summaryrefslogtreecommitdiff
path: root/theories/Strings
diff options
context:
space:
mode:
Diffstat (limited to 'theories/Strings')
-rw-r--r--theories/Strings/Ascii.v143
-rw-r--r--theories/Strings/String.v52
-rw-r--r--theories/Strings/vo.itarget2
3 files changed, 107 insertions, 90 deletions
diff --git a/theories/Strings/Ascii.v b/theories/Strings/Ascii.v
index 1c02be7f..9e760d21 100644
--- a/theories/Strings/Ascii.v
+++ b/theories/Strings/Ascii.v
@@ -1,3 +1,4 @@
+(* -*- coding: utf-8 -*- *)
(************************************************************************)
(* v * The Coq Proof Assistant / The Coq Development Team *)
(* <O___,, * CNRS-Ecole Polytechnique-INRIA Futurs-Universite Paris Sud *)
@@ -6,39 +7,26 @@
(* * GNU Lesser General Public License Version 2.1 *)
(************************************************************************)
-(* $Id: Ascii.v 9245 2006-10-17 12:53:34Z notin $ *)
+(* $Id$ *)
-(** Contributed by Laurent Théry (INRIA);
+(** Contributed by Laurent Théry (INRIA);
Adapted to Coq V8 by the Coq Development Team *)
-Require Import Bool.
-Require Import BinPos.
+Require Import Bool BinPos BinNat Nnat.
+Declare ML Module "ascii_syntax_plugin".
(** * Definition of ascii characters *)
(** Definition of ascii character as a 8 bits constructor *)
-
+
Inductive ascii : Set := Ascii (_ _ _ _ _ _ _ _ : bool).
Delimit Scope char_scope with char.
Bind Scope char_scope with ascii.
-
+
Definition zero := Ascii false false false false false false false false.
-
+
Definition one := Ascii true false false false false false false false.
-
-Definition app1 (f : bool -> bool) (a : ascii) :=
- match a with
- | Ascii a1 a2 a3 a4 a5 a6 a7 a8 =>
- Ascii (f a1) (f a2) (f a3) (f a4) (f a5) (f a6) (f a7) (f a8)
- end.
-
-Definition app2 (f : bool -> bool -> bool) (a b : ascii) :=
- match a, b with
- | Ascii a1 a2 a3 a4 a5 a6 a7 a8, Ascii b1 b2 b3 b4 b5 b6 b7 b8 =>
- Ascii (f a1 b1) (f a2 b2) (f a3 b3) (f a4 b4)
- (f a5 b5) (f a6 b6) (f a7 b7) (f a8 b8)
- end.
Definition shift (c : bool) (a : ascii) :=
match a with
@@ -46,7 +34,7 @@ Definition shift (c : bool) (a : ascii) :=
end.
(** Definition of a decidable function that is effective *)
-
+
Definition ascii_dec : forall a b : ascii, {a = b} + {a <> b}.
decide equality; apply bool_dec.
Defined.
@@ -54,60 +42,85 @@ Defined.
(** * Conversion between natural numbers modulo 256 and ascii characters *)
(** Auxillary function that turns a positive into an ascii by
- looking at the last n bits, ie z mod 2^n *)
-
-Fixpoint ascii_of_pos_aux (res acc : ascii) (z : positive)
- (n : nat) {struct n} : ascii :=
+ looking at the last 8 bits, ie z mod 2^8 *)
+
+Definition ascii_of_pos : positive -> ascii :=
+ let loop := fix loop n p :=
+ match n with
+ | O => zero
+ | S n' =>
+ match p with
+ | xH => one
+ | xI p' => shift true (loop n' p')
+ | xO p' => shift false (loop n' p')
+ end
+ end
+ in loop 8.
+
+(** Conversion from [N] to [ascii] *)
+
+Definition ascii_of_N (n : N) :=
match n with
- | O => res
- | S n1 =>
- match z with
- | xH => app2 orb res acc
- | xI z' => ascii_of_pos_aux (app2 orb res acc) (shift false acc) z' n1
- | xO z' => ascii_of_pos_aux res (shift false acc) z' n1
- end
+ | N0 => zero
+ | Npos p => ascii_of_pos p
end.
+(** Same for [nat] *)
-(** Function that turns a positive into an ascii by
- looking at the last 8 bits, ie a mod 8 *)
-
-Definition ascii_of_pos (a : positive) := ascii_of_pos_aux zero one a 8.
+Definition ascii_of_nat (a : nat) := ascii_of_N (N_of_nat a).
-(** Function that turns a Peano number into an ascii by converting it
- to positive *)
+(** The opposite functions *)
-Definition ascii_of_nat (a : nat) :=
- match a with
- | O => zero
- | S a' => ascii_of_pos (P_of_succ_nat a')
- end.
-
-(** The opposite function *)
-
-Definition nat_of_ascii (a : ascii) : nat :=
- let (a1, a2, a3, a4, a5, a6, a7, a8) := a in
- 2 *
- (2 *
- (2 *
- (2 *
- (2 *
- (2 *
- (2 * (if a8 then 1 else 0)
- + (if a7 then 1 else 0))
- + (if a6 then 1 else 0))
- + (if a5 then 1 else 0))
- + (if a4 then 1 else 0))
- + (if a3 then 1 else 0))
- + (if a2 then 1 else 0))
- + (if a1 then 1 else 0).
-
-Theorem ascii_nat_embedding :
+Local Open Scope list_scope.
+
+Fixpoint N_of_digits (l:list bool) : N :=
+ match l with
+ | nil => 0
+ | b :: l' => (if b then 1 else 0) + 2*(N_of_digits l')
+ end%N.
+
+Definition N_of_ascii (a : ascii) : N :=
+ let (a0,a1,a2,a3,a4,a5,a6,a7) := a in
+ N_of_digits (a0::a1::a2::a3::a4::a5::a6::a7::nil).
+
+Definition nat_of_ascii (a : ascii) : nat := nat_of_N (N_of_ascii a).
+
+(** Proofs that we have indeed opposite function (below 256) *)
+
+Theorem ascii_N_embedding :
+ forall a : ascii, ascii_of_N (N_of_ascii a) = a.
+Proof.
+ destruct a as [[|][|][|][|][|][|][|][|]]; vm_compute; reflexivity.
+Qed.
+
+Theorem N_ascii_embedding :
+ forall n:N, (n < 256)%N -> N_of_ascii (ascii_of_N n) = n.
+Proof.
+destruct n.
+reflexivity.
+do 8 (destruct p; [ | | intros; vm_compute; reflexivity ]);
+ intro H; vm_compute in H; destruct p; discriminate.
+Qed.
+
+Theorem ascii_nat_embedding :
forall a : ascii, ascii_of_nat (nat_of_ascii a) = a.
Proof.
destruct a as [[|][|][|][|][|][|][|][|]]; compute; reflexivity.
Qed.
+Theorem nat_ascii_embedding :
+ forall n : nat, n < 256 -> nat_of_ascii (ascii_of_nat n) = n.
+Proof.
+ intros. unfold nat_of_ascii, ascii_of_nat.
+ rewrite N_ascii_embedding.
+ apply nat_of_N_of_nat.
+ unfold Nlt.
+ change 256%N with (N_of_nat 256).
+ rewrite <- N_of_nat_compare.
+ rewrite <- Compare_dec.nat_compare_lt. auto.
+Qed.
+
+
(** * Concrete syntax *)
(**
@@ -123,7 +136,7 @@ Qed.
Notice that the ascii characters of code >= 128 do not denote
stand-alone utf8 characters so that only the notation "nnn" is
available for them (unless your terminal is able to represent them,
- which is typically not the case in coqide).
+ which is typically not the case in coqide).
*)
Open Local Scope char_scope.
diff --git a/theories/Strings/String.v b/theories/Strings/String.v
index 00f28a9c..15f29821 100644
--- a/theories/Strings/String.v
+++ b/theories/Strings/String.v
@@ -1,3 +1,4 @@
+(* -*- coding: utf-8 -*- *)
(************************************************************************)
(* v * The Coq Proof Assistant / The Coq Development Team *)
(* <O___,, * CNRS-Ecole Polytechnique-INRIA Futurs-Universite Paris Sud *)
@@ -6,18 +7,19 @@
(* * GNU Lesser General Public License Version 2.1 *)
(************************************************************************)
-(* $Id: String.v 11206 2008-07-04 16:21:28Z letouzey $ *)
+(* $Id$ *)
-(** Contributed by Laurent Théry (INRIA);
+(** Contributed by Laurent Théry (INRIA);
Adapted to Coq V8 by the Coq Development Team *)
Require Import Arith.
Require Import Ascii.
+Declare ML Module "string_syntax_plugin".
(** *** Definition of strings *)
(** Implementation of string as list of ascii characters *)
-
+
Inductive string : Set :=
| EmptyString : string
| String : ascii -> string -> string.
@@ -36,7 +38,7 @@ Defined.
Reserved Notation "x ++ y" (right associativity, at level 60).
-Fixpoint append (s1 s2 : string) {struct s1} : string :=
+Fixpoint append (s1 s2 : string) : string :=
match s1 with
| EmptyString => s2
| String c s1' => String c (s1' ++ s2)
@@ -47,7 +49,7 @@ where "s1 ++ s2" := (append s1 s2) : string_scope.
(******************************)
(** Length *)
(******************************)
-
+
Fixpoint length (s : string) : nat :=
match s with
| EmptyString => 0
@@ -57,7 +59,7 @@ Fixpoint length (s : string) : nat :=
(******************************)
(** Nth character of a string *)
(******************************)
-
+
Fixpoint get (n : nat) (s : string) {struct s} : option ascii :=
match s with
| EmptyString => None
@@ -68,7 +70,7 @@ Fixpoint get (n : nat) (s : string) {struct s} : option ascii :=
end.
(** Two lists that are identical through get are syntactically equal *)
-
+
Theorem get_correct :
forall s1 s2 : string, (forall n : nat, get n s1 = get n s2) <-> s1 = s2.
Proof.
@@ -89,7 +91,7 @@ rewrite H1; auto.
Qed.
(** The first elements of [s1 ++ s2] are the ones of [s1] *)
-
+
Theorem append_correct1 :
forall (s1 s2 : string) (n : nat),
n < length s1 -> get n s1 = get n (s1 ++ s2).
@@ -102,7 +104,7 @@ apply lt_S_n; auto.
Qed.
(** The last elements of [s1 ++ s2] are the ones of [s2] *)
-
+
Theorem append_correct2 :
forall (s1 s2 : string) (n : nat),
get n s2 = get (n + length s1) (s1 ++ s2).
@@ -119,8 +121,8 @@ Qed.
(** [substring n m s] returns the substring of [s] that starts
at position [n] and of length [m];
if this does not make sense it returns [""] *)
-
-Fixpoint substring (n m : nat) (s : string) {struct s} : string :=
+
+Fixpoint substring (n m : nat) (s : string) : string :=
match n, m, s with
| 0, 0, _ => EmptyString
| 0, S m', EmptyString => s
@@ -130,7 +132,7 @@ Fixpoint substring (n m : nat) (s : string) {struct s} : string :=
end.
(** The substring is included in the initial string *)
-
+
Theorem substring_correct1 :
forall (s : string) (n m p : nat),
p < m -> get p (substring n m s) = get (p + n) s.
@@ -148,7 +150,7 @@ intros n' m p H; rewrite <- Plus.plus_Snm_nSm; simpl in |- *; auto.
Qed.
(** The substring has at most [m] elements *)
-
+
Theorem substring_correct2 :
forall (s : string) (n m p : nat), m <= p -> get p (substring n m s) = None.
Proof.
@@ -166,7 +168,7 @@ Qed.
(** *** Test functions *)
(** Test if [s1] is a prefix of [s2] *)
-
+
Fixpoint prefix (s1 s2 : string) {struct s2} : bool :=
match s1 with
| EmptyString => true
@@ -183,7 +185,7 @@ Fixpoint prefix (s1 s2 : string) {struct s2} : bool :=
(** If [s1] is a prefix of [s2], it is the [substring] of length
[length s1] starting at position [O] of [s2] *)
-
+
Theorem prefix_correct :
forall s1 s2 : string,
prefix s1 s2 = true <-> substring 0 (length s1) s2 = s1.
@@ -202,8 +204,8 @@ Qed.
(** Test if, starting at position [n], [s1] occurs in [s2]; if
so it returns the position *)
-
-Fixpoint index (n : nat) (s1 s2 : string) {struct s2} : option nat :=
+
+Fixpoint index (n : nat) (s1 s2 : string) : option nat :=
match s2, n with
| EmptyString, 0 =>
match s1 with
@@ -211,7 +213,7 @@ Fixpoint index (n : nat) (s1 s2 : string) {struct s2} : option nat :=
| String a s1' => None
end
| EmptyString, S n' => None
- | String b s2', 0 =>
+ | String b s2', 0 =>
if prefix s1 s2 then Some 0
else
match index 0 s1 s2' with
@@ -229,7 +231,7 @@ Fixpoint index (n : nat) (s1 s2 : string) {struct s2} : option nat :=
Opaque prefix.
(** If the result of [index] is [Some m], [s1] in [s2] at position [m] *)
-
+
Theorem index_correct1 :
forall (n m : nat) (s1 s2 : string),
index n s1 s2 = Some m -> substring m (length s1) s2 = s1.
@@ -259,9 +261,9 @@ intros x H H1; apply H; injection H1; intros H2; injection H2; auto.
intros; discriminate.
Qed.
-(** If the result of [index] is [Some m],
+(** If the result of [index] is [Some m],
[s1] does not occur in [s2] before [m] *)
-
+
Theorem index_correct2 :
forall (n m : nat) (s1 s2 : string),
index n s1 s2 = Some m ->
@@ -304,9 +306,9 @@ apply Lt.lt_S_n; auto.
intros; discriminate.
Qed.
-(** If the result of [index] is [None], [s1] does not occur in [s2]
+(** If the result of [index] is [None], [s1] does not occur in [s2]
after [n] *)
-
+
Theorem index_correct3 :
forall (n m : nat) (s1 s2 : string),
index n s1 s2 = None ->
@@ -348,7 +350,7 @@ Transparent prefix.
(** If we are searching for the [Empty] string and the answer is no
this means that [n] is greater than the size of [s] *)
-
+
Theorem index_correct4 :
forall (n : nat) (s : string),
index n EmptyString s = None -> length s < n.
@@ -367,7 +369,7 @@ Qed.
(** Same as [index] but with no optional type, we return [0] when it
does not occur *)
-
+
Definition findex n s1 s2 :=
match index n s1 s2 with
| Some n => n
diff --git a/theories/Strings/vo.itarget b/theories/Strings/vo.itarget
new file mode 100644
index 00000000..20813b42
--- /dev/null
+++ b/theories/Strings/vo.itarget
@@ -0,0 +1,2 @@
+Ascii.vo
+String.vo