racket/collects/web-server/tests/tmp/htmlprag/htmlprag.ss
Jay McCarthy 19d59da08b Moving temporary code
svn: r7822
2007-11-23 18:56:31 +00:00

2384 lines
105 KiB
Scheme

;; THIS FILE IS GENERATED
(module htmlprag mzscheme
;;; @Package HtmlPrag
;;; @Subtitle Pragmatic Parsing and Emitting of HTML using SXML and SHTML
;;; @HomePage http://www.neilvandyke.org/htmlprag/
;;; @Author Neil W. Van Dyke
;;; @AuthorEmail neil@@neilvandyke.org
;;; @Version 0.16
;;; @Date 2005-12-18
;; $Id: htmlprag.scm,v 1.385 2005/12/19 03:28:28 neil Exp $
;;; @legal
;;; Copyright @copyright{} 2003 - 2005 Neil W. Van Dyke. This program is Free
;;; Software; you can redistribute it and/or modify it under the terms of the
;;; GNU Lesser General Public License as published by the Free Software
;;; Foundation; either version 2.1 of the License, or (at your option) any
;;; later version. This program is distributed in the hope that it will be
;;; useful, but without any warranty; without even the implied warranty of
;;; merchantability or fitness for a particular purpose. See
;;; @indicateurl{http://www.gnu.org/copyleft/lesser.html} for details. For
;;; other license options and consulting, contact the author.
;;; @end legal
(define-syntax %htmlprag:testeez
(syntax-rules () ((_ x ...)
;; (testeez x ...)
(error "Tests disabled.")
)))
;;; @section Introduction
;;; HtmlPrag provides permissive HTML parsing and emitting capability to Scheme
;;; programs. The parser is useful for software agent extraction of
;;; information from Web pages, for programmatically transforming HTML files,
;;; and for implementing interactive Web browsers. HtmlPrag emits ``SHTML,''
;;; which is an encoding of HTML in
;;; @uref{http://pobox.com/~oleg/ftp/Scheme/SXML.html, SXML}, so that
;;; conventional HTML may be processed with XML tools such as
;;; @uref{http://pair.com/lisovsky/query/sxpath/, SXPath}. Like Oleg
;;; Kiselyov's @uref{http://pobox.com/~oleg/ftp/Scheme/xml.html#HTML-parser,
;;; SSAX-based HTML parser}, HtmlPrag provides a permissive tokenizer, but also
;;; attempts to recover structure. HtmlPrag also includes procedures for
;;; encoding SHTML in HTML syntax.
;;;
;;; The HtmlPrag parsing behavior is permissive in that it accepts erroneous
;;; HTML, handling several classes of HTML syntax errors gracefully, without
;;; yielding a parse error. This is crucial for parsing arbitrary real-world
;;; Web pages, since many pages actually contain syntax errors that would
;;; defeat a strict or validating parser. HtmlPrag's handling of errors is
;;; intended to generally emulate popular Web browsers' interpretation of the
;;; structure of erroneous HTML. We euphemistically term this kind of parse
;;; ``pragmatic.''
;;;
;;; HtmlPrag also has some support for XHTML, although XML namespace qualifiers
;;; are currently accepted but stripped from the resulting SHTML. Note that
;;; valid XHTML input is of course better handled by a validating XML parser
;;; like Kiselyov's
;;; @uref{http://pobox.com/~oleg/ftp/Scheme/xml.html#XML-parser, SSAX}.
;;;
;;; HtmlPrag requires R5RS, SRFI-6, and SRFI-23.
;; The following bindings are used internally by HtmlPrag for portability, with
;; the intention that packagings of HtmlPrag use more efficient procedures for
;; the particular Scheme implementation. This is waiting on universal support
;; of SRFI-0.
;; @defproc %htmlprag:a2c num
;;
;; Returns the character with ASCII value @var{num}. In most Scheme
;; implementations, this is the same as @code{integer->char}. Two exceptions
;; are Scheme 48 0.57 and Scsh 0.6.3, for which the user must manually edit
;; file @code{htmlprag.scm} to bind this variable to @code{ascii->char}. A
;; future version of HtmlPrag will automatically use @code{ascii->char} where
;; available.
(define %htmlprag:a2c integer->char)
;; @defproc %htmlprag:append! a b
;;
;; Returns a concatenation of lists @var{a} and @var{b}, modifying the tail of
;; @var{a} to point to the head of @var{b} if both lists are non-null. A
;; future version should use the more general @code{append!} where available.
(define (%htmlprag:append! a b)
(append a b))
;; @defproc %htmlprag:reverse!ok lst
;;
;; Returns a reversed list @var{lst}, possibly destructive. A future version
;; will use @code{reverse!} where available, and @code{reverse} elsewhere.
(define %htmlprag:reverse!ok reverse)
;; @defproc %htmlprag:down str
;;
;; Returns a string that is equivalent to @var{str} with all characters mapped
;; to lowercase, as if by @code{char-downcase}, without mutating @var{str}. A
;; future version should use the Scheme implementation's native nondestructive
;; procedure where available.
;; @defproc %htmlprag:error proc-str msg obj
;;
;; For Bigloo, this is changed to:
;;
;; @lisp
;; (define %htmlprag:error error)
;; @end lisp
;; TODO: Maybe go back to requiring a SRFI-23 "error".
(define-syntax %htmlprag:error
(syntax-rules ()
((_ p m o) (error (string-append p " : " m) o))
;; ((_ p m o) (error p m o))))
))
(define (%htmlprag:down s)
(list->string (map char-downcase (string->list s))))
;; @defproc %htmlprag:down!ok str
;;
;; Returns a string that is equivalent to @var{str} with all characters mapped
;; to lowercase, as if by @code{char-downcase}, possibly mutating @var{str}.
;; A future version should use the Scheme implementation's native destructive
;; or nondestructive procedure where available.
(define %htmlprag:down!ok %htmlprag:down)
;; @defproc %htmlprag:gosc os
;;
;; One-shot version of the conventional @code{get-output-string}. The result
;; of any subsequent attempt to write to the port or get the output string is
;; undefined. This may or may not free up resources.
(define (%htmlprag:gosc os)
(let ((str (get-output-string os)))
;; Note: By default, we don't call close-output-port, since at least one
;; tested Scheme implementation barfs on that.
;;
;; (close-output-port os)
str))
;;; @section SHTML and SXML
;;; SHTML is a variant of SXML, with two minor but useful extensions:
;;;
;;; @enumerate
;;;
;;; @item
;;; The SXML keyword symbols, such as @code{*TOP*}, are defined to be in all
;;; uppercase, regardless of the case-sensitivity of the reader of the hosting
;;; Scheme implementation in any context. This avoids several pitfalls.
;;;
;;; @item
;;; Since not all character entity references used in HTML can be converted to
;;; Scheme characters in all R5RS Scheme implementations, nor represented in
;;; conventional text files or other common external text formats to which one
;;; might wish to write SHTML, SHTML adds a special @code{&} syntax for
;;; non-ASCII (or non-Extended-ASCII) characters. The syntax is @code{(&
;;; @var{val})}, where @var{val} is a symbol or string naming with the symbolic
;;; name of the character, or an integer with the numeric value of the
;;; character.
;;;
;;; @end enumerate
;;; @defvar shtml-comment-symbol
;;; @defvarx shtml-decl-symbol
;;; @defvarx shtml-empty-symbol
;;; @defvarx shtml-end-symbol
;;; @defvarx shtml-entity-symbol
;;; @defvarx shtml-pi-symbol
;;; @defvarx shtml-start-symbol
;;; @defvarx shtml-text-symbol
;;; @defvarx shtml-top-symbol
;;;
;;; These variables are bound to the following case-sensitive symbols used in
;;; SHTML, respectively: @code{*COMMENT*}, @code{*DECL*}, @code{*EMPTY*},
;;; @code{*END*}, @code{*ENTITY*}, @code{*PI*}, @code{*START*}, @code{*TEXT*},
;;; and @code{*TOP*}. These can be used in lieu of the literal symbols in
;;; programs read by a case-insensitive Scheme reader.@footnote{Scheme
;;; implementators who have not yet made @code{read} case-sensitive by default
;;; are encouraged to do so.}
(define shtml-comment-symbol (string->symbol "*COMMENT*"))
(define shtml-decl-symbol (string->symbol "*DECL*"))
(define shtml-empty-symbol (string->symbol "*EMPTY*"))
(define shtml-end-symbol (string->symbol "*END*"))
(define shtml-entity-symbol (string->symbol "*ENTITY*"))
(define shtml-pi-symbol (string->symbol "*PI*"))
(define shtml-start-symbol (string->symbol "*START*"))
(define shtml-text-symbol (string->symbol "*TEXT*"))
(define shtml-top-symbol (string->symbol "*TOP*"))
;;; @defvar shtml-named-char-id
;;; @defvarx shtml-numeric-char-id
;;;
;;; These variables are bound to the SHTML entity public identifier strings
;;; used in SHTML @code{*ENTITY*} named and numeric character entity
;;; references.
(define shtml-named-char-id "shtml-named-char")
(define shtml-numeric-char-id "shtml-numeric-char")
;;; @defproc make-shtml-entity val
;;;
;;; Yields an SHTML character entity reference for @var{val}. For example:
;;;
;;; @lisp
;;; (make-shtml-entity "rArr") @result{} (& rArr)
;;; (make-shtml-entity (string->symbol "rArr")) @result{} (& rArr)
;;; (make-shtml-entity 151) @result{} (& 151)
;;; @end lisp
(define (make-shtml-entity val)
(list '& (cond ((symbol? val) val)
((integer? val) val)
((string? val) (string->symbol val))
(else (%htmlprag:error "make-shtml-entity"
"invalid SHTML entity value:"
val)))))
;; TODO:
;;
;; (define (shtml-entity? x)
;; (and (shtml-entity-value entity) #t))
;;; @defproc shtml-entity-value obj
;;;
;;; Yields the value for the SHTML entity @var{obj}, or @code{#f} if @var{obj}
;;; is not a recognized entity. Values of named entities are symbols, and
;;; values of numeric entities are numbers. An error may raised if @var{obj}
;;; is an entity with system ID inconsistent with its public ID. For example:
;;;
;;; @lisp
;;; (define (f s) (shtml-entity-value (cadr (html->shtml s))))
;;; (f " ") @result{} nbsp
;;; (f "ߐ") @result{} 2000
;;; @end lisp
(define (shtml-entity-value entity)
(cond ((not (pair? entity)) #f)
((null? (cdr entity)) #f)
((eqv? (car entity) '&)
;; TODO: Error-check for extraneous list members?
(let ((val (cadr entity)))
(cond ((symbol? val) val)
((integer? val) val)
((string? val) (string->symbol val))
(else #f))))
((eqv? (car entity) shtml-entity-symbol)
(if (null? (cddr entity))
#f
(let ((public-id (list-ref entity 1))
(system-id (list-ref entity 2)))
;; TODO: Error-check for extraneous list members?
(cond ((equal? public-id shtml-named-char-id)
(string->symbol system-id))
((equal? public-id shtml-numeric-char-id)
(string->number system-id))
(else #f)))))
(else #f)))
;;; @section Tokenizing
;;; The tokenizer is used by the higher-level structural parser, but can also
;;; be called directly for debugging purposes or unusual applications. Some of
;;; the list structure of tokens, such as for start tag tokens, is mutated and
;;; incorporated into the SHTML list structure emitted by the parser.
;; TODO: Document the token format.
;;; @defproc make-html-tokenizer in normalized?
;;;
;;; Constructs an HTML tokenizer procedure on input port @var{in}. If boolean
;;; @var{normalized?} is true, then tokens will be in a format conducive to use
;;; with a parser emitting normalized SXML. Each call to the resulting
;;; procedure yields a successive token from the input. When the tokens have
;;; been exhausted, the procedure returns the null list. For example:
;;;
;;; @lisp
;;; (define input (open-input-string "<a href=\"foo\">bar</a>"))
;;; (define next (make-html-tokenizer input #f))
;;; (next) @result{} (a (@@ (href "foo")))
;;; (next) @result{} "bar"
;;; (next) @result{} (*END* a)
;;; (next) @result{} ()
;;; (next) @result{} ()
;;; @end lisp
(define make-html-tokenizer
;; TODO: Have the tokenizer replace contiguous whitespace within individual
;; text tokens with single space characters (except for when in `pre' and
;; verbatim elements). The parser will introduce new contiguous whitespace
;; (e.g., when text tokens are concatenated, invalid end tags are removed,
;; whitespace is irrelevant between certain elements), but then the parser
;; only has to worry about the first and last character of each string.
;; Perhaps the text tokens should have both leading and trailing whitespace
;; stripped, and contain flags for whether or not leading and trailing
;; whitespace occurred.
(letrec ((no-token '())
;; TODO: Maybe make these three variables options.
(verbatim-to-eof-elems '(plaintext))
(verbatim-pair-elems '(script server style xmp))
(ws-chars (list #\space
(%htmlprag:a2c 9)
(%htmlprag:a2c 10)
(%htmlprag:a2c 11)
(%htmlprag:a2c 12)
(%htmlprag:a2c 13)))
(gosc/string-or-false
(lambda (os)
(let ((s (%htmlprag:gosc os)))
(if (string=? s "") #f s))))
(gosc/symbol-or-false
(lambda (os)
(let ((s (gosc/string-or-false os)))
(if s (string->symbol s) #f))))
)
(lambda (in normalized?)
;; TODO: Make a tokenizer option that causes XML namespace qualifiers to
;; be ignored.
(letrec
(
;; Port buffer with inexpensive unread of one character and slightly
;; more expensive pushback of second character to unread. The
;; procedures themselves do no consing. The tokenizer currently
;; needs two-symbol lookahead, due to ambiguous "/" while parsing
;; element and attribute names, which could be either empty-tag
;; syntax or XML qualified names.
(c #f)
(next-c #f)
(c-consumed? #t)
(read-c (lambda ()
(if c-consumed?
(if next-c
(begin (set! c next-c)
(set! next-c #f))
(set! c (read-char in)))
(set! c-consumed? #t))))
(unread-c (lambda ()
(if c-consumed?
(set! c-consumed? #f)
;; TODO: Procedure name in error message really
;; isn't "make-html-tokenizer"...
(%htmlprag:error "make-html-tokenizer"
"already unread:"
c))))
(push-c (lambda (new-c)
(if c-consumed?
(begin (set! c new-c)
(set! c-consumed? #f))
(if next-c
(%htmlprag:error
"make-html-tokenizer"
"pushback full:"
c)
(begin (set! next-c c)
(set! c new-c)
(set! c-consumed? #f))))))
;; TODO: These procedures are a temporary convenience for
;; enumerating the pertinent character classes, with an eye towards
;; removing redundant tests of character class. These procedures
;; should be eliminated in a future version.
(c-eof? (lambda () (eof-object? c)))
(c-amp? (lambda () (eqv? c #\&)))
(c-apos? (lambda () (eqv? c #\')))
(c-bang? (lambda () (eqv? c #\!)))
(c-colon? (lambda () (eqv? c #\:)))
(c-quot? (lambda () (eqv? c #\")))
(c-equals? (lambda () (eqv? c #\=)))
(c-gt? (lambda () (eqv? c #\>)))
(c-lsquare? (lambda () (eqv? c #\[)))
(c-lt? (lambda () (eqv? c #\<)))
(c-minus? (lambda () (eqv? c #\-)))
(c-pound? (lambda () (eqv? c #\#)))
(c-ques? (lambda () (eqv? c #\?)))
(c-semi? (lambda () (eqv? c #\;)))
(c-slash? (lambda () (eqv? c #\/)))
(c-splat? (lambda () (eqv? c #\*)))
(c-lf? (lambda () (eqv? c #\newline)))
(c-angle? (lambda () (memv c '(#\< #\>))))
(c-ws? (lambda () (memv c ws-chars)))
(c-alpha? (lambda () (char-alphabetic? c)))
(c-digit? (lambda () (char-numeric? c)))
(c-alphanum? (lambda () (or (c-alpha?) (c-digit?))))
(c-hexlet? (lambda () (memv c '(#\a #\b #\c #\d #\e #\f
#\A #\B #\C #\D #\E #\F))))
(skip-ws (lambda () (read-c) (if (c-ws?) (skip-ws) (unread-c))))
(if-read-chars
(lambda (match-chars yes-thunk no-proc)
(let loop ((chars match-chars)
(match-count 0))
(if (null? chars)
(yes-thunk)
(begin (read-c)
(if (eqv? c (car chars))
(begin (loop (cdr chars) (+ 1 match-count)))
(begin (unread-c)
(no-proc match-chars match-count))))))))
(write-chars-count
(lambda (chars count port)
(let loop ((chars chars)
(count count))
(or (zero? count)
(begin (write-char (car chars) port)
(loop (cdr chars)
(- count 1)))))))
(make-start-token
(if normalized?
(lambda (name ns attrs)
(list name (cons '@ attrs)))
(lambda (name ns attrs)
(if (null? attrs)
(list name)
(list name (cons '@ attrs))))))
(make-empty-token
(lambda (name ns attrs)
(cons shtml-empty-symbol
(make-start-token name ns attrs))))
(make-end-token
(if normalized?
(lambda (name ns attrs)
(list shtml-end-symbol
name
(cons '@ attrs)))
(lambda (name ns attrs)
(if (null? attrs)
(list shtml-end-symbol name)
(list shtml-end-symbol
name
(cons '@ attrs))))))
(make-comment-token
(lambda (str) (list shtml-comment-symbol str)))
(make-decl-token
(lambda (parts) (cons shtml-decl-symbol parts)))
(scan-qname
;; TODO: Make sure we don't accept local names that have "*", since
;; this can break SXML tools. Have to validate this afterwards if
;; "verbatim-safe?". Also check for "@" and maybe "@@". Check
;; qname parsing code, especially for verbatim mode. This is
;; important!
(lambda (verbatim-safe?)
;; Note: If we accept some invalid local names, we only need two
;; symbols of lookahead to determine the end of a qname.
(letrec ((os #f)
(ns '())
(vcolons 0)
(good-os (lambda ()
(or os
(begin (set! os (open-output-string))
os)))))
(let loop ()
(read-c)
(cond ((c-eof?) #f)
((or (c-ws?) (c-splat?))
(if verbatim-safe?
(unread-c)))
((or (c-angle?) (c-equals?) (c-quot?) (c-apos?))
(unread-c))
((c-colon?)
(or (null? ns)
(set! ns (cons ":" ns)))
(if os
(begin
(set! ns (cons (%htmlprag:gosc os)
ns))
(set! os #f)))
(loop))
((c-slash?)
(read-c)
(cond ((or (c-eof?)
(c-ws?)
(c-equals?)
(c-apos?)
(c-quot?)
(c-angle?)
(c-splat?))
(unread-c)
(push-c #\/))
(else (write-char #\/ (good-os))
(write-char c os)
(loop))))
(else (write-char c (good-os))
(loop))))
(let ((ns (if (null? ns)
#f
(apply string-append
(%htmlprag:reverse!ok ns))))
(local (if os (%htmlprag:gosc os) #f)))
(if verbatim-safe?
;; TODO: Make sure we don't have ambiguous ":" or drop
;; any characters!
(cons ns local)
;; Note: We represent "xml:" and "xmlns:" syntax as
;; normal qnames, for lack of something better to do with
;; them when we don't support XML namespaces.
;;
;; TODO: Local names are currently forced to lowercase,
;; since HTML is usually case-insensitive. If XML
;; namespaces are used, we might wish to keep local names
;; case-sensitive.
(if local
(if ns
(if (or (string=? ns "xml")
(string=? ns "xmlns"))
(string->symbol (string-append ns ":" local))
(cons ns
(string->symbol
(%htmlprag:down!ok
local))))
(string->symbol
(%htmlprag:down!ok local)))
(if ns
(string->symbol
(%htmlprag:down!ok ns))
;; TODO: Ensure in rest of code that returning #f
;; as a name here is OK.
#f)))))))
(scan-tag
(lambda (start?)
(skip-ws)
(let ((tag-name (scan-qname #f))
(tag-ns #f)
(tag-attrs #f)
(tag-empty? #f))
;; Scan element name.
(if (pair? tag-name)
(begin (set! tag-ns (car tag-name))
(set! tag-name (cdr tag-name))))
;; TODO: Ensure there's no case in which a #f tag-name isn't
;; compensated for later.
;;
;; Scan element attributes.
(set! tag-attrs
(let scan-attr-list ()
(read-c)
(cond ((c-eof?) '())
((c-angle?) (unread-c) '())
((c-slash?)
(set! tag-empty? #t)
(scan-attr-list))
((c-alpha?)
(unread-c)
(let ((attr (scan-attr)))
(cons attr (scan-attr-list))))
(else (scan-attr-list)))))
;; Find ">" or unnatural end.
(let loop ()
(read-c)
(cond ((c-eof?) no-token)
((c-slash?) (set! tag-empty? #t) (loop))
((c-gt?) #f)
((c-ws?) (loop))
(else (unread-c))))
;; Change the tokenizer mode if necessary.
(cond ((not start?) #f)
(tag-empty? #f)
;; TODO: Maybe make one alist lookup here, instead of
;; two.
((memq tag-name verbatim-to-eof-elems)
(set! nexttok verbeof-nexttok))
((memq tag-name verbatim-pair-elems)
(set! nexttok (make-verbpair-nexttok tag-name))))
;; Return a token object.
(if start?
(if tag-empty?
(make-empty-token tag-name tag-ns tag-attrs)
(make-start-token tag-name tag-ns tag-attrs))
(make-end-token tag-name tag-ns tag-attrs)))))
(scan-attr
(lambda ()
(let ((name (scan-qname #f))
(val #f))
(if (pair? name)
(set! name (cdr name)))
(let loop-equals-or-end ()
(read-c)
(cond ((c-eof?) no-token)
((c-ws?) (loop-equals-or-end))
((c-equals?)
(let loop-quote-or-unquoted ()
(read-c)
(cond ((c-eof?) no-token)
((c-ws?) (loop-quote-or-unquoted))
((or (c-apos?) (c-quot?))
(let ((term c))
(set! val (open-output-string))
(let loop-quoted-val ()
(read-c)
(cond ((c-eof?) #f)
((eqv? c term) #f)
(else (write-char c val)
(loop-quoted-val))))))
((c-angle?) (unread-c))
(else
(set! val (open-output-string))
(write-char c val)
(let loop-unquoted-val ()
(read-c)
(cond ((c-eof?) no-token)
((c-apos?) #f)
((c-quot?) #f)
((or (c-ws?) (c-angle?)
;;(c-slash?)
)
(unread-c))
;; Note: We can treat a slash in an
;; unquoted attribute value as a
;; value constituent because the
;; slash is specially-handled only
;; for XHTML, and XHTML attribute
;; values must always be quoted. We
;; could do lookahead for "/>", but
;; that wouldn't let us parse HTML
;; "<a href=/>" correctly, so this is
;; an easier and more correct way to
;; do things.
(else (write-char c val)
(loop-unquoted-val))))))))
(else (unread-c))))
(if normalized?
(list name (if val
(%htmlprag:gosc val)
(symbol->string name)))
(if val
(list name (%htmlprag:gosc val))
(list name))))))
(scan-comment
;; TODO: Rewrite this to use tail recursion rather than a state
;; variable.
(lambda ()
(let ((os (open-output-string))
(state 'start-minus))
(let loop ()
(read-c)
(cond ((c-eof?) #f)
((c-minus?)
(set! state
(case state
((start-minus) 'start-minus-minus)
((start-minus-minus body) 'end-minus)
((end-minus) 'end-minus-minus)
((end-minus-minus) (write-char #\- os) state)
(else (%htmlprag:error
"make-html-tokenizer"
"invalid state:"
state))))
(loop))
((and (c-gt?) (eq? state 'end-minus-minus)) #f)
(else (case state
((end-minus) (write-char #\- os))
((end-minus-minus) (display "--" os)))
(set! state 'body)
(write-char c os)
(loop))))
(make-comment-token (%htmlprag:gosc os)))))
(scan-possible-cdata
(lambda ()
;; Read "<!" and current character is "[", so try to read the
;; rest of the CDATA start delimeter.
(if-read-chars
'(#\C #\D #\A #\T #\A #\[)
(lambda ()
;; Successfully read CDATA section start delimiter, so read
;; the section.
(scan-cdata))
(lambda (chars count)
;; Did not read rest of CDATA section start delimiter, so
;; return a string for what we did read.
(let ((os (open-output-string)))
(display "<![" os)
(write-chars-count chars count os)
(%htmlprag:gosc os))))))
(scan-cdata
(lambda ()
(let ((os (open-output-string)))
(let loop ()
(if-read-chars
'(#\] #\] #\>)
(lambda () (%htmlprag:gosc os))
(lambda (chars count)
(if (zero? count)
(if (eof-object? c)
(%htmlprag:gosc os)
(begin (write-char c os)
(read-c)
(loop)))
(begin (write-char #\] os)
(if (= count 2)
(push-c #\]))
(loop)))))))))
(scan-pi
(lambda ()
(skip-ws)
(let ((name (open-output-string))
(val (open-output-string)))
(let scan-name ()
(read-c)
(cond ((c-eof?) #f)
((c-ws?) #f)
((c-alpha?) (write-char c name) (scan-name))
(else (unread-c))))
;; TODO: Do we really want to emit #f for PI name?
(set! name (gosc/symbol-or-false name))
(let scan-val ()
(read-c)
(cond ((c-eof?) #f)
;; ((c-amp?) (display (scan-entity) val)
;; (scan-val))
((c-ques?)
(read-c)
(cond ((c-eof?) (write-char #\? val))
((c-gt?) #f)
(else (write-char #\? val)
(unread-c)
(scan-val))))
(else (write-char c val) (scan-val))))
(list shtml-pi-symbol
name
(%htmlprag:gosc val)))))
(scan-decl
;; TODO: Find if SXML includes declaration forms, and if so, use
;; whatever format SXML wants.
;;
;; TODO: Rewrite to eliminate state variables.
(letrec
((scan-parts
(lambda ()
(let ((part (open-output-string))
(nonsymbol? #f)
(state 'before)
(last? #f))
(let loop ()
(read-c)
(cond ((c-eof?) #f)
((c-ws?)
(case state
((before) (loop))
((quoted) (write-char c part) (loop))))
((and (c-gt?) (not (eq? state 'quoted)))
(set! last? #t))
((and (c-lt?) (not (eq? state 'quoted)))
(unread-c))
((c-quot?)
(case state
((before) (set! state 'quoted) (loop))
((unquoted) (unread-c))
((quoted) #f)))
(else
(if (eq? state 'before)
(set! state 'unquoted))
(set! nonsymbol? (or nonsymbol?
(not (c-alphanum?))))
(write-char c part)
(loop))))
(set! part (%htmlprag:gosc part))
(if (string=? part "")
'()
(cons (if (or (eq? state 'quoted) nonsymbol?)
part
;; TODO: Normalize case of things we make
;; into symbols here.
(string->symbol part))
(if last?
'()
(scan-parts))))))))
(lambda () (make-decl-token (scan-parts)))))
(scan-entity
(lambda ()
(read-c)
(cond ((c-eof?) "&")
((c-alpha?)
;; TODO: Do entity names have a maximum length?
(let ((name (open-output-string)))
(write-char c name)
(let loop ()
(read-c)
(cond ((c-eof?) #f)
((c-alpha?) (write-char c name) (loop))
((c-semi?) #f)
(else (unread-c))))
(set! name (%htmlprag:gosc name))
;; TODO: Make the entity map an option.
(let ((pair (assoc name '(("amp" . "&")
("apos" . "'")
("gt" . ">")
("lt" . "<")
("quot" . "\"")))))
(if pair
(cdr pair)
(make-shtml-entity name)))))
((c-pound?)
(let ((num (open-output-string))
(hex? #f))
(read-c)
(cond ((c-eof?) #f)
((memv c '(#\x #\X)) (set! hex? #t) (read-c)))
(let loop ()
(cond ((c-eof?) #f)
((c-semi?) #f)
((or (c-digit?) (and hex? (c-hexlet?)))
(write-char c num)
(read-c)
(loop))
(else (unread-c))))
(set! num (%htmlprag:gosc num))
(if (string=? num "")
"&#;"
(let ((n (string->number num (if hex? 16 10))))
(if (<= 32 n 126)
;; (and (<= 32 n 255) (not (= n 127)))
(string (%htmlprag:a2c n))
(make-shtml-entity n))))))
(else (unread-c) "&"))))
(normal-nexttok
(lambda ()
(read-c)
(cond ((c-eof?) no-token)
((c-lt?)
(let loop ()
(read-c)
(cond ((c-eof?) "<")
((c-ws?) (loop))
((c-slash?) (scan-tag #f))
((c-ques?) (scan-pi))
((c-alpha?) (unread-c) (scan-tag #t))
((c-bang?)
(read-c)
(if (c-lsquare?)
(scan-possible-cdata)
(let loop ()
(cond ((c-eof?) no-token)
((c-ws?) (read-c) (loop))
((c-minus?) (scan-comment))
(else (unread-c)
(scan-decl))))))
(else (unread-c) "<"))))
((c-gt?) ">")
(else (let ((os (open-output-string)))
(let loop ()
(cond ((c-eof?) #f)
((c-angle?) (unread-c))
((c-amp?)
(let ((entity (scan-entity)))
(if (string? entity)
(begin (display entity os)
(read-c)
(loop))
(let ((saved-nexttok nexttok))
(set! nexttok
(lambda ()
(set! nexttok
saved-nexttok)
entity))))))
(else (write-char c os)
(or (c-lf?)
(begin (read-c) (loop))))))
(let ((text (%htmlprag:gosc os)))
(if (equal? text "")
(nexttok)
text)))))))
(verbeof-nexttok
(lambda ()
(read-c)
(if (c-eof?)
no-token
(let ((os (open-output-string)))
(let loop ()
(or (c-eof?)
(begin (write-char c os)
(or (c-lf?)
(begin (read-c) (loop))))))
(%htmlprag:gosc os)))))
(make-verbpair-nexttok
(lambda (elem-name)
(lambda ()
(let ((os (open-output-string)))
;; Accumulate up to a newline-terminated line.
(let loop ()
(read-c)
(cond ((c-eof?)
;; Got EOF in verbatim context, so set the normal
;; nextok procedure, then fall out of loop.
(set! nexttok normal-nexttok))
((c-lt?)
;; Got "<" in verbatim context, so get next
;; character.
(read-c)
(cond ((c-eof?)
;; Got "<" then EOF, so set to the normal
;; nexttok procedure, add the "<" to the
;; verbatim string, and fall out of loop.
(set! nexttok normal-nexttok)
(write-char #\< os))
((c-slash?)
;; Got "</", so...
(read-c)
(cond
((c-eof?)
(display "</" os))
((c-alpha?)
;; Got "</" followed by alpha, so unread
;; the alpha, scan qname, compare...
(unread-c)
(let* ((vqname (scan-qname #t))
(ns (car vqname))
(local (cdr vqname)))
;; Note: We ignore XML namespace
;; qualifier for purposes of comparison.
;;
;; Note: We're interning strings here for
;; comparison when in theory there could
;; be many such unique interned strings
;; in a valid HTML document, although in
;; practice this should not be a problem.
(if (and local
(eqv? (string->symbol
(%htmlprag:down
local))
elem-name))
;; This is the terminator tag, so
;; scan to the end of it, set the
;; nexttok, and fall out of the loop.
(begin
(let scan-to-end ()
(read-c)
(cond ((c-eof?) #f)
((c-gt?) #f)
((c-lt?) (unread-c))
((c-alpha?)
(unread-c)
;; Note: This is an
;; expensive way to skip
;; over an attribute, but
;; in practice more
;; verbatim end tags will
;; not have attributes.
(scan-attr)
(scan-to-end))
(else (scan-to-end))))
(set! nexttok
(lambda ()
(set! nexttok
normal-nexttok)
(make-end-token
elem-name #f '()))))
;; This isn't the terminator tag, so
;; add to the verbatim string the
;; "</" and the characters of what we
;; were scanning as a qname, and
;; recurse in the loop.
(begin
(display "</" os)
(if ns
(begin (display ns os)
(display ":" os)))
(if local
(display local os))
(loop)))))
(else
;; Got "</" and non-alpha, so unread new
;; character, add the "</" to verbatim
;; string, then loop.
(unread-c)
(display "</" os)
(loop))))
(else
;; Got "<" and non-slash, so unread the new
;; character, write the "<" to the verbatim
;; string, then loop.
(unread-c)
(write-char #\< os)
(loop))))
(else
;; Got non-"<" in verbatim context, so just add it
;; to the buffer, then, if it's not a linefeed, fall
;; out of the loop so that the token can be
;; returned.
(write-char c os)
(or (c-lf?) (loop)))))
;; Return the accumulated line string, if non-null, or call
;; nexttok.
(or (gosc/string-or-false os) (nexttok))))))
(nexttok #f))
(set! nexttok normal-nexttok)
(lambda () (nexttok))))))
;;; @defproc tokenize-html in normalized?
;;;
;;; Returns a list of tokens from input port @var{in}, normalizing according to
;;; boolean @var{normalized?}. This is probably most useful as a debugging
;;; convenience. For example:
;;;
;;; @lisp
;;; (tokenize-html (open-input-string "<a href=\"foo\">bar</a>") #f)
;;; @result{} ((a (@@ (href "foo"))) "bar" (*END* a))
;;; @end lisp
(define (tokenize-html in normalized?)
(let ((next-tok (make-html-tokenizer in normalized?)))
(let loop ((tok (next-tok)))
(if (null? tok)
'()
(cons tok (loop (next-tok)))))))
;;; @defproc shtml-token-kind token
;;;
;;; Returns a symbol indicating the kind of tokenizer @var{token}:
;;; @code{*COMMENT*}, @code{*DECL*}, @code{*EMPTY*}, @code{*END*},
;;; @code{*ENTITY*}, @code{*PI*}, @code{*START*}, @code{*TEXT*}.
;;; This is used by higher-level parsing code. For example:
;;;
;;; @lisp
;;; (map shtml-token-kind
;;; (tokenize-html (open-input-string "<a<b>><c</</c") #f))
;;; @result{} (*START* *START* *TEXT* *START* *END* *END*)
;;; @end lisp
(define (shtml-token-kind token)
(cond ((string? token) shtml-text-symbol)
((list? token)
(let ((s (list-ref token 0)))
(if (memq s `(,shtml-comment-symbol
,shtml-decl-symbol
,shtml-empty-symbol
,shtml-end-symbol
,shtml-entity-symbol
,shtml-pi-symbol))
s
shtml-start-symbol)))
(else (%htmlprag:error "shtml-token-kind"
"unrecognized token kind:"
token))))
;;; @section Parsing
;;; Most applications will call a parser procedure such as
;;; @code{html->shtml} rather than calling the tokenizer directly.
;; @defvar %htmlprag:empty-elements
;;
;; List of names of HTML element types that have no content, represented as a
;; list of symbols. This is used internally by the parser and encoder. The
;; effect of mutating this list is undefined.
;; TODO: Document exactly which elements these are, after we make the new
;; parameterized parser constructor.
(define %htmlprag:empty-elements
'(& area base br frame hr img input isindex keygen link meta object param
spacer wbr))
;;; @defproc parse-html/tokenizer tokenizer normalized?
;;;
;;; Emits a parse tree like @code{html->shtml} and related procedures, except
;;; using @var{tokenizer} as a source of tokens, rather than tokenizing from an
;;; input port. This procedure is used internally, and generally should not be
;;; called directly.
(define parse-html/tokenizer
;; TODO: Document the algorithm, then see if rewriting as idiomatic Scheme
;; can make it more clear.
(letrec ((empty-elements
;; TODO: Maybe make this an option. This might also be an
;; acceptable way to parse old HTML that uses the `p' element as a
;; paragraph terminator.
%htmlprag:empty-elements)
(parent-constraints
;; TODO: Maybe make this an option.
'((area . (map))
(body . (html))
(caption . (table))
(colgroup . (table))
(dd . (dl))
(dt . (dl))
(frame . (frameset))
(head . (html))
(isindex . (head))
(li . (dir menu ol ul))
(meta . (head))
(noframes . (frameset))
(option . (select))
(p . (body td th))
(param . (applet))
(tbody . (table))
(td . (tr))
(th . (tr))
(thead . (table))
(title . (head))
(tr . (table tbody thead))))
(start-tag-name (lambda (tag-token) (car tag-token)))
(end-tag-name (lambda (tag-token) (list-ref tag-token 1))))
(lambda (tokenizer normalized?)
;; Example `begs' value:
;;
;; ( ((head ...) . ( (title ...) ))
;; ((html ...) . ( (head ...) (*COMMENT* ...) ))
;; (#f . ( (html ...) (*DECL* doctype ...) )) )
(let ((begs (mcons (mcons #f '()) null)))
(letrec ((add-to-current-beg
(lambda (tok)
(set-mcdr! (mcar begs) (mcons tok (mcdr (mcar begs))))))
(finish-all-begs
(lambda ()
(let ((toplist #f))
(map (lambda (beg) (set! toplist (finish-beg beg)))
begs)
toplist)))
(finish-beg
(lambda (beg)
(let ((start-tok (car beg)))
(if start-tok
(%htmlprag:append!
(car beg)
(%htmlprag:reverse!ok (cdr beg)))
(%htmlprag:reverse!ok (cdr beg))))))
(finish-begs-to
(lambda (name lst)
(let* ((top (car lst))
(starttag (car top)))
(cond ((not starttag) #f)
((eqv? name (start-tag-name starttag))
(set! begs (cdr lst))
(finish-beg top)
#t)
(else (if (finish-begs-to name (cdr lst))
(begin (finish-beg top) #t)
#f))))))
(finish-begs-upto
(lambda (parents lst)
(let* ((top (car lst))
(starttag (car top)))
(cond ((not starttag) #f)
((memq (start-tag-name starttag) parents)
(set! begs lst)
#t)
(else (if (finish-begs-upto parents (cdr lst))
(begin (finish-beg top) #t)
#f)))))))
(let loop ()
(let ((tok (tokenizer)))
(if (null? tok)
(finish-all-begs)
(let ((kind (shtml-token-kind tok)))
(cond ((memv kind `(,shtml-comment-symbol
,shtml-decl-symbol
,shtml-entity-symbol
,shtml-pi-symbol
,shtml-text-symbol))
(add-to-current-beg tok))
((eqv? kind shtml-start-symbol)
(let* ((name (start-tag-name tok))
(cell (assq name parent-constraints)))
(and cell (finish-begs-upto (cons 'div (cdr cell))
begs))
(add-to-current-beg tok)
(or (memq name empty-elements)
(set! begs (cons (cons tok '()) begs)))))
((eqv? kind shtml-empty-symbol)
;; Empty tag token, so just add it to current
;; beginning while stripping off leading `*EMPTY*'
;; symbol so that the token becomes normal SXML
;; element syntax.
(add-to-current-beg (cdr tok)))
((eqv? kind shtml-end-symbol)
(let ((name (end-tag-name tok)))
(if name
;; Try to finish to a start tag matching this
;; end tag. If none, just drop the token,
;; though we used to add it to the current
;; beginning.
(finish-begs-to name begs)
;; We have an anonymous end tag, so match it
;; with the most recent beginning. If no
;; beginning to match, then just drop the
;; token, though we used to add it to the
;; current beginning.
(and (car (car begs))
(begin (finish-beg (car begs))
(set! begs (cdr begs)))))))
(else (%htmlprag:error "parse-html/tokenizer"
"unknown tag kind:"
kind)))
(loop))))))))))
;; @defproc %htmlprag:parse-html input normalized? top?
;;
;; This procedure is now used internally by @code{html->shtml} and its
;; variants, and should not be used directly by programs. The interface is
;; likely to change in future versions of HtmlPrag.
(define (%htmlprag:parse-html input normalized? top?)
(let ((parse
(lambda ()
(parse-html/tokenizer
(make-html-tokenizer
(cond ((input-port? input) input)
((string? input) (open-input-string input))
(else (%htmlprag:error
"%htmlprag:parse-html"
"invalid input type:"
input)))
normalized?)
normalized?))))
(if top?
(cons shtml-top-symbol (parse))
(parse))))
;;; @defproc html->sxml-0nf input
;;; @defprocx html->sxml-1nf input
;;; @defprocx html->sxml-2nf input
;;; @defprocx html->sxml input
;;; @defprocx html->shtml input
;;;
;;; Permissively parse HTML from @var{input}, which is either an input port or
;;; a string, and emit an SHTML equivalent or approximation. To borrow and
;;; slightly modify an example from Kiselyov's discussion of his HTML parser:
;;;
;;; @lisp
;;; (html->shtml
;;; "<html><head><title></title><title>whatever</title></head><body>
;;; <a href=\"url\">link</a><p align=center><ul compact style=\"aa\">
;;; <p>BLah<!-- comment <comment> --> <i> italic <b> bold <tt> ened</i>
;;; still &lt; bold </b></body><P> But not done yet...")
;;; @result{}
;;; (*TOP* (html (head (title) (title "whatever"))
;;; (body "\n"
;;; (a (@@ (href "url")) "link")
;;; (p (@@ (align "center"))
;;; (ul (@@ (compact) (style "aa")) "\n"))
;;; (p "BLah"
;;; (*COMMENT* " comment <comment> ")
;;; " "
;;; (i " italic " (b " bold " (tt " ened")))
;;; "\n"
;;; "still < bold "))
;;; (p " But not done yet...")))
;;; @end lisp
;;;
;;; Note that in the emitted SHTML the text token @code{"still < bold"} is
;;; @emph{not} inside the @code{b} element, which represents an unfortunate
;;; failure to emulate all the quirks-handling behavior of some popular Web
;;; browsers.
;;;
;;; The procedures @code{html->sxml-@var{n}nf} for @var{n} 0 through 2
;;; correspond to 0th through 2nd normal forms of SXML as specified in SXML,
;;; and indicate the minimal requirements of the emitted SXML.
;;;
;;; @code{html->sxml} and @code{html->shtml} are currently aliases for
;;; @code{html->sxml-0nf}, and can be used in scripts and interactively, when
;;; terseness is important and any normal form of SXML would suffice.
(define (html->sxml-0nf input) (%htmlprag:parse-html input #f #t))
(define (html->sxml-1nf input) (%htmlprag:parse-html input #f #t))
(define (html->sxml-2nf input) (%htmlprag:parse-html input #t #t))
(define html->sxml html->sxml-0nf)
(define html->shtml html->sxml-0nf)
;;; @section Emitting HTML
;;; Two procedures encoding the SHTML representation as conventional HTML,
;;; @code{write-shtml-as-html} and @code{shtml->html}. These are perhaps most
;;; useful for emitting the result of parsed and transformed input HTML. They
;;; can also be used for emitting HTML from generated or handwritten SHTML.
;;; @defproc write-shtml-as-html shtml [out [foreign-filter]]
;;;
;;; Writes a conventional HTML transliteration of the SHTML @var{shtml} to
;;; output port @var{out}. If @var{out} is not specified, the default is the
;;; current output port. HTML elements of types that are always empty are
;;; written using HTML4-compatible XHTML tag syntax.
;;;
;;; If @var{foreign-filter} is specified, it is a procedure of two argument
;;; that is applied to any non-SHTML (``foreign'') object encountered in
;;; @var{shtml}, and should yield SHTML. The first argument is the object, and
;;; the second argument is a boolean for whether or not the object is part of
;;; an attribute value.
;;;
;;; No inter-tag whitespace or line breaks not explicit in @var{shtml} is
;;; emitted. The @var{shtml} should normally include a newline at the end of
;;; the document. For example:
;;;
;;; @lisp
;;; (write-shtml-as-html
;;; '((html (head (title "My Title"))
;;; (body (@@ (bgcolor "white"))
;;; (h1 "My Heading")
;;; (p "This is a paragraph.")
;;; (p "This is another paragraph.")))))
;;; @print{} <html><head><title>My Title</title></head><body bgcolor="whi
;;; @print{} te"><h1>My Heading</h1><p>This is a paragraph.</p><p>This is
;;; @print{} another paragraph.</p></body></html>
;;; @end lisp
(define (%htmlprag:write-shtml-as-html/fixed shtml out foreign-filter)
(letrec
((write-shtml-text
(lambda (str out)
(let ((len (string-length str)))
(let loop ((i 0))
(if (< i len)
(begin (display (let ((c (string-ref str i)))
(case c
;; ((#\") "&quot;")
((#\&) "&amp;")
((#\<) "&lt;")
((#\>) "&gt;")
(else c)))
out)
(loop (+ 1 i))))))))
(write-dquote-ampified
(lambda (str out)
;; TODO: If we emit "&quot;", we really should parse it, and HTML
;; 4.01 says we should, but anachronisms in HTML create the potential
;; for nasty mutilation of URI in attribute values.
(let ((len (string-length str)))
(let loop ((i 0))
(if (< i len)
(begin (display (let ((c (string-ref str i)))
(if (eqv? c #\") "&quot;" c))
out)
(loop (+ 1 i))))))))
(do-thing
(lambda (thing)
(cond ((string? thing) (write-shtml-text thing out))
((list? thing) (if (not (null? thing))
(do-list-thing thing)))
(else (do-thing (foreign-filter thing #f))))))
(do-list-thing
(lambda (thing)
(let ((head (car thing)))
(cond ((symbol? head)
;; Head is a symbol, so...
(cond ((eq? head shtml-comment-symbol)
;; TODO: Make sure the comment text doesn't contain a
;; comment end sequence.
(display "<!-- " out)
(let ((text (car (cdr thing))))
(if (string? text)
;; TODO: Enforce whitespace safety without
;; padding unnecessarily.
;;
;; (let ((len (string-length text)))
;; (if (= len 0)
;; (display #\space out)
;; (begin (if (not (eqv?
;; (string-ref text 0)
;; #\space))
(display text out)
(%htmlprag:error
"write-shtml-as-html"
"invalid SHTML comment text:"
thing)))
(or (null? (cdr (cdr thing)))
(%htmlprag:error
"write-shtml-as-html"
"invalid SHTML comment body:"
thing))
(display " -->" out))
((eq? head shtml-decl-symbol)
(let ((head (car (cdr thing))))
(display "<!" out)
(display (symbol->string head) out)
(for-each
(lambda (n)
(cond ((symbol? n)
(display #\space out)
(display (symbol->string n) out))
((string? n)
(display " \"" out)
(write-dquote-ampified n out)
(display #\" out))
(else (%htmlprag:error
"write-shtml-as-html"
"invalid SHTML decl:"
thing))))
(cdr (cdr thing)))
(display #\> out)))
((eq? head shtml-pi-symbol)
(display "<?" out)
(display (symbol->string (car (cdr thing))) out)
(display #\space out)
(display (car (cdr (cdr thing))) out)
;; TODO: Error-check that no more rest of PI.
(display "?>" out))
((eq? head shtml-top-symbol)
(for-each do-thing (cdr thing)))
((eq? head shtml-empty-symbol)
#f)
((eq? head '@)
(%htmlprag:error
"write-shtml-as-html"
"illegal position of SHTML attributes:"
thing))
((or (eq? head '&) (eq? head shtml-entity-symbol))
(let ((val (shtml-entity-value thing)))
(if val
(begin (write-char #\& out)
(if (integer? val)
(write-char #\# out))
(display val out)
(write-char #\; out))
(%htmlprag:error
"write-shtml-as-html"
"invalid SHTML entity reference:"
thing))))
((memq head `(,shtml-end-symbol
,shtml-start-symbol
,shtml-text-symbol))
(%htmlprag:error "write-shtml-as-html"
"invalid SHTML symbol:"
head))
(else
(display #\< out)
(display head out)
(let* ((rest (cdr thing)))
(if (not (null? rest))
(let ((second (car rest)))
(and (list? second)
(not (null? second))
(eq? (car second)
'@)
(begin (for-each do-attr (cdr second))
(set! rest (cdr rest))))))
(if (memq head
%htmlprag:empty-elements)
;; TODO: Error-check to make sure the element
;; has no content other than attributes. We
;; have to test for cases like: (br (@) ()
;; (()))
(display " />" out)
(begin (display #\> out)
(for-each do-thing rest)
(display "</" out)
(display (symbol->string head) out)
(display #\> out)))))))
;; ((or (list? head) (string? head))
;;
;; Head is a list or string, which might occur as the result
;; of an SXML transform, so we'll cope.
(else
;; Head is not a symbol, which might occur as the result of
;; an SXML transform, so we'll cope.
(for-each do-thing thing))
;;(else
;; ;; Head is NOT a symbol, list, or string, so error.
;; (%htmlprag:error "write-shtml-as-html"
;; "invalid SHTML list:"
;; thing))
))))
(write-attr-val-dquoted
(lambda (str out)
(display #\" out)
(display str out)
(display #\" out)))
(write-attr-val-squoted
(lambda (str out)
(display #\' out)
(display str out)
(display #\' out)))
(write-attr-val-dquoted-and-amped
(lambda (str out)
(display #\" out)
(write-dquote-ampified str out)
(display #\" out)))
(write-attr-val
(lambda (str out)
(let ((len (string-length str)))
(let find-dquote-and-squote ((i 0))
(if (= i len)
(write-attr-val-dquoted str out)
(let ((c (string-ref str i)))
(cond ((eqv? c #\")
(let find-squote ((i (+ 1 i)))
(if (= i len)
(write-attr-val-squoted str out)
(if (eqv? (string-ref str i) #\')
(write-attr-val-dquoted-and-amped str
out)
(find-squote (+ 1 i))))))
((eqv? c #\')
(let find-dquote ((i (+ 1 i)))
(if (= i len)
(write-attr-val-dquoted str out)
(if (eqv? (string-ref str i) #\")
(write-attr-val-dquoted-and-amped str
out)
(find-dquote (+ 1 i))))))
(else (find-dquote-and-squote (+ 1 i))))))))))
(collect-and-write-attr-val
;; TODO: Take another look at this.
(lambda (lst out)
(let ((os #f))
(let do-list ((lst lst))
(for-each
(lambda (thing)
(let do-thing ((thing thing))
(cond ((string? thing)
(or os (set! os (open-output-string)))
(display thing os))
((list? thing)
(do-list thing))
((eq? thing #t)
#f)
(else
(do-thing (foreign-filter thing #t))))))
lst))
(if os
(begin
(display #\= out)
(write-attr-val (%htmlprag:gosc os) out))))))
(do-attr
(lambda (attr)
(or (list? attr)
(%htmlprag:error "write-shtml-as-html"
"invalid SHTML attribute:"
attr))
(if (not (null? attr))
(let ((name (car attr)))
(or (symbol? name)
(%htmlprag:error
"write-shtml-as-html"
"invalid name in SHTML attribute:"
attr))
(if (not (eq? name '@))
(begin
(display #\space out)
(display name out)
(collect-and-write-attr-val (cdr attr) out)
)))))))
(do-thing shtml)
(if #f #f)))
(define write-shtml-as-html
(letrec ((error-foreign-filter
(lambda (foreign-object in-attribute-value?)
(%htmlprag:error
"write-shtml-as-html"
(if in-attribute-value?
"unhandled foreign object in shtml attribute value:"
"unhandled foreign object in shtml:")
foreign-object))))
(lambda (shtml . rest)
(case (length rest)
((0) (%htmlprag:write-shtml-as-html/fixed
shtml
(current-output-port)
error-foreign-filter))
((1) (%htmlprag:write-shtml-as-html/fixed
shtml
(car rest)
error-foreign-filter))
((2) (%htmlprag:write-shtml-as-html/fixed
shtml
(car rest)
(cadr rest)))
(else
(%htmlprag:error "write-shtml-as-html"
"extraneous arguments:"
(cddr rest)))))))
;;; @defproc shtml->html shtml
;;;
;;; Yields an HTML encoding of SHTML @var{shtml} as a string. For example:
;;;
;;; @lisp
;;; (shtml->html
;;; (html->shtml
;;; "<P>This is<br<b<I>bold </foo>italic</ b > text.</p>"))
;;; @result{} "<p>This is<br /><b><i>bold italic</i></b> text.</p>"
;;; @end lisp
;;;
;;; Note that, since this procedure constructs a string, it should normally
;;; only be used when the HTML is relatively small. When encoding HTML
;;; documents of conventional size and larger, @code{write-shtml-as-html} is
;;; much more efficient.
(define (shtml->html shtml)
(let ((os (open-output-string)))
(write-shtml-as-html shtml os)
(%htmlprag:gosc os)))
;;; @section Tests
;;; The HtmlPrag test suite can be enabled by editing the source code file and
;;; loading @uref{http://www.neilvandyke.org/testeez/, Testeez}.
(define (%htmlprag:test)
(%htmlprag:testeez
"HtmlPrag"
(test-define "" lf (string (%htmlprag:a2c 10)))
(test/equal "" (html->shtml "<a>>") `(,shtml-top-symbol (a ">")))
(test/equal "" (html->shtml "<a<>") `(,shtml-top-symbol (a "<" ">")))
(test/equal "" (html->shtml "<>") `(,shtml-top-symbol "<" ">"))
(test/equal "" (html->shtml "< >") `(,shtml-top-symbol "<" ">"))
(test/equal "" (html->shtml "< a>") `(,shtml-top-symbol (a)))
(test/equal "" (html->shtml "< a / >") `(,shtml-top-symbol (a)))
(test/equal "" (html->shtml "<a<") `(,shtml-top-symbol (a "<")))
(test/equal "" (html->shtml "<a<b") `(,shtml-top-symbol (a (b))))
(test/equal "" (html->shtml "><a>") `(,shtml-top-symbol ">" (a)))
(test/equal "" (html->shtml "</>") `(,shtml-top-symbol))
(test/equal "" (html->shtml "<\">") `(,shtml-top-symbol "<" "\"" ">"))
(test/equal ""
(html->shtml (string-append "<a>xxx<plaintext>aaa" lf
"bbb" lf
"c<c<c"))
`(,shtml-top-symbol
(a "xxx" (plaintext ,(string-append "aaa" lf)
,(string-append "bbb" lf)
"c<c<c"))))
(test/equal ""
(html->shtml "aaa<!-- xxx -->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol " xxx ") "bbb"))
(test/equal ""
(html->shtml "aaa<! -- xxx -->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol " xxx ") "bbb"))
(test/equal ""
(html->shtml "aaa<!-- xxx --->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol " xxx -") "bbb"))
(test/equal ""
(html->shtml "aaa<!-- xxx ---->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol " xxx --") "bbb"))
(test/equal ""
(html->shtml "aaa<!-- xxx -y-->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol " xxx -y") "bbb"))
(test/equal ""
(html->shtml "aaa<!----->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol "-") "bbb"))
(test/equal ""
(html->shtml "aaa<!---->bbb")
`(,shtml-top-symbol
"aaa" (,shtml-comment-symbol "") "bbb"))
(test/equal ""
(html->shtml "aaa<!--->bbb")
`(,shtml-top-symbol "aaa" (,shtml-comment-symbol "->bbb")))
(test/equal "" (html->shtml "<hr>") `(,shtml-top-symbol (hr)))
(test/equal "" (html->shtml "<hr/>") `(,shtml-top-symbol (hr)))
(test/equal "" (html->shtml "<hr />") `(,shtml-top-symbol (hr)))
(test/equal ""
(html->shtml "<hr noshade>")
`(,shtml-top-symbol (hr (@ (noshade)))))
(test/equal ""
(html->shtml "<hr noshade/>")
`(,shtml-top-symbol (hr (@ (noshade)))))
(test/equal ""
(html->shtml "<hr noshade />")
`(,shtml-top-symbol (hr (@ (noshade)))))
(test/equal ""
(html->shtml "<hr noshade / >")
`(,shtml-top-symbol (hr (@ (noshade)))))
(test/equal ""
(html->shtml "<hr noshade=1 />")
`(,shtml-top-symbol (hr (@ (noshade "1")))))
(test/equal ""
(html->shtml "<hr noshade=1/>")
`(,shtml-top-symbol (hr (@ (noshade "1/")))))
(test/equal ""
(html->shtml "<q>aaa<p/>bbb</q>ccc</p>ddd")
`(,shtml-top-symbol (q "aaa" (p) "bbb") "ccc" "ddd"))
(test/equal "" (html->shtml "&lt;") `(,shtml-top-symbol "<"))
(test/equal "" (html->shtml "&gt;") `(,shtml-top-symbol ">"))
(test/equal ""
(html->shtml "Gilbert &amp; Sullivan")
`(,shtml-top-symbol "Gilbert & Sullivan"))
(test/equal ""
(html->shtml "Gilbert &amp Sullivan")
`(,shtml-top-symbol "Gilbert & Sullivan"))
(test/equal ""
(html->shtml "Gilbert & Sullivan")
`(,shtml-top-symbol "Gilbert & Sullivan"))
(test/equal ""
(html->shtml "Copyright &copy; Foo")
`(,shtml-top-symbol "Copyright "
(& ,(string->symbol "copy"))
" Foo"))
(test/equal ""
(html->shtml "aaa&copy;bbb")
`(,shtml-top-symbol
"aaa" (& ,(string->symbol "copy")) "bbb"))
(test/equal ""
(html->shtml "aaa&copy")
`(,shtml-top-symbol
"aaa" (& ,(string->symbol "copy"))))
(test/equal "" (html->shtml "&#42;") `(,shtml-top-symbol "*"))
(test/equal "" (html->shtml "&#42") `(,shtml-top-symbol "*"))
(test/equal "" (html->shtml "&#42x") `(,shtml-top-symbol "*x"))
(test/equal "" (html->shtml "&#151") `(,shtml-top-symbol
(& 151)
;; ,(string (%htmlprag:a2c 151))
))
(test/equal "" (html->shtml "&#1000") `(,shtml-top-symbol (& 1000)))
(test/equal "" (html->shtml "&#x42") `(,shtml-top-symbol "B"))
(test/equal "" (html->shtml "&#xA2") `(,shtml-top-symbol
(& 162)
;; ,(string (%htmlprag:a2c 162))
))
(test/equal "" (html->shtml "&#xFF") `(,shtml-top-symbol
(& 255)
;; ,(string (%htmlprag:a2c 255))
))
(test/equal "" (html->shtml "&#x100") `(,shtml-top-symbol (& 256)))
(test/equal "" (html->shtml "&#X42") `(,shtml-top-symbol "B"))
(test/equal "" (html->shtml "&42;") `(,shtml-top-symbol "&42;"))
(test/equal ""
(html->shtml (string-append "aaa&copy;bbb&amp;ccc&lt;ddd&&gt;"
"eee&#42;fff&#1000;ggg&#x5a;hhh"))
`(,shtml-top-symbol
"aaa"
(& ,(string->symbol "copy"))
"bbb&ccc<ddd&>eee*fff"
(& 1000)
"gggZhhh"))
(test/equal ""
(html->shtml
(string-append
"<IMG src=\"http://e.e/aw/pics/listings/"
"ebayLogo_38x16.gif\" border=0 width=\"38\" height=\"16\" "
"HSPACE=5 VSPACE=0\">2</FONT>"))
`(,shtml-top-symbol
(img (@
(src
"http://e.e/aw/pics/listings/ebayLogo_38x16.gif")
(border "0") (width "38") (height "16")
(hspace "5") (vspace "0")))
"2"))
(test/equal ""
(html->shtml "<aaa bbb=ccc\"ddd>eee")
`(,shtml-top-symbol (aaa (@ (bbb "ccc") (ddd)) "eee")))
(test/equal ""
(html->shtml "<aaa bbb=ccc \"ddd>eee")
`(,shtml-top-symbol (aaa (@ (bbb "ccc") (ddd)) "eee")))
(test/equal ""
(html->shtml
(string-append
"<HTML><Head><Title>My Title</Title></Head>"
"<Body BGColor=\"white\" Foo=42>"
"This is a <B><I>bold-italic</B></I> test of </Erk>"
"broken HTML.<br>Yes it is.</Body></HTML>"))
`(,shtml-top-symbol
(html (head (title "My Title"))
(body (@ (bgcolor "white") (foo "42"))
"This is a "
(b (i "bold-italic"))
" test of "
"broken HTML."
(br)
"Yes it is."))))
(test/equal ""
(html->shtml
(string-append
"<!DOCTYPE html PUBLIC \"-//W3C//DTD XHTML 1.0 Strict//EN\""
" \"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd\">"))
`(,shtml-top-symbol
(,shtml-decl-symbol
,(string->symbol "DOCTYPE")
html
,(string->symbol "PUBLIC")
"-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd")))
(test/equal ""
(html->shtml
(string-append
"<html xmlns=\"http://www.w3.org/1999/xhtml\" "
"xml:lang=\"en\" "
"lang=\"en\">"))
`(,shtml-top-symbol
(html (@ (xmlns "http://www.w3.org/1999/xhtml")
(xml:lang "en") (lang "en")))))
(test/equal
""
(html->shtml
(string-append
"<html:html xmlns:html=\"http://www.w3.org/TR/REC-html40\">"
"<html:head><html:title>Frobnostication</html:title></html:head>"
"<html:body><html:p>Moved to <html:a href=\"http://frob.com\">"
"here.</html:a></html:p></html:body></html:html>"))
`(,shtml-top-symbol
(html (@ (xmlns:html "http://www.w3.org/TR/REC-html40"))
(head (title "Frobnostication"))
(body (p "Moved to "
(a (@ (href "http://frob.com"))
"here."))))))
(test/equal ""
(html->shtml
(string-append
"<RESERVATION xmlns:HTML=\"http://www.w3.org/TR/REC-html40\">"
"<NAME HTML:CLASS=\"largeSansSerif\">Layman, A</NAME>"
"<SEAT CLASS=\"Y\" HTML:CLASS=\"largeMonotype\">33B</SEAT>"
"<HTML:A HREF=\"/cgi-bin/ResStatus\">Check Status</HTML:A>"
"<DEPARTURE>1997-05-24T07:55:00+1</DEPARTURE></RESERVATION>"))
`(,shtml-top-symbol
(reservation (@ (,(string->symbol "xmlns:HTML")
"http://www.w3.org/TR/REC-html40"))
(name (@ (class "largeSansSerif"))
"Layman, A")
(seat (@ (class "Y") (class "largeMonotype"))
"33B")
(a (@ (href "/cgi-bin/ResStatus"))
"Check Status")
(departure "1997-05-24T07:55:00+1"))))
(test/equal
""
(html->shtml
(string-append
"<html><head><title></title><title>whatever</title></head><body>"
"<a href=\"url\">link</a><p align=center><ul compact style=\"aa\">"
"<p>BLah<!-- comment <comment> --> <i> italic <b> bold <tt> ened </i>"
" still &lt; bold </b></body><P> But not done yet..."))
`(,shtml-top-symbol
(html (head (title) (title "whatever"))
(body (a (@ (href "url")) "link")
(p (@ (align "center"))
(ul (@ (compact) (style "aa"))))
(p "BLah"
(,shtml-comment-symbol " comment <comment> ")
" "
(i " italic " (b " bold " (tt " ened ")))
" still < bold "))
(p " But not done yet..."))))
(test/equal ""
(html->shtml "<?xml version=\"1.0\" encoding=\"UTF-8\"?>")
`(,shtml-top-symbol
(,shtml-pi-symbol xml "version=\"1.0\" encoding=\"UTF-8\"")))
(test/equal ""
(html->shtml "<?php php_info(); ?>")
`(,shtml-top-symbol (,shtml-pi-symbol php "php_info(); ")))
(test/equal ""
(html->shtml "<?php php_info(); ?")
`(,shtml-top-symbol (,shtml-pi-symbol php "php_info(); ?")))
(test/equal ""
(html->shtml "<?php php_info(); ")
`(,shtml-top-symbol (,shtml-pi-symbol php "php_info(); ")))
(test/equal ""
(html->shtml "<?foo bar ? baz > blort ?>")
`(,shtml-top-symbol
(,shtml-pi-symbol foo "bar ? baz > blort ")))
(test/equal ""
(html->shtml "<?foo b?>x")
`(,shtml-top-symbol (,shtml-pi-symbol foo "b") "x"))
(test/equal ""
(html->shtml "<?foo ?>x")
`(,shtml-top-symbol (,shtml-pi-symbol foo "") "x"))
(test/equal ""
(html->shtml "<?foo ?>x")
`(,shtml-top-symbol (,shtml-pi-symbol foo "") "x"))
(test/equal ""
(html->shtml "<?foo?>x")
`(,shtml-top-symbol (,shtml-pi-symbol foo "") "x"))
(test/equal ""
(html->shtml "<?f?>x")
`(,shtml-top-symbol (,shtml-pi-symbol f "") "x"))
(test/equal ""
(html->shtml "<??>x")
`(,shtml-top-symbol (,shtml-pi-symbol #f "") "x"))
(test/equal ""
(html->shtml "<?>x")
`(,shtml-top-symbol (,shtml-pi-symbol #f ">x")))
(test/equal ""
(html->shtml "<foo bar=\"baz\">blort")
`(,shtml-top-symbol (foo (@ (bar "baz")) "blort")))
(test/equal ""
(html->shtml "<foo bar='baz'>blort")
`(,shtml-top-symbol (foo (@ (bar "baz")) "blort")))
(test/equal ""
(html->shtml "<foo bar=\"baz'>blort")
`(,shtml-top-symbol (foo (@ (bar "baz'>blort")))))
(test/equal ""
(html->shtml "<foo bar='baz\">blort")
`(,shtml-top-symbol (foo (@ (bar "baz\">blort")))))
(test/equal ""
(html->shtml (string-append "<p>A</p>"
"<script>line0 <" lf
"line1" lf
"<line2></script>"
"<p>B</p>"))
`(,shtml-top-symbol (p "A")
(script ,(string-append "line0 <" lf)
,(string-append "line1" lf)
"<line2>")
(p "B")))
(test/equal ""
(html->shtml "<xmp>a<b>c</XMP>d")
`(,shtml-top-symbol (xmp "a<b>c") "d"))
(test/equal ""
(html->shtml "<XMP>a<b>c</xmp>d")
`(,shtml-top-symbol (xmp "a<b>c") "d"))
(test/equal ""
(html->shtml "<xmp>a<b>c</foo:xmp>d")
`(,shtml-top-symbol (xmp "a<b>c") "d"))
(test/equal ""
(html->shtml "<foo:xmp>a<b>c</xmp>d")
`(,shtml-top-symbol (xmp "a<b>c") "d"))
(test/equal ""
(html->shtml "<foo:xmp>a<b>c</foo:xmp>d")
`(,shtml-top-symbol (xmp "a<b>c") "d"))
(test/equal ""
(html->shtml "<foo:xmp>a<b>c</bar:xmp>d")
`(,shtml-top-symbol (xmp "a<b>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b>c</xmp>d")
`(,shtml-top-symbol (xmp "a</b>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b >c</xmp>d")
`(,shtml-top-symbol (xmp "a</b >c") "d"))
(test/equal ""
(html->shtml "<xmp>a</ b>c</xmp>d")
`(,shtml-top-symbol (xmp "a</ b>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</ b >c</xmp>d")
`(,shtml-top-symbol (xmp "a</ b >c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b:x>c</xmp>d")
`(,shtml-top-symbol (xmp "a</b:x>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b::x>c</xmp>d")
`(,shtml-top-symbol (xmp "a</b::x>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b:::x>c</xmp>d")
`(,shtml-top-symbol (xmp "a</b:::x>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b:>c</xmp>d")
`(,shtml-top-symbol (xmp "a</b:>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</b::>c</xmp>d")
`(,shtml-top-symbol (xmp "a</b::>c") "d"))
(test/equal ""
(html->shtml "<xmp>a</xmp:b>c</xmp>d")
`(,shtml-top-symbol (xmp "a</xmp:b>c") "d"))
(test-define "expected output for next two tests"
expected
`(,shtml-top-symbol (p "real1")
,lf
(xmp ,lf
,(string-append "alpha" lf)
,(string-append "<P>fake</P>" lf)
,(string-append "bravo" lf))
(p "real2")))
(test/equal ""
(html->shtml (string-append "<P>real1</P>" lf
"<XMP>" lf
"alpha" lf
"<P>fake</P>" lf
"bravo" lf
"</XMP " lf
"<P>real2</P>"))
expected)
(test/equal ""
(html->shtml (string-append "<P>real1</P>" lf
"<XMP>" lf
"alpha" lf
"<P>fake</P>" lf
"bravo" lf
"</XMP" lf
"<P>real2</P>"))
expected)
(test/equal ""
(html->shtml "<xmp>a</xmp>x")
`(,shtml-top-symbol (xmp "a") "x"))
(test/equal ""
(html->shtml (string-append "<xmp>a" lf "</xmp>x"))
`(,shtml-top-symbol (xmp ,(string-append "a" lf)) "x"))
(test/equal ""
(html->shtml "<xmp></xmp>x")
`(,shtml-top-symbol (xmp) "x"))
(test/equal ""
(html->shtml "<xmp>a</xmp") `(,shtml-top-symbol (xmp "a")))
(test/equal ""
(html->shtml "<xmp>a</xm") `(,shtml-top-symbol (xmp "a</xm")))
(test/equal ""
(html->shtml "<xmp>a</x") `(,shtml-top-symbol (xmp "a</x")))
(test/equal ""
(html->shtml "<xmp>a</") `(,shtml-top-symbol (xmp "a</")))
(test/equal ""
(html->shtml "<xmp>a<") `(,shtml-top-symbol (xmp "a<")))
(test/equal ""
(html->shtml "<xmp>a") `(,shtml-top-symbol (xmp "a")))
(test/equal ""
(html->shtml "<xmp>") `(,shtml-top-symbol (xmp)))
(test/equal ""
(html->shtml "<xmp") `(,shtml-top-symbol (xmp)))
(test/equal ""
(html->shtml "<xmp x=42 ")
`(,shtml-top-symbol (xmp (@ (x "42")))))
(test/equal ""
(html->shtml "<xmp x= ") `(,shtml-top-symbol (xmp (@ (x)))))
(test/equal ""
(html->shtml "<xmp x ") `(,shtml-top-symbol (xmp (@ (x)))))
(test/equal ""
(html->shtml "<xmp x") `(,shtml-top-symbol (xmp (@ (x)))))
(test/equal ""
(html->shtml "<script>xxx")
`(,shtml-top-symbol (script "xxx")))
(test/equal ""
(html->shtml "<script/>xxx")
`(,shtml-top-symbol (script) "xxx"))
(test/equal ""
(html->shtml "<html xml:lang=\"en\" lang=\"en\">")
`(,shtml-top-symbol (html (@ (xml:lang "en") (lang "en")))))
(test/equal ""
(html->shtml "<a href=/foo.html>")
`(,shtml-top-symbol (a (@ (href "/foo.html")))))
(test/equal ""
(html->shtml "<a href=/>foo.html")
`(,shtml-top-symbol (a (@ (href "/")) "foo.html")))
;; TODO: Add verbatim-pair cases with attributes in the end tag.
(test/equal ""
(shtml->html '(p)) "<p></p>")
(test/equal ""
(shtml->html '(p "CONTENT")) "<p>CONTENT</p>")
(test/equal ""
(shtml->html '(br)) "<br />")
(test/equal ""
(shtml->html '(br "CONTENT")) "<br />")
(test/equal ""
(shtml->html `(hr (@ (clear "all"))))
"<hr clear=\"all\" />")
(test/equal ""
(shtml->html `(hr (@ (noshade))))
"<hr noshade />")
(test/equal ""
(shtml->html `(hr (@ (noshade #t))))
"<hr noshade />") ;; TODO: Maybe lose this test.
(test/equal ""
(shtml->html `(hr (@ (noshade "noshade"))))
"<hr noshade=\"noshade\" />")
(test/equal ""
(shtml->html `(hr (@ (aaa "bbbccc"))))
"<hr aaa=\"bbbccc\" />")
(test/equal ""
(shtml->html `(hr (@ (aaa "bbb'ccc"))))
"<hr aaa=\"bbb'ccc\" />")
(test/equal ""
(shtml->html `(hr (@ (aaa "bbb\"ccc"))))
"<hr aaa='bbb\"ccc' />")
(test/equal ""
(shtml->html `(hr (@ (aaa "bbb\"ccc'ddd"))))
"<hr aaa=\"bbb&quot;ccc'ddd\" />")
(test/equal "" (shtml->html '(& "copy")) "&copy;")
(test/equal "" (shtml->html '(& "rArr")) "&rArr;")
(test/equal "" (shtml->html `(& ,(string->symbol "rArr"))) "&rArr;")
(test/equal "" (shtml->html '(& 151)) "&#151;")
(test/equal ""
(html->shtml "&copy;")
`(,shtml-top-symbol (& ,(string->symbol "copy"))))
(test/equal ""
(html->shtml "&rArr;")
`(,shtml-top-symbol (& ,(string->symbol "rArr"))))
(test/equal ""
(html->shtml "&#151;")
`(,shtml-top-symbol
(& 151)
;; ,(string (%htmlprag:a2c 151))
))
(test/equal ""
(html->shtml "&#999;")
`(,shtml-top-symbol (& 999)))
(test/equal ""
(shtml->html
`(,shtml-pi-symbol xml "version=\"1.0\" encoding=\"UTF-8\""))
"<?xml version=\"1.0\" encoding=\"UTF-8\"?>")
(test/equal ""
(shtml->html
`(,shtml-decl-symbol
,(string->symbol "DOCTYPE")
html
,(string->symbol "PUBLIC")
"-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"))
(string-append
"<!DOCTYPE html PUBLIC \"-//W3C//DTD XHTML 1.0 Strict//EN\""
" \"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd\">"))
(test/equal ""
(shtml-entity-value '(*ENTITY* "shtml-named-char" "rArr"))
(string->symbol "rArr"))
(test/equal ""
(shtml-entity-value '(& "rArr"))
(string->symbol "rArr"))
(test/equal ""
(shtml-entity-value `(& ,(string->symbol "rArr")))
(string->symbol "rArr"))
(test/equal ""
(html->shtml "xxx<![CDATA[abc]]>yyy")
`(,shtml-top-symbol "xxx" "abc" "yyy"))
(test/equal ""
(html->shtml "xxx<![CDATA[ab]c]]>yyy")
`(,shtml-top-symbol "xxx" "ab]c" "yyy"))
(test/equal ""
(html->shtml "xxx<![CDATA[ab]]c]]>yyy")
`(,shtml-top-symbol "xxx" "ab]]c" "yyy"))
(test/equal ""
(html->shtml "xxx<![CDATA[]]]>yyy")
`(,shtml-top-symbol "xxx" "]" "yyy"))
(test/equal ""
(html->shtml "xxx<![CDATAyyy")
`(,shtml-top-symbol "xxx" "<![CDATA" "yyy"))
(test/equal "parent constraints with div"
(html->shtml "<html><div><p>P1</p><p>P2</p></div><p>P3</p>")
`(,shtml-top-symbol (html (div (p "P1")
(p "P2"))
(p "P3"))))
(test/equal "we no longer convert character references above 126 to string"
(html->shtml "&#151;")
`(,shtml-top-symbol (& 151)))
;; TODO: Write more test cases for HTML encoding.
;; TODO: Write test cases for foreign-filter of HTML writing.
;; TODO: Write test cases for attribute values that aren't simple strings.
;; TODO: Document this.
;;
;; (define html-1 "<myelem myattr=\"&\">")
;; (define shtml (html->shtml html-1))
;; shtml
;; (define html-2 (shtml->html shtml))
;; html-2
))
;;; @unnumberedsec History
;;; @table @asis
;;;
;;; @item Version 0.16 --- 2005-12-18
;;; Documentation fix.
;;;
;;; @item Version 0.15 --- 2005-12-18
;;; In the HTML parent element constraints that are used for structure
;;; recovery, @code{div} is now always permitted as a parent, as a stopgap
;;; measure until substantial time can be spent reworking the algorithm to
;;; better support @code{div} (bug reported by Corey Sweeney and Jepri). Also
;;; no longer convert to Scheme character any HTML numeric character reference
;;; with value above 126, to avoid Unicode problem with PLT 299/300 (bug
;;; reported by Corey Sweeney).
;;;
;;; @item Version 0.14 --- 2005-06-16
;;; XML CDATA sections are now tokenized. Thanks to Alejandro Forero Cuervo
;;; for suggesting this feature. The deprecated procedures @code{sxml->html}
;;; and @code{write-sxml-html} have been removed. Minor documentation changes.
;;;
;;; @item Version 0.13 --- 2005-02-23
;;; HtmlPrag now requires @code{syntax-rules}, and a reader that can read
;;; @code{@@} as a symbol. SHTML now has a special @code{&} element for
;;; character entities, and it is emitted by the parser rather than the old
;;; @code{*ENTITY*} kludge. @code{shtml-entity-value} supports both the new
;;; and the old character entity representations. @code{shtml-entity-value}
;;; now yields @code{#f} on invalid SHTML entity, rather than raising an error.
;;; @code{write-shtml-as-html} now has a third argument, @code{foreign-filter}.
;;; @code{write-shtml-as-html} now emits SHTML @code{&} entity references.
;;; Changed @code{shtml-named-char-id} and @code{shtml-numeric-char-id}, as
;;; previously warned. Testeez is now used for the test suite. Test procedure
;;; is now the internal @code{%htmlprag:test}. Documentation changes.
;;; Notably, much documentation about using HtmlPrag under various particular
;;; Scheme implementations has been removed.
;;;
;;; @item Version 0.12 --- 2004-07-12
;;; Forward-slash in an unquoted attribute value is now considered a value
;;; constituent rather than an unconsumed terminator of the value (thanks to
;;; Maurice Davis for reporting and a suggested fix). @code{xml:} is now
;;; preserved as a namespace qualifier (thanks to Peter Barabas for
;;; reporting). Output port term of @code{write-shtml-as-html} is now
;;; optional. Began documenting loading for particular implementation-specific
;;; packagings.
;;;
;;; @item Version 0.11 --- 2004-05-13
;;; To reduce likely namespace collisions with SXML tools, and in anticipation
;;; of a forthcoming set of new features, introduced the concept of ``SHTML,''
;;; which will be elaborated upon in a future version of HtmlPrag. Renamed
;;; @code{sxml-@var{x}-symbol} to @code{shtml-@var{x}-symbol},
;;; @code{sxml-html-@var{x}} to @code{shtml-@var{x}}, and
;;; @code{sxml-token-kind} to @code{shtml-token-kind}. @code{html->shtml},
;;; @code{shtml->html}, and @code{write-shtml-as-html} have been added as
;;; names. Considered deprecated but still defined (see the ``Deprecated''
;;; section of this documentation) are @code{sxml->html} and
;;; @code{write-sxml-html}. The growing pains should now be all but over.
;;; Internally, @code{htmlprag-internal:error} introduced for Bigloo
;;; portability. SISC returned to the test list; thanks to Scott G. Miller
;;; for his help. Fixed a new character @code{eq?} bug, thanks to SISC.
;;;
;;; @item Version 0.10 --- 2004-05-11
;;; All public identifiers have been renamed to drop the ``@code{htmlprag:}''
;;; prefix. The portability identifiers have been renamed to begin with an
;;; @code{htmlprag-internal:} prefix, are now considered strictly
;;; internal-use-only, and have otherwise been changed. @code{parse-html} and
;;; @code{always-empty-html-elements} are no longer public.
;;; @code{test-htmlprag} now tests @code{html->sxml} rather than
;;; @code{parse-html}. SISC temporarily removed from the test list, until an
;;; open source Java that works correctly is found.
;;;
;;; @item Version 0.9 --- 2004-05-07
;;; HTML encoding procedures added. Added
;;; @code{htmlprag:sxml-html-entity-value}. Upper-case @code{X} in hexadecimal
;;; character entities is now parsed, in addition to lower-case @code{x}.
;;; Added @code{htmlprag:always-empty-html-elements}. Added additional
;;; portability bindings. Added more test cases.
;;;
;;; @item Version 0.8 --- 2004-04-27
;;; Entity references (symbolic, decimal numeric, hexadecimal numeric) are now
;;; parsed into @code{*ENTITY*} SXML. SXML symbols like @code{*TOP*} are now
;;; always upper-case, regardless of the Scheme implementation. Identifiers
;;; such as @code{htmlprag:sxml-top-symbol} are bound to the upper-case
;;; symbols. Procedures @code{htmlprag:html->sxml-0nf},
;;; @code{htmlprag:html->sxml-1nf}, and @code{htmlprag:html->sxml-2nf} have
;;; been added. @code{htmlprag:html->sxml} now an alias for
;;; @code{htmlprag:html->sxml-0nf}. @code{htmlprag:parse} has been refashioned
;;; as @code{htmlprag:parse-html} and should no longer be directly. A number
;;; of identifiers have been renamed to be more appropriate when the
;;; @code{htmlprag:} prefix is dropped in some implementation-specific
;;; packagings of HtmlPrag: @code{htmlprag:make-tokenizer} to
;;; @code{htmlprag:make-html-tokenizer}, @code{htmlprag:parse/tokenizer} to
;;; @code{htmlprag:parse-html/tokenizer}, @code{htmlprag:html->token-list} to
;;; @code{htmlprag:tokenize-html}, @code{htmlprag:token-kind} to
;;; @code{htmlprag:sxml-token-kind}, and @code{htmlprag:test} to
;;; @code{htmlprag:test-htmlprag}. Verbatim elements with empty-element tag
;;; syntax are handled correctly. New versions of Bigloo and RScheme tested.
;;;
;;; @item Version 0.7 --- 2004-03-10
;;; Verbatim pair elements like @code{script} and @code{xmp} are now parsed
;;; correctly. Two Scheme implementations have temporarily been dropped from
;;; regression testing: Kawa, due to a Java bytecode verifier error likely due
;;; to a Java installation problem on the test machine; and SXM 1.1, due to
;;; hitting a limit on the number of literals late in the test suite code.
;;; Tested newer versions of Bigloo, Chicken, Gauche, Guile, MIT Scheme, PLT
;;; MzScheme, RScheme, SISC, and STklos. RScheme no longer requires the
;;; ``@code{(define get-output-string close-output-port)}'' workaround.
;;;
;;; @item Version 0.6 --- 2003-07-03
;;; Fixed uses of @code{eq?} in character comparisons, thanks to Scott G.
;;; Miller. Added @code{htmlprag:html->normalized-sxml} and
;;; @code{htmlprag:html->nonnormalized-sxml}. Started to add
;;; @code{close-output-port} to uses of output strings, then reverted due to
;;; bug in one of the supported dialects. Tested newer versions of Bigloo,
;;; Gauche, PLT MzScheme, RScheme.
;;;
;;; @item Version 0.5 --- 2003-02-26
;;; Removed uses of @code{call-with-values}. Re-ordered top-level definitions,
;;; for portability. Now tests under Kawa 1.6.99, RScheme 0.7.3.2, Scheme 48
;;; 0.57, SISC 1.7.4, STklos 0.54, and SXM 1.1.
;;;
;;; @item Version 0.4 --- 2003-02-19
;;; Apostrophe-quoted element attribute values are now handled. A bug that
;;; incorrectly assumed left-to-right term evaluation order has been fixed
;;; (thanks to MIT Scheme for confronting us with this). Now also tests OK
;;; under Gauche 0.6.6 and MIT Scheme 7.7.1. Portability improvement for
;;; implementations (e.g., RScheme 0.7.3.2.b6, Stalin 0.9) that cannot read
;;; @code{@@} as a symbol (although those implementations tend to present other
;;; portability issues, as yet unresolved).
;;;
;;; @item Version 0.3 --- 2003-02-05
;;; A test suite with 66 cases has been added, and necessary changes have been
;;; made for the suite to pass on five popular Scheme implementations. XML
;;; processing instructions are now parsed. Parent constraints have been added
;;; for @code{colgroup}, @code{tbody}, and @code{thead} elements. Erroneous
;;; input, including invalid hexadecimal entity reference syntax and extraneous
;;; double quotes in element tags, is now parsed better.
;;; @code{htmlprag:token-kind} emits symbols more consistent with SXML.
;;;
;;; @item Version 0.2 --- 2003-02-02
;;; Portability improvements.
;;;
;;; @item Version 0.1 --- 2003-01-31
;;; Dusted off old Guile-specific code from April 2001, converted to emit SXML,
;;; mostly ported to R5RS and SRFI-6, added some XHTML support and
;;; documentation. A little preliminary testing has been done, and the package
;;; is already useful for some applications, but this release should be
;;; considered a preview to invite comments.
;;;
;;; @end table
(provide (all-defined)))