1 /* HTML parser for Wget.
2 Copyright (C) 1998, 2000, 2003 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or (at
9 your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
30 /* The only entry point to this module is map_html_tags(), which see. */
34 - Allow hooks for callers to process contents outside tags. This
35 is needed to implement handling <style> and <script>. The
36 taginfo structure already carries the information about where the
37 tags are, but this is not enough, because one would also want to
38 skip the comments. (The funny thing is that for <style> and
39 <script> you *don't* want to skip comments!)
41 - Create a test suite for regression testing. */
45 This is the third HTML parser written for Wget. The first one was
46 written some time during the Geturl 1.0 beta cycle, and was very
47 inefficient and buggy. It also contained some very complex code to
48 remember a list of parser states, because it was supposed to be
51 The second HTML parser was written for Wget 1.4 (the first version
52 by the name `Wget'), and was a complete rewrite. Although the new
53 parser behaved much better and made no claims of reentrancy, it
54 still shared many of the fundamental flaws of the old version -- it
55 only regarded HTML in terms tag-attribute pairs, where the
56 attribute's value was a URL to be returned. Any other property of
57 HTML, such as <base href=...>, or strange way to specify a URL,
58 such as <meta http-equiv=Refresh content="0; URL=..."> had to be
59 crudely hacked in -- and the caller had to be aware of these hacks.
60 Like its predecessor, this parser did not support HTML comments.
62 After Wget 1.5.1 was released, I set out to write a third HTML
63 parser. The objectives of the new parser were to: (1) provide a
64 clean way to analyze HTML lexically, (2) separate interpretation of
65 the markup from the parsing process, (3) be as correct as possible,
66 e.g. correctly skipping comments and other SGML declarations, (4)
67 understand the most common errors in markup and skip them or be
68 relaxed towrds them, and (5) be reasonably efficient (no regexps,
69 minimum copying and minimum or no heap allocation).
71 I believe this parser meets all of the above goals. It is
72 reasonably well structured, and could be relatively easily
73 separated from Wget and used elsewhere. While some of its
74 intrinsic properties limit its value as a general-purpose HTML
75 parser, I believe that, with minimum modifications, it could serve
78 Due to time and other constraints, this parser was not integrated
79 into Wget until the version 1.7. */
83 The single entry point of this parser is map_html_tags(), which
84 works by calling a function you specify for each tag. The function
85 gets called with the pointer to a structure describing the tag and
88 /* To test as standalone, compile with `-DSTANDALONE -I.'. You'll
89 still need Wget headers to compile. */
94 # define I_REALLY_WANT_CTYPE_MACROS
102 # include <strings.h>
107 #include "html-parse.h"
113 # define xmalloc malloc
114 # define xrealloc realloc
125 # define ISSPACE(x) isspace (x)
126 # define ISDIGIT(x) isdigit (x)
127 # define ISXDIGIT(x) isxdigit (x)
128 # define ISALPHA(x) isalpha (x)
129 # define ISALNUM(x) isalnum (x)
130 # define TOLOWER(x) tolower (x)
131 # define TOUPPER(x) toupper (x)
133 static struct options opt;
134 #endif /* STANDALONE */
136 /* Pool support. A pool is a resizable chunk of memory. It is first
137 allocated on the stack, and moved to the heap if it needs to be
138 larger than originally expected. map_html_tags() uses it to store
139 the zero-terminated names and values of tags and attributes.
141 Thus taginfo->name, and attr->name and attr->value for each
142 attribute, do not point into separately allocated areas, but into
143 different parts of the pool, separated only by terminating zeros.
144 This ensures minimum amount of allocation and, for most tags, no
145 allocation because the entire pool is kept on the stack. */
148 char *contents; /* pointer to the contents. */
149 int size; /* size of the pool. */
150 int tail; /* next available position index. */
151 int resized; /* whether the pool has been resized
154 char *orig_contents; /* original pool contents, usually
155 stack-allocated. used by POOL_FREE
156 to restore the pool to the initial
161 /* Initialize the pool to hold INITIAL_SIZE bytes of storage. */
163 #define POOL_INIT(p, initial_storage, initial_size) do { \
164 struct pool *P = (p); \
165 P->contents = (initial_storage); \
166 P->size = (initial_size); \
169 P->orig_contents = P->contents; \
170 P->orig_size = P->size; \
173 /* Grow the pool to accomodate at least SIZE new bytes. If the pool
174 already has room to accomodate SIZE bytes of data, this is a no-op. */
176 #define POOL_GROW(p, increase) \
177 GROW_ARRAY ((p)->contents, (p)->size, (p)->tail + (increase), \
180 /* Append text in the range [beg, end) to POOL. No zero-termination
183 #define POOL_APPEND(p, beg, end) do { \
184 const char *PA_beg = (beg); \
185 int PA_size = (end) - PA_beg; \
186 POOL_GROW (p, PA_size); \
187 memcpy ((p)->contents + (p)->tail, PA_beg, PA_size); \
188 (p)->tail += PA_size; \
191 /* Append one character to the pool. Can be used to zero-terminate
194 #define POOL_APPEND_CHR(p, ch) do { \
195 char PAC_char = (ch); \
197 (p)->contents[(p)->tail++] = PAC_char; \
200 /* Forget old pool contents. The allocated memory is not freed. */
201 #define POOL_REWIND(p) (p)->tail = 0
203 /* Free heap-allocated memory for contents of POOL. This calls
204 xfree() if the memory was allocated through malloc. It also
205 restores `contents' and `size' to their original, pre-malloc
206 values. That way after POOL_FREE, the pool is fully usable, just
207 as if it were freshly initialized with POOL_INIT. */
209 #define POOL_FREE(p) do { \
210 struct pool *P = p; \
212 xfree (P->contents); \
213 P->contents = P->orig_contents; \
214 P->size = P->orig_size; \
219 /* Used for small stack-allocated memory chunks that might grow. Like
220 DO_REALLOC, this macro grows BASEVAR as necessary to take
221 NEEDED_SIZE items of TYPE.
223 The difference is that on the first resize, it will use
224 malloc+memcpy rather than realloc. That way you can stack-allocate
225 the initial chunk, and only resort to heap allocation if you
226 stumble upon large data.
228 After the first resize, subsequent ones are performed with realloc,
229 just like DO_REALLOC. */
231 #define GROW_ARRAY(basevar, sizevar, needed_size, resized, type) do { \
232 long ga_needed_size = (needed_size); \
233 long ga_newsize = (sizevar); \
234 while (ga_newsize < ga_needed_size) \
236 if (ga_newsize != (sizevar)) \
239 basevar = (type *)xrealloc (basevar, ga_newsize * sizeof (type)); \
242 void *ga_new = xmalloc (ga_newsize * sizeof (type)); \
243 memcpy (ga_new, basevar, (sizevar) * sizeof (type)); \
244 (basevar) = ga_new; \
247 (sizevar) = ga_newsize; \
251 #define AP_DOWNCASE 1
252 #define AP_PROCESS_ENTITIES 2
253 #define AP_TRIM_BLANKS 4
255 /* Copy the text in the range [BEG, END) to POOL, optionally
256 performing operations specified by FLAGS. FLAGS may be any
257 combination of AP_DOWNCASE, AP_PROCESS_ENTITIES and AP_TRIM_BLANKS
258 with the following meaning:
260 * AP_DOWNCASE -- downcase all the letters;
262 * AP_PROCESS_ENTITIES -- process the SGML entities and write out
263 the decoded string. Recognized entities are <, >, &, ",
264   and the numerical entities.
266 * AP_TRIM_BLANKS -- ignore blanks at the beginning and at the end
270 convert_and_copy (struct pool *pool, const char *beg, const char *end, int flags)
272 int old_tail = pool->tail;
275 /* First, skip blanks if required. We must do this before entities
276 are processed, so that blanks can still be inserted as, for
277 instance, ` '. */
278 if (flags & AP_TRIM_BLANKS)
280 while (beg < end && ISSPACE (*beg))
282 while (end > beg && ISSPACE (end[-1]))
287 if (flags & AP_PROCESS_ENTITIES)
289 /* Grow the pool, then copy the text to the pool character by
290 character, processing the encountered entities as we go
293 It's safe (and necessary) to grow the pool in advance because
294 processing the entities can only *shorten* the string, it can
295 never lengthen it. */
296 const char *from = beg;
299 POOL_GROW (pool, end - beg);
300 to = pool->contents + pool->tail;
308 const char *save = from;
315 /* Process numeric entities "&#DDD;" and "&#xHH;". */
318 int numeric = 0, digits = 0;
323 for (; from < end && ISXDIGIT (*from); from++, digits++)
324 numeric = (numeric << 4) + XDIGIT_TO_NUM (*from);
328 for (; from < end && ISDIGIT (*from); from++, digits++)
329 numeric = (numeric * 10) + (*from - '0');
336 #define FROB(x) (remain >= (sizeof (x) - 1) \
337 && 0 == memcmp (from, x, sizeof (x) - 1) \
338 && (*(from + sizeof (x) - 1) == ';' \
339 || remain == sizeof (x) - 1 \
340 || !ISALNUM (*(from + sizeof (x) - 1))))
341 else if (FROB ("lt"))
342 *to++ = '<', from += 2;
343 else if (FROB ("gt"))
344 *to++ = '>', from += 2;
345 else if (FROB ("amp"))
346 *to++ = '&', from += 3;
347 else if (FROB ("quot"))
348 *to++ = '\"', from += 4;
349 /* We don't implement the proposed "Added Latin 1"
350 entities (except for nbsp), because it is unnecessary
351 in the context of Wget, and would require hashing to
353 else if (FROB ("nbsp"))
354 *to++ = 160, from += 4;
358 /* If the entity was followed by `;', we step over the
359 `;'. Otherwise, it was followed by either a
360 non-alphanumeric or EOB, in which case we do nothing. */
361 if (from < end && *from == ';')
366 /* This was not an entity after all. Back out. */
371 /* Verify that we haven't exceeded the original size. (It
372 shouldn't happen, hence the assert.) */
373 assert (to - (pool->contents + pool->tail) <= end - beg);
375 /* Make POOL's tail point to the position following the string
377 pool->tail = to - pool->contents;
378 POOL_APPEND_CHR (pool, '\0');
382 /* Just copy the text to the pool. */
383 POOL_APPEND (pool, beg, end);
384 POOL_APPEND_CHR (pool, '\0');
387 if (flags & AP_DOWNCASE)
389 char *p = pool->contents + old_tail;
395 /* Check whether the contents of [POS, POS+LENGTH) match any of the
396 strings in the ARRAY. */
398 array_allowed (const char **array, const char *beg, const char *end)
400 int length = end - beg;
403 for (; *array; array++)
404 if (length >= strlen (*array)
405 && !strncasecmp (*array, beg, length))
413 /* Originally we used to adhere to rfc 1866 here, and allowed only
414 letters, digits, periods, and hyphens as names (of tags or
415 attributes). However, this broke too many pages which used
416 proprietary or strange attributes, e.g. <img src="a.gif"
417 v:shapes="whatever">.
419 So now we allow any character except:
421 * 8-bit and control chars
422 * characters that clearly cannot be part of name:
425 This only affects attribute and tag names; attribute values allow
426 an even greater variety of characters. */
428 #define NAME_CHAR_P(x) ((x) > 32 && (x) < 127 \
429 && (x) != '=' && (x) != '>' && (x) != '/')
432 static int comment_backout_count;
435 /* Advance over an SGML declaration, such as <!DOCTYPE ...>. In
436 strict comments mode, this is used for skipping over comments as
439 To recap: any SGML declaration may have comments associated with
441 <!MY-DECL -- isn't this fun? -- foo bar>
443 An HTML comment is merely an empty declaration (<!>) with a comment
445 <!-- some stuff here -->
447 Several comments may be embedded in one comment declaration:
448 <!-- have -- -- fun -->
450 Whitespace is allowed between and after the comments, but not
451 before the first comment. Additionally, this function attempts to
452 handle double quotes in SGML declarations correctly. */
455 advance_declaration (const char *beg, const char *end)
458 char quote_char = '\0'; /* shut up, gcc! */
481 /* It looked like a good idea to write this as a state machine, but
484 while (state != AC_S_DONE && state != AC_S_BACKOUT)
487 state = AC_S_BACKOUT;
497 state = AC_S_DEFAULT;
500 state = AC_S_BACKOUT;
522 if (NAME_CHAR_P (ch))
523 state = AC_S_DCLNAME;
525 state = AC_S_BACKOUT;
532 else if (NAME_CHAR_P (ch))
535 state = AC_S_DEFAULT;
538 /* We must use 0x22 because broken assert macros choke on
540 assert (ch == '\'' || ch == 0x22);
541 quote_char = ch; /* cheating -- I really don't feel like
542 introducing more different states for
543 different quote characters. */
545 state = AC_S_IN_QUOTE;
548 if (ch == quote_char)
554 assert (ch == quote_char);
556 state = AC_S_DEFAULT;
568 state = AC_S_COMMENT;
571 state = AC_S_BACKOUT;
595 state = AC_S_DEFAULT;
598 state = AC_S_COMMENT;
605 if (state == AC_S_BACKOUT)
608 ++comment_backout_count;
615 /* Find the first occurrence of the substring "-->" in [BEG, END) and
616 return the pointer to the character after the substring. If the
617 substring is not found, return NULL. */
620 find_comment_end (const char *beg, const char *end)
622 /* Open-coded Boyer-Moore search for "-->". Examine the third char;
623 if it's not '>' or '-', advance by three characters. Otherwise,
624 look at the preceding characters and try to find a match. */
626 const char *p = beg - 1;
628 while ((p += 3) < end)
632 if (p[-1] == '-' && p[-2] == '-')
640 if (++p == end) return NULL;
643 case '>': return p + 1;
644 case '-': goto at_dash_dash;
649 if ((p += 2) >= end) return NULL;
664 /* Advance P (a char pointer), with the explicit intent of being able
665 to read the next character. If this is not possible, go to finish. */
667 #define ADVANCE(p) do { \
673 /* Skip whitespace, if any. */
675 #define SKIP_WS(p) do { \
676 while (ISSPACE (*p)) { \
681 /* Skip non-whitespace, if any. */
683 #define SKIP_NON_WS(p) do { \
684 while (!ISSPACE (*p)) { \
690 static int tag_backout_count;
693 /* Map MAPFUN over HTML tags in TEXT, which is SIZE characters long.
694 MAPFUN will be called with two arguments: pointer to an initialized
695 struct taginfo, and CLOSURE.
697 ALLOWED_TAG_NAMES should be a NULL-terminated array of tag names to
698 be processed by this function. If it is NULL, all the tags are
699 allowed. The same goes for attributes and ALLOWED_ATTRIBUTE_NAMES.
701 (Obviously, the caller can filter out unwanted tags and attributes
702 just as well, but this is just an optimization designed to avoid
703 unnecessary copying for tags/attributes which the caller doesn't
704 want to know about. These lists are searched linearly; therefore,
705 if you're interested in a large number of tags or attributes, you'd
706 better set these to NULL and filter them out yourself with a
707 hashing process most appropriate for your application.) */
710 map_html_tags (const char *text, int size,
711 const char **allowed_tag_names,
712 const char **allowed_attribute_names,
713 void (*mapfun) (struct taginfo *, void *),
716 /* storage for strings passed to MAPFUN callback; if 256 bytes is
717 too little, POOL_APPEND allocates more with malloc. */
718 char pool_initial_storage[256];
721 const char *p = text;
722 const char *end = text + size;
724 struct attr_pair attr_pair_initial_storage[8];
725 int attr_pair_size = countof (attr_pair_initial_storage);
726 int attr_pair_resized = 0;
727 struct attr_pair *pairs = attr_pair_initial_storage;
732 POOL_INIT (&pool, pool_initial_storage, countof (pool_initial_storage));
736 const char *tag_name_begin, *tag_name_end;
737 const char *tag_start_position;
738 int uninteresting_tag;
746 /* Find beginning of tag. We use memchr() instead of the usual
747 looping with ADVANCE() for speed. */
748 p = memchr (p, '<', end - p);
752 tag_start_position = p;
755 /* Establish the type of the tag (start-tag, end-tag or
759 if (!opt.strict_comments
760 && p < end + 3 && p[1] == '-' && p[2] == '-')
762 /* If strict comments are not enforced and if we know
763 we're looking at a comment, simply look for the
764 terminating "-->". Non-strict is the default because
765 it works in other browsers and most HTML writers can't
766 be bothered with getting the comments right. */
767 const char *comment_end = find_comment_end (p + 3, end);
773 /* Either in strict comment mode or looking at a non-empty
774 declaration. Real declarations are much less likely to
775 be misused the way comments are, so advance over them
776 properly regardless of strictness. */
777 p = advance_declaration (p, end);
789 while (NAME_CHAR_P (*p))
791 if (p == tag_name_begin)
795 if (end_tag && *p != '>')
798 if (!array_allowed (allowed_tag_names, tag_name_begin, tag_name_end))
799 /* We can't just say "goto look_for_tag" here because we need
800 the loop below to properly advance over the tag's attributes. */
801 uninteresting_tag = 1;
804 uninteresting_tag = 0;
805 convert_and_copy (&pool, tag_name_begin, tag_name_end, AP_DOWNCASE);
808 /* Find the attributes. */
811 const char *attr_name_begin, *attr_name_end;
812 const char *attr_value_begin, *attr_value_end;
813 const char *attr_raw_value_begin, *attr_raw_value_end;
814 int operation = AP_DOWNCASE; /* stupid compiler. */
820 /* A slash at this point means the tag is about to be
821 closed. This is legal in XML and has been popularized
822 in HTML via XHTML. */
823 /* <foo a=b c=d /> */
831 /* Check for end of tag definition. */
835 /* Establish bounds of attribute name. */
836 attr_name_begin = p; /* <foo bar ...> */
838 while (NAME_CHAR_P (*p))
840 attr_name_end = p; /* <foo bar ...> */
842 if (attr_name_begin == attr_name_end)
845 /* Establish bounds of attribute value. */
847 if (NAME_CHAR_P (*p) || *p == '/' || *p == '>')
849 /* Minimized attribute syntax allows `=' to be omitted.
850 For example, <UL COMPACT> is a valid shorthand for <UL
851 COMPACT="compact">. Even if such attributes are not
852 useful to Wget, we need to support them, so that the
853 tags containing them can be parsed correctly. */
854 attr_raw_value_begin = attr_value_begin = attr_name_begin;
855 attr_raw_value_end = attr_value_end = attr_name_end;
861 if (*p == '\"' || *p == '\'')
863 int newline_seen = 0;
864 char quote_char = *p;
865 attr_raw_value_begin = p;
867 attr_value_begin = p; /* <foo bar="baz"> */
869 while (*p != quote_char)
871 if (!newline_seen && *p == '\n')
873 /* If a newline is seen within the quotes, it
874 is most likely that someone forgot to close
875 the quote. In that case, we back out to
876 the value beginning, and terminate the tag
877 at either `>' or the delimiter, whichever
878 comes first. Such a tag terminated at `>'
880 p = attr_value_begin;
884 else if (newline_seen && *p == '>')
888 attr_value_end = p; /* <foo bar="baz"> */
890 if (*p == quote_char)
894 attr_raw_value_end = p; /* <foo bar="baz"> */
896 /* The AP_TRIM_BLANKS is there for buggy HTML
897 generators that generate <a href=" foo"> instead of
898 <a href="foo"> (Netscape ignores spaces as well.)
899 If you really mean space, use &32; or %20. */
900 operation = AP_PROCESS_ENTITIES | AP_TRIM_BLANKS;
904 attr_value_begin = p; /* <foo bar=baz> */
906 /* According to SGML, a name token should consist only
907 of alphanumerics, . and -. However, this is often
908 violated by, for instance, `%' in `width=75%'.
909 We'll be liberal and allow just about anything as
910 an attribute value. */
911 while (!ISSPACE (*p) && *p != '>')
913 attr_value_end = p; /* <foo bar=baz qux=quix> */
915 if (attr_value_begin == attr_value_end)
919 attr_raw_value_begin = attr_value_begin;
920 attr_raw_value_end = attr_value_end;
921 operation = AP_PROCESS_ENTITIES;
926 /* We skipped the whitespace and found something that is
927 neither `=' nor the beginning of the next attribute's
929 goto backout_tag; /* <foo bar [... */
933 /* If we're not interested in the tag, don't bother with any
934 of the attributes. */
935 if (uninteresting_tag)
938 /* If we aren't interested in the attribute, skip it. We
939 cannot do this test any sooner, because our text pointer
940 needs to correctly advance over the attribute. */
941 if (allowed_attribute_names
942 && !array_allowed (allowed_attribute_names, attr_name_begin,
946 GROW_ARRAY (pairs, attr_pair_size, nattrs + 1, attr_pair_resized,
949 pairs[nattrs].name_pool_index = pool.tail;
950 convert_and_copy (&pool, attr_name_begin, attr_name_end, AP_DOWNCASE);
952 pairs[nattrs].value_pool_index = pool.tail;
953 convert_and_copy (&pool, attr_value_begin, attr_value_end, operation);
954 pairs[nattrs].value_raw_beginning = attr_raw_value_begin;
955 pairs[nattrs].value_raw_size = (attr_raw_value_end
956 - attr_raw_value_begin);
960 if (uninteresting_tag)
966 /* By now, we have a valid tag with a name and zero or more
967 attributes. Fill in the data and call the mapper function. */
970 struct taginfo taginfo;
972 taginfo.name = pool.contents;
973 taginfo.end_tag_p = end_tag;
974 taginfo.nattrs = nattrs;
975 /* We fill in the char pointers only now, when pool can no
976 longer get realloc'ed. If we did that above, we could get
977 hosed by reallocation. Obviously, after this point, the pool
978 may no longer be grown. */
979 for (i = 0; i < nattrs; i++)
981 pairs[i].name = pool.contents + pairs[i].name_pool_index;
982 pairs[i].value = pool.contents + pairs[i].value_pool_index;
984 taginfo.attrs = pairs;
985 taginfo.start_position = tag_start_position;
986 taginfo.end_position = p + 1;
988 (*mapfun) (&taginfo, closure);
997 /* The tag wasn't really a tag. Treat its contents as ordinary
999 p = tag_start_position + 1;
1005 if (attr_pair_resized)
1015 test_mapper (struct taginfo *taginfo, void *arg)
1019 printf ("%s%s", taginfo->end_tag_p ? "/" : "", taginfo->name);
1020 for (i = 0; i < taginfo->nattrs; i++)
1021 printf (" %s=%s", taginfo->attrs[i].name, taginfo->attrs[i].value);
1029 char *x = (char *)xmalloc (size);
1032 int tag_counter = 0;
1034 while ((read_count = fread (x + length, 1, size - length, stdin)))
1036 length += read_count;
1038 x = (char *)xrealloc (x, size);
1041 map_html_tags (x, length, NULL, NULL, test_mapper, &tag_counter);
1042 printf ("TAGS: %d\n", tag_counter);
1043 printf ("Tag backouts: %d\n", tag_backout_count);
1044 printf ("Comment backouts: %d\n", comment_backout_count);
1047 #endif /* STANDALONE */