_dec a_eq_dec x ls')
adamc@72 421 </pre># *)
adamc@72 422
adamc@72 423
adamc@72 424 (** * Partial Subset Types *)
adamc@72 425
adamc@73 426 (** Our final implementation of dependent predecessor used a very specific argument type to ensure that execution could always complete normally. Sometimes we want to allow execution to fail, and we want a more principled way of signaling that than returning a default value, as [pred] does for [0]. One approach is to define this type family [maybe], which is a version of [sig] that allows obligation-free failure. *)
adamc@73 427
adamc@72 428 Inductive maybe (A : Type) (P : A -> Prop) : Set :=
adamc@72 429 | Unknown : maybe P
adamc@72 430 | Found : forall x : A, P x -> maybe P.
adamc@72 431
adamc@73 432 (** We can define some new notations, analogous to those we defined for subset types. *)
adamc@73 433
adamc@72 434 Notation "{{ x | P }}" := (maybe (fun x => P)).
adamc@72 435 Notation "??" := (Unknown _).
adamc@72 436 Notation "[[ x ]]" := (Found _ x _).
adamc@72 437
adamc@73 438 (** Now our next version of [pred] is trivial to write. *)
adamc@73 439
adamc@73 440 Definition pred_strong6 (n : nat) : {{m | n = S m}}.
adamc@73 441 refine (fun n =>
adamc@73 442 match n return {{m | n = S m}} with
adamc@73 443 | O => ??
adamc@73 444 | S n' => [[n']]
adamc@73 445 end); trivial.
adamc@73 446 Defined.
adamc@73 447
adamc@73 448 (** Because we used [maybe], one valid implementation of the type we gave [pred_strong6] would return [??] in every case. We can strengthen the type to rule out such vacuous implementations, and the type family [sumor] from the standard library provides the easiest starting point. For type [A] and proposition [B], [A + {B}] desugars to [sumor A B], whose values are either values of [A] or proofs of [B]. *)
adamc@73 449
adamc@73 450 Print sumor.
adamc@73 451 (** [[
adamc@73 452
adamc@73 453 Inductive sumor (A : Type) (B : Prop) : Type :=
adamc@73 454 inleft : A -> A + {B} | inright : B -> A + {B}
adamc@73 455 For inleft: Argument A is implicit
adamc@73 456 For inright: Argument B is implicit
adamc@73 457 ]] *)
adamc@73 458
adamc@73 459 (** We add notations for easy use of the [sumor] constructors. The second notation is specialized to [sumor]s whose [A] parameters are instantiated with regular subset types, since this is how we will use [sumor] below. *)
adamc@73 460
adamc@73 461 Notation "!!" := (inright _ _).
adamc@73 462 Notation "[[[ x ]]]" := (inleft _ [x]).
adamc@73 463
adamc@73 464 (** Now we are ready to give the final version of possibly-failing predecessor. The [sumor]-based type that we use is maximally expressive; any implementation of the type has the same input-output behavior. *)
adamc@73 465
adamc@73 466 Definition pred_strong7 (n : nat) : {m : nat | n = S m} + {n = 0}.
adamc@73 467 refine (fun n =>
adamc@73 468 match n return {m : nat | n = S m} + {n = 0} with
adamc@73 469 | O => !!
adamc@73 470 | S n' => [[[n']]]
adamc@73 471 end); trivial.
adamc@73 472 Defined.
adamc@73 473
adamc@73 474
adamc@73 475 (** * Monadic Notations *)
adamc@73 476
adamc@73 477 (** We can treat [maybe] like a monad, in the same way that the Haskell [Maybe] type is interpreted as a failure monad. Our [maybe] has the wrong type to be a literal monad, but a "bind"-like notation will still be helpful. *)
adamc@73 478
adamc@72 479 Notation "x <- e1 ; e2" := (match e1 with
adamc@72 480 | Unknown => ??
adamc@72 481 | Found x _ => e2
adamc@72 482 end)
adamc@72 483 (right associativity, at level 60).
adamc@72 484
adamc@73 485 (** The meaning of [x <- e1; e2] is: First run [e1]. If it fails to find an answer, then announce failure for our derived computation, too. If [e1] %\textit{%#<i>#does#</i>#%}% find an answer, pass that answer on to [e2] to find the final result. The variable [x] can be considered bound in [e2].
adamc@73 486
adamc@73 487 This notation is very helpful for composing richly-typed procedures. For instance, here is a very simple implementation of a function to take the predecessors of two naturals at once. *)
adamc@73 488
adamc@73 489 Definition doublePred (n1 n2 : nat) : {{p | n1 = S (fst p) /\ n2 = S (snd p)}}.
adamc@73 490 refine (fun n1 n2 =>
adamc@73 491 m1 <- pred_strong6 n1;
adamc@73 492 m2 <- pred_strong6 n2;
adamc@73 493 [[(m1, m2)]]); tauto.
adamc@73 494 Defined.
adamc@73 495
adamc@73 496 (** We can build a [sumor] version of the "bind" notation and use it to write a similarly straightforward version of this function. *)
adamc@73 497
adamc@73 498 (** printing <-- $\longleftarrow$ *)
adamc@73 499
adamc@73 500 Notation "x <-- e1 ; e2" := (match e1 with
adamc@73 501 | inright _ => !!
adamc@73 502 | inleft (exist x _) => e2
adamc@73 503 end)
adamc@73 504 (right associativity, at level 60).
adamc@73 505
adamc@73 506 (** printing * $\times$ *)
adamc@73 507
adamc@73 508 Definition doublePred' (n1 n2 : nat) : {p : nat * nat | n1 = S (fst p) /\ n2 = S (snd p)}
adamc@73 509 + {n1 = 0 \/ n2 = 0}.
adamc@73 510 refine (fun n1 n2 =>
adamc@73 511 m1 <-- pred_strong7 n1;
adamc@73 512 m2 <-- pred_strong7 n2;
adamc@73 513 [[[(m1, m2)]]]); tauto.
adamc@73 514 Defined.
adamc@72 515
adamc@72 516
adamc@72 517 (** * A Type-Checking Example *)
adamc@72 518
adamc@75 519 (** We can apply these specification types to build a certified type-checker for a simple expression language. *)
adamc@75 520
adamc@72 521 Inductive exp : Set :=
adamc@72 522 | Nat : nat -> exp
adamc@72 523 | Plus : exp -> exp -> exp
adamc@72 524 | Bool : bool -> exp
adamc@72 525 | And : exp -> exp -> exp.
adamc@72 526
adamc@75 527 (** We define a simple language of types and its typing rules, in the style introduced in Chapter 4. *)
adamc@75 528
adamc@72 529 Inductive type : Set := TNat | TBool.
adamc@72 530
adamc@72 531 Inductive hasType : exp -> type -> Prop :=
adamc@72 532 | HtNat : forall n,
adamc@72 533 hasType (Nat n) TNat
adamc@72 534 | HtPlus : forall e1 e2,
adamc@72 535 hasType e1 TNat
adamc@72 536 -> hasType e2 TNat
adamc@72 537 -> hasType (Plus e1 e2) TNat
adamc@72 538 | HtBool : forall b,
adamc@72 539 hasType (Bool b) TBool
adamc@72 540 | HtAnd : forall e1 e2,
adamc@72 541 hasType e1 TBool
adamc@72 542 -> hasType e2 TBool
adamc@72 543 -> hasType (And e1 e2) TBool.
adamc@72 544
adamc@75 545 (** It will be helpful to have a function for comparing two types. We build one using [decide equality]. *)
adamc@75 546
adamc@75 547 Definition eq_type_dec : forall t1 t2 : type, {t1 = t2} + {t1 <> t2}.
adamc@72 548 decide equality.
adamc@72 549 Defined.
adamc@72 550
adamc@75 551 (** Another notation complements the monadic notation for [maybe] that we defined earlier. Sometimes we want to be to include "assertions" in our procedures. That is, we want to run a decision procedure and fail if it fails; otherwise, we want to continue, with the proof that it produced made available to us. This infix notation captures that, for a procedure that returns an arbitrary two-constructor type. *)
adamc@75 552
adamc@73 553 Notation "e1 ;; e2" := (if e1 then e2 else ??)
adamc@73 554 (right associativity, at level 60).
adamc@73 555
adamc@75 556 (** With that notation defined, we can implement a [typeCheck] function, whose code is only more complex than what we would write in ML because it needs to include some extra type annotations. Every [[[e]]] expression adds a [hasType] proof obligation, and [crush] makes short work of them when we add [hasType]'s constructors as hints. *)
adamc@75 557
adamc@72 558 Definition typeCheck (e : exp) : {{t | hasType e t}}.
adamc@72 559 Hint Constructors hasType.
adamc@72 560
adamc@72 561 refine (fix F (e : exp) : {{t | hasType e t}} :=
adamc@72 562 match e return {{t | hasType e t}} with
adamc@72 563 | Nat _ => [[TNat]]
adamc@72 564 | Plus e1 e2 =>
adamc@72 565 t1 <- F e1;
adamc@72 566 t2 <- F e2;
adamc@72 567 eq_type_dec t1 TNat;;
adamc@72 568 eq_type_dec t2 TNat;;
adamc@72 569 [[TNat]]
adamc@72 570 | Bool _ => [[TBool]]
adamc@72 571 | And e1 e2 =>
adamc@72 572 t1 <- F e1;
adamc@72 573 t2 <- F e2;
adamc@72 574 eq_type_dec t1 TBool;;
adamc@72 575 eq_type_dec t2 TBool;;
adamc@72 576 [[TBool]]
adamc@72 577 end); crush.
adamc@72 578 Defined.
adamc@72 579
adamc@75 580 (** Despite manipulating proofs, our type checker is easy to run. *)
adamc@75 581
adamc@72 582 Eval simpl in typeCheck (Nat 0).
adamc@75 583 (** [[
adamc@75 584
adamc@75 585 = [[TNat]]
adamc@75 586 : {{t | hasType (Nat 0) t}}
adamc@75 587 ]] *)
adamc@75 588
adamc@72 589 Eval simpl in typeCheck (Plus (Nat 1) (Nat 2)).
adamc@75 590 (** [[
adamc@75 591
adamc@75 592 = [[TNat]]
adamc@75 593 : {{t | hasType (Plus (Nat 1) (Nat 2)) t}}
adamc@75 594 ]] *)
adamc@75 595
adamc@72 596 Eval simpl in typeCheck (Plus (Nat 1) (Bool false)).
adamc@75 597 (** [[
adamc@75 598
adamc@75 599 = ??
adamc@75 600 : {{t | hasType (Plus (Nat 1) (Bool false)) t}}
adamc@75 601 ]] *)
adamc@75 602
adamc@75 603 (** The type-checker also extracts to some reasonable OCaml code. *)
adamc@75 604
adamc@75 605 Extraction typeCheck.
adamc@75 606
adamc@75 607 (** %\begin{verbatim}
adamc@75 608 (** val typeCheck : exp -> type0 maybe **)
adamc@75 609
adamc@75 610 let rec typeCheck = function
adamc@75 611 | Nat n -> Found TNat
adamc@75 612 | Plus (e1, e2) ->
adamc@75 613 (match typeCheck e1 with
adamc@75 614 | Unknown -> Unknown
adamc@75 615 | Found t1 ->
adamc@75 616 (match typeCheck e2 with
adamc@75 617 | Unknown -> Unknown
adamc@75 618 | Found t2 ->
adamc@75 619 (match eq_type_dec t1 TNat with
adamc@75 620 | true ->
adamc@75 621 (match eq_type_dec t2 TNat with
adamc@75 622 | true -> Found TNat
adamc@75 623 | false -> Unknown)
adamc@75 624 | false -> Unknown)))
adamc@75 625 | Bool b -> Found TBool
adamc@75 626 | And (e1, e2) ->
adamc@75 627 (match typeCheck e1 with
adamc@75 628 | Unknown -> Unknown
adamc@75 629 | Found t1 ->
adamc@75 630 (match typeCheck e2 with
adamc@75 631 | Unknown -> Unknown
adamc@75 632 | Found t2 ->
adamc@75 633 (match eq_type_dec t1 TBool with
adamc@75 634 | true ->
adamc@75 635 (match eq_type_dec t2 TBool with
adamc@75 636 | true -> Found TBool
adamc@75 637 | false -> Unknown)
adamc@75 638 | false -> Unknown)))
adamc@75 639 \end{verbatim}%
adamc@75 640
adamc@75 641 #<pre>
adamc@75 642 (** val typeCheck : exp -> type0 maybe **)
adamc@75 643
adamc@75 644 let rec typeCheck = function
adamc@75 645 | Nat n -> Found TNat
adamc@75 646 | Plus (e1, e2) ->
adamc@75 647 (match typeCheck e1 with
adamc@75 648 | Unknown -> Unknown
adamc@75 649 | Found t1 ->
adamc@75 650 (match typeCheck e2 with
adamc@75 651 | Unknown -> Unknown
adamc@75 652 | Found t2 ->
adamc@75 653 (match eq_type_dec t1 TNat with
adamc@75 654 | true ->
adamc@75 655 (match eq_type_dec t2 TNat with
adamc@75 656 | true -> Found TNat
adamc@75 657 | false -> Unknown)
adamc@75 658 | false -> Unknown)))
adamc@75 659 | Bool b -> Found TBool
adamc@75 660 | And (e1, e2) ->
adamc@75 661 (match typeCheck e1 with
adamc@75 662 | Unknown -> Unknown
adamc@75 663 | Found t1 ->
adamc@75 664 (match typeCheck e2 with
adamc@75 665 | Unknown -> Unknown
adamc@75 666 | Found t2 ->
adamc@75 667 (match eq_type_dec t1 TBool with
adamc@75 668 | true ->
adamc@75 669 (match eq_type_dec t2 TBool with
adamc@75 670 | true -> Found TBool
adamc@75 671 | false -> Unknown)
adamc@75 672 | false -> Unknown)))
adamc@75 673 </pre># *)
adamc@75 674
adamc@75 675 (** %\smallskip%
adamc@75 676
adamc@75 677 We can adapt this implementation to use [sumor], so that we know our type-checker only fails on ill-typed inputs. First, we define an analogue to the "assertion" notation. *)
adamc@73 678
adamc@73 679 Notation "e1 ;;; e2" := (if e1 then e2 else !!)
adamc@73 680 (right associativity, at level 60).
adamc@73 681
adamc@75 682 (** Next, we prove a helpful lemma, which states that a given expression can have at most one type. *)
adamc@75 683
adamc@75 684 Lemma hasType_det : forall e t1,
adamc@73 685 hasType e t1
adamc@73 686 -> forall t2,
adamc@73 687 hasType e t2
adamc@73 688 -> t1 = t2.
adamc@73 689 induction 1; inversion 1; crush.
adamc@73 690 Qed.
adamc@73 691
adamc@75 692 (** Now we can define the type-checker. Its type expresses that it only fails on untypable expressions. *)
adamc@75 693
adamc@73 694 Definition typeCheck' (e : exp) : {t : type | hasType e t} + {forall t, ~hasType e t}.
adamc@73 695 Hint Constructors hasType.
adamc@75 696 (** We register all of the typing rules as hints. *)
adamc@75 697
adamc@73 698 Hint Resolve hasType_det.
adamc@75 699 (** [hasType_det] will also be useful for proving proof obligations with contradictory contexts. Since its statement includes [forall]-bound variables that do not appear in its conclusion, only [eauto] will apply this hint. *)
adamc@73 700
adamc@75 701 (** Finally, the implementation of [typeCheck] can be transcribed literally, simply switching notations as needed. *)
adamc@73 702 refine (fix F (e : exp) : {t : type | hasType e t} + {forall t, ~hasType e t} :=
adamc@73 703 match e return {t : type | hasType e t} + {forall t, ~hasType e t} with
adamc@73 704 | Nat _ => [[[TNat]]]
adamc@73 705 | Plus e1 e2 =>
adamc@73 706 t1 <-- F e1;
adamc@73 707 t2 <-- F e2;
adamc@73 708 eq_type_dec t1 TNat;;;
adamc@73 709 eq_type_dec t2 TNat;;;
adamc@73 710 [[[TNat]]]
adamc@73 711 | Bool _ => [[[TBool]]]
adamc@73 712 | And e1 e2 =>
adamc@73 713 t1 <-- F e1;
adamc@73 714 t2 <-- F e2;
adamc@73 715 eq_type_dec t1 TBool;;;
adamc@73 716 eq_type_dec t2 TBool;;;
adamc@73 717 [[[TBool]]]
adamc@73 718 end); clear F; crush' tt hasType; eauto.
adamc@75 719
adamc@75 720 (** We clear [F], the local name for the recursive function, to avoid strange proofs that refer to recursive calls that we never make. The [crush] variant [crush'] helps us by performing automatic inversion on instances of the predicates specified in its second argument. Once we throw in [eauto] to apply [hasType_det] for us, we have discharged all the subgoals. *)
adamc@73 721 Defined.
adamc@73 722
adamc@75 723 (** The short implementation here hides just how time-saving automation is. Every use of one of the notations adds a proof obligation, giving us 12 in total. Most of these obligations require multiple inversions and either uses of [hasType_det] or applications of [hasType] rules.
adamc@75 724
adamc@75 725 The results of simplifying calls to [typeCheck'] look deceptively similar to the results for [typeCheck], but now the types of the results provide more information. *)
adamc@75 726
adamc@73 727 Eval simpl in typeCheck' (Nat 0).
adamc@75 728 (** [[
adamc@75 729
adamc@75 730 = [[[TNat]]]
adamc@75 731 : {t : type | hasType (Nat 0) t} +
adamc@75 732 {(forall t : type, ~ hasType (Nat 0) t)}
adamc@75 733 ]] *)
adamc@75 734
adamc@73 735 Eval simpl in typeCheck' (Plus (Nat 1) (Nat 2)).
adamc@75 736 (** [[
adamc@75 737
adamc@75 738 = [[[TNat]]]
adamc@75 739 : {t : type | hasType (Plus (Nat 1) (Nat 2)) t} +
adamc@75 740 {(forall t : type, ~ hasType (Plus (Nat 1) (Nat 2)) t)}
adamc@75 741 ]] *)
adamc@75 742
adamc@73 743 Eval simpl in typeCheck' (Plus (Nat 1) (Bool false)).
adamc@75 744 (** [[
adamc@75 745
adamc@75 746 = !!
adamc@75 747 : {t : type | hasType (Plus (Nat 1) (Bool false)) t} +
adamc@75 748 {(forall t : type, ~ hasType (Plus (Nat 1) (Bool false)) t)}
adamc@75 749 ]] *)