aboutsummaryrefslogtreecommitdiffhomepage
path: root/parsing
diff options
context:
space:
mode:
authorGravatar filliatr <filliatr@85f007b7-540e-0410-9357-904b9bb8a0f7>1999-09-06 15:14:07 +0000
committerGravatar filliatr <filliatr@85f007b7-540e-0410-9357-904b9bb8a0f7>1999-09-06 15:14:07 +0000
commiteec1ca7cdb14eadb6b6f671fe993e0a8b9489758 (patch)
tree160d8b746b03ff1785effb965d99c35857be0d4d /parsing
parent05074a8ee576e155596cf0c8bb3c8372a4dfa086 (diff)
debut d'un lexer
git-svn-id: svn+ssh://scm.gforge.inria.fr/svn/coq/trunk@37 85f007b7-540e-0410-9357-904b9bb8a0f7
Diffstat (limited to 'parsing')
-rw-r--r--parsing/.cvsignore1
-rw-r--r--parsing/lexer.mll133
2 files changed, 134 insertions, 0 deletions
diff --git a/parsing/.cvsignore b/parsing/.cvsignore
new file mode 100644
index 000000000..c1104f3ab
--- /dev/null
+++ b/parsing/.cvsignore
@@ -0,0 +1 @@
+lexer.ml
diff --git a/parsing/lexer.mll b/parsing/lexer.mll
new file mode 100644
index 000000000..bdd8d8016
--- /dev/null
+++ b/parsing/lexer.mll
@@ -0,0 +1,133 @@
+
+(* $Id$ *)
+
+{
+open Util
+
+type error =
+ | Illegal_character
+ | Unterminated_comment
+ | Unterminated_string
+
+exception Error of error * int * int
+
+let is_keyword =
+ let table = Hashtbl.create 149 in
+ List.iter (fun kw -> Hashtbl.add table kw ())
+ [ "Grammar"; "Syntax"; "Quit"; "Load"; "Compile";
+ "of"; "with"; "end"; "as"; "in"; "using";
+ "Cases"; "Fixpoint"; "CoFixpoint";
+ "Definition"; "Inductive"; "CoInductive";
+ "Theorem"; "Variable"; "Axiom"; "Parameter"; "Hypothesis";
+ "Orelse"; "Proof"; "Qed";
+ "Prop"; "Set"; "Type" ];
+ fun s -> try Hashtbl.find table s; true with Not_found -> false
+
+let char_for_backslash =
+ match Sys.os_type with
+ | "Unix" | "Win32" ->
+ begin function
+ | 'n' -> '\010'
+ | 'r' -> '\013'
+ | 'b' -> '\008'
+ | 't' -> '\009'
+ | c -> c
+ end
+ | "MacOS" ->
+ begin function
+ | 'n' -> '\013'
+ | 'r' -> '\010'
+ | 'b' -> '\008'
+ | 't' -> '\009'
+ | c -> c
+ end
+ | x -> error "Lexer: unknown system type"
+
+let char_for_decimal_code lexbuf i =
+ let c = 100 * (Char.code(Lexing.lexeme_char lexbuf i) - 48) +
+ 10 * (Char.code(Lexing.lexeme_char lexbuf (i+1)) - 48) +
+ (Char.code(Lexing.lexeme_char lexbuf (i+2)) - 48) in
+ Char.chr(c land 0xFF)
+
+let string_buffer = Buffer.create 80
+let string_start_pos = ref 0
+
+let comment_depth = ref 0
+let comment_start_pos = ref 0
+
+}
+
+let blank = [' ' '\010' '\013' '\009' '\012']
+let firstchar =
+ ['A'-'Z' 'a'-'z' '\192'-'\214' '\216'-'\246' '\248'-'\255']
+let identchar =
+ ['A'-'Z' 'a'-'z' '_' '\192'-'\214' '\216'-'\246' '\248'-'\255' '\'' '0'-'9']
+let symbolchar =
+ ['!' '$' '%' '&' '*' '+' '-' '.' '/' ':' '<' '=' '>' '?' '@' '^' '|' '~']
+
+rule token = parse
+ | blank+
+ { token lexbuf }
+ | firstchar identchar*
+ { let s = Lexing.lexeme lexbuf in
+ if is_keyword s then ("",s) else ("IDENT",s) }
+ | symbolchar+
+ { ("SPECIAL", Lexing.lexeme lexbuf) }
+ | '`' [^'`']* '`'
+ { ("QUOTED", Lexing.lexeme lexbuf) }
+ | "\""
+ { Buffer.reset string_buffer;
+ let string_start = Lexing.lexeme_start lexbuf in
+ string_start_pos := string_start;
+ string lexbuf;
+ ("STRING", Buffer.contents string_buffer) }
+ | "(*"
+ { comment_depth := 1;
+ comment_start_pos := Lexing.lexeme_start lexbuf;
+ comment lexbuf;
+ token lexbuf }
+
+and comment = parse
+ "(*"
+ { comment_depth := succ !comment_depth; comment lexbuf }
+ | "*)"
+ { comment_depth := pred !comment_depth;
+ if !comment_depth > 0 then comment lexbuf }
+ | "\""
+ { Buffer.reset string_buffer;
+ string_start_pos := Lexing.lexeme_start lexbuf;
+ string lexbuf;
+ comment lexbuf }
+ | "''"
+ { comment lexbuf }
+ | "'" [^ '\\' '\''] "'"
+ { comment lexbuf }
+ | "'\\" ['\\' '\'' 'n' 't' 'b' 'r'] "'"
+ { comment lexbuf }
+ | "'\\" ['0'-'9'] ['0'-'9'] ['0'-'9'] "'"
+ { comment lexbuf }
+ | eof
+ { raise (Error (Unterminated_comment,
+ !comment_start_pos, !comment_start_pos+2)) }
+ | _
+ { comment lexbuf }
+
+and string = parse
+ '"'
+ { () }
+ | '\\' ("\010" | "\013" | "\013\010") [' ' '\009'] *
+ { string lexbuf }
+ | '\\' ['\\' '"' 'n' 't' 'b' 'r']
+ { let c = char_for_backslash (Lexing.lexeme_char lexbuf 1) in
+ Buffer.add_char string_buffer c;
+ string lexbuf }
+ | '\\' ['0'-'9'] ['0'-'9'] ['0'-'9']
+ { Buffer.add_char string_buffer (char_for_decimal_code lexbuf 1);
+ string lexbuf }
+ | eof
+ { raise (Error (Unterminated_string,
+ !string_start_pos, !string_start_pos+1)) }
+ | _
+ { Buffer.add_char string_buffer (Lexing.lexeme_char lexbuf 0);
+ string lexbuf }
+