[racket] Look-ahead in parser-tools?

From: Dmitry Pavlov (dpavlov at ipa.nw.ru)
Date: Tue Jan 17 07:53:19 EST 2012

Hello Simon,

Some observations that maybe will help:

1. Since the problem is obviously in the lexer,
you would probably prefer testing the lexer instead
of parser:

(define (test-lexer str)
   (let ((p (open-input-string str)))
     (port-count-lines! p)
     (let loop ()
       (let ((tok (position-token-token (toy-lexer p))))
         (printf "~a\n" tok)
         (unless (equal? tok 'eof)
           (loop))))))

2. You probably do not want the aliases to contain
whitespace:

(define-lex-abbrevs
   [...]
   (lex:whitespace (:or #\newline #\return #\tab #\space #\vtab)))


(define toy-lexer
   (lexer-src-pos
    [...]
    ((:&
     (:* (char-complement lex:whitespace))
     (complement (:: any-string "is" any-string))) (token-alias lexeme))
    [...]))

3. The problem with lexer is I think that it can not tell
"fact" from "alias" (because it can not look-ahead whether
there is a following "is" or not).

For example, if the input string is "remember somefact",
it gets "remember", skips the whitespace, and then
gets "somefact" as alias. To work that around, I would
change the lexer rules by putting mandatory quotes around
facts or something like that.

Best regards,

Dmitry



On 01/17/2012 05:35 AM, Simon Haines wrote:
> I've been playing around with parser-tools and am having difficulty
> expressing the following language:
>
> "remember <alias> is <email>"
> "remember <fact>"
>
> where <alias> is any string that does not contain the word 'is', <email>
> is a well-formed email address and <fact> is any string that does not
> match the previous constraints.
>
> Here's (stripped down) version of what I have so far:
> #lang racket
>
> (require parser-tools/lex
>           parser-tools/yacc
>           (prefix-in : parser-tools/lex-sre))
>
> (define-lex-abbrevs
>    (atext (:+ (:or alphabetic (:/ #\0 #\9) (char-set
> "!#$%&'*+-/=?^_`{|}~"))))
>    (dot-atom (:: atext (:* #\. atext))))
>
> (define-tokens toy-tokens (addr-spec alias fact))
> (define-empty-tokens empty-toy-tokens (eof REMEMBER IS))
>
> (define toy-lexer
>    (lexer-src-pos
>     ; Consume whitespace
>     ((:or #\tab #\space) (return-without-pos (toy-lexer input-port)))
>     ; Email addresses
>     ((:: dot-atom #\@ dot-atom) (token-addr-spec lexeme))
>
>     ; Commands
>     ("remember" 'REMEMBER)
>     ("is" 'IS)
>     ; ??? what to lex here ???
>     ((complement (:: any-string "is" any-string)) (token-alias lexeme))
>     (any-string (token-fact lexeme))))
>
> (define toy-parser
>    (parser
>     (tokens toy-tokens empty-toy-tokens)
>     (start start)
>     (end eof)
>     (error (lambda (a b c d e) (display (format "~a ~a ~a ~a ~a" a b c
>                                                 (position-offset d)
>                                                 (position-offset e)))))
>     (src-pos)
>     (grammar
>      (start (() #f)
>             ((REMEMBER alias IS addr-spec) `(alias ,$2 ,$4))
>             ((REMEMBER fact) `(fact ,$2))))))
>
> ; test
> (define (test str)
>    (let ((p (open-input-string str)))
>      (port-count-lines! p)
>      (toy-parser (lambda () (toy-lexer p)))))
>
> The problem I'm having is that the 'fact' lexer rule always matches
> without giving a chance for the other rules to attempt a match. Perhaps
> it is my ignorance with BNF. Can this language be expressed in this way?
> An alternative I've thought of is to create a lexer rule to just match
> "remember" then pass the port to another lexer that tries to look for
> "is" or (eof) and munge the result into a token. Alternatively I could
> try to regex the <alias>, <email> or <fact> clauses out and parse them
> separately, but I'd like to compose this toy parser into a larger one if
> possible. Yet I feel there is a simple technique here that I've missed
> in my ignorance. Any ideas?
> Many thanks, Simon.
>
>
>
>
> ____________________
>    Racket Users list:
>    http://lists.racket-lang.org/users



Posted on the users mailing list.