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)
137 hash_table_get (const struct hash_table *ht, void *ptr)
141 #else /* not STANDALONE */
145 /* Pool support. A pool is a resizable chunk of memory. It is first
146 allocated on the stack, and moved to the heap if it needs to be
147 larger than originally expected. map_html_tags() uses it to store
148 the zero-terminated names and values of tags and attributes.
150 Thus taginfo->name, and attr->name and attr->value for each
151 attribute, do not point into separately allocated areas, but into
152 different parts of the pool, separated only by terminating zeros.
153 This ensures minimum amount of allocation and, for most tags, no
154 allocation because the entire pool is kept on the stack. */
157 char *contents; /* pointer to the contents. */
158 int size; /* size of the pool. */
159 int tail; /* next available position index. */
160 int resized; /* whether the pool has been resized
163 char *orig_contents; /* original pool contents, usually
164 stack-allocated. used by POOL_FREE
165 to restore the pool to the initial
170 /* Initialize the pool to hold INITIAL_SIZE bytes of storage. */
172 #define POOL_INIT(p, initial_storage, initial_size) do { \
173 struct pool *P = (p); \
174 P->contents = (initial_storage); \
175 P->size = (initial_size); \
178 P->orig_contents = P->contents; \
179 P->orig_size = P->size; \
182 /* Grow the pool to accomodate at least SIZE new bytes. If the pool
183 already has room to accomodate SIZE bytes of data, this is a no-op. */
185 #define POOL_GROW(p, increase) \
186 GROW_ARRAY ((p)->contents, (p)->size, (p)->tail + (increase), \
189 /* Append text in the range [beg, end) to POOL. No zero-termination
192 #define POOL_APPEND(p, beg, end) do { \
193 const char *PA_beg = (beg); \
194 int PA_size = (end) - PA_beg; \
195 POOL_GROW (p, PA_size); \
196 memcpy ((p)->contents + (p)->tail, PA_beg, PA_size); \
197 (p)->tail += PA_size; \
200 /* Append one character to the pool. Can be used to zero-terminate
203 #define POOL_APPEND_CHR(p, ch) do { \
204 char PAC_char = (ch); \
206 (p)->contents[(p)->tail++] = PAC_char; \
209 /* Forget old pool contents. The allocated memory is not freed. */
210 #define POOL_REWIND(p) (p)->tail = 0
212 /* Free heap-allocated memory for contents of POOL. This calls
213 xfree() if the memory was allocated through malloc. It also
214 restores `contents' and `size' to their original, pre-malloc
215 values. That way after POOL_FREE, the pool is fully usable, just
216 as if it were freshly initialized with POOL_INIT. */
218 #define POOL_FREE(p) do { \
219 struct pool *P = p; \
221 xfree (P->contents); \
222 P->contents = P->orig_contents; \
223 P->size = P->orig_size; \
228 /* Used for small stack-allocated memory chunks that might grow. Like
229 DO_REALLOC, this macro grows BASEVAR as necessary to take
230 NEEDED_SIZE items of TYPE.
232 The difference is that on the first resize, it will use
233 malloc+memcpy rather than realloc. That way you can stack-allocate
234 the initial chunk, and only resort to heap allocation if you
235 stumble upon large data.
237 After the first resize, subsequent ones are performed with realloc,
238 just like DO_REALLOC. */
240 #define GROW_ARRAY(basevar, sizevar, needed_size, resized, type) do { \
241 long ga_needed_size = (needed_size); \
242 long ga_newsize = (sizevar); \
243 while (ga_newsize < ga_needed_size) \
245 if (ga_newsize != (sizevar)) \
248 basevar = (type *)xrealloc (basevar, ga_newsize * sizeof (type)); \
251 void *ga_new = xmalloc (ga_newsize * sizeof (type)); \
252 memcpy (ga_new, basevar, (sizevar) * sizeof (type)); \
253 (basevar) = ga_new; \
256 (sizevar) = ga_newsize; \
260 #define AP_DOWNCASE 1
261 #define AP_PROCESS_ENTITIES 2
262 #define AP_TRIM_BLANKS 4
264 /* Copy the text in the range [BEG, END) to POOL, optionally
265 performing operations specified by FLAGS. FLAGS may be any
266 combination of AP_DOWNCASE, AP_PROCESS_ENTITIES and AP_TRIM_BLANKS
267 with the following meaning:
269 * AP_DOWNCASE -- downcase all the letters;
271 * AP_PROCESS_ENTITIES -- process the SGML entities and write out
272 the decoded string. Recognized entities are <, >, &, ",
273   and the numerical entities.
275 * AP_TRIM_BLANKS -- ignore blanks at the beginning and at the end
279 convert_and_copy (struct pool *pool, const char *beg, const char *end, int flags)
281 int old_tail = pool->tail;
284 /* First, skip blanks if required. We must do this before entities
285 are processed, so that blanks can still be inserted as, for
286 instance, ` '. */
287 if (flags & AP_TRIM_BLANKS)
289 while (beg < end && ISSPACE (*beg))
291 while (end > beg && ISSPACE (end[-1]))
296 if (flags & AP_PROCESS_ENTITIES)
298 /* Grow the pool, then copy the text to the pool character by
299 character, processing the encountered entities as we go
302 It's safe (and necessary) to grow the pool in advance because
303 processing the entities can only *shorten* the string, it can
304 never lengthen it. */
305 const char *from = beg;
308 POOL_GROW (pool, end - beg);
309 to = pool->contents + pool->tail;
317 const char *save = from;
324 /* Process numeric entities "&#DDD;" and "&#xHH;". */
327 int numeric = 0, digits = 0;
332 for (; from < end && ISXDIGIT (*from); from++, digits++)
333 numeric = (numeric << 4) + XDIGIT_TO_NUM (*from);
337 for (; from < end && ISDIGIT (*from); from++, digits++)
338 numeric = (numeric * 10) + (*from - '0');
345 #define FROB(x) (remain >= (sizeof (x) - 1) \
346 && 0 == memcmp (from, x, sizeof (x) - 1) \
347 && (*(from + sizeof (x) - 1) == ';' \
348 || remain == sizeof (x) - 1 \
349 || !ISALNUM (*(from + sizeof (x) - 1))))
350 else if (FROB ("lt"))
351 *to++ = '<', from += 2;
352 else if (FROB ("gt"))
353 *to++ = '>', from += 2;
354 else if (FROB ("amp"))
355 *to++ = '&', from += 3;
356 else if (FROB ("quot"))
357 *to++ = '\"', from += 4;
358 /* We don't implement the proposed "Added Latin 1"
359 entities (except for nbsp), because it is unnecessary
360 in the context of Wget, and would require hashing to
362 else if (FROB ("nbsp"))
363 *to++ = 160, from += 4;
367 /* If the entity was followed by `;', we step over the
368 `;'. Otherwise, it was followed by either a
369 non-alphanumeric or EOB, in which case we do nothing. */
370 if (from < end && *from == ';')
375 /* This was not an entity after all. Back out. */
380 /* Verify that we haven't exceeded the original size. (It
381 shouldn't happen, hence the assert.) */
382 assert (to - (pool->contents + pool->tail) <= end - beg);
384 /* Make POOL's tail point to the position following the string
386 pool->tail = to - pool->contents;
387 POOL_APPEND_CHR (pool, '\0');
391 /* Just copy the text to the pool. */
392 POOL_APPEND (pool, beg, end);
393 POOL_APPEND_CHR (pool, '\0');
396 if (flags & AP_DOWNCASE)
398 char *p = pool->contents + old_tail;
404 /* Originally we used to adhere to rfc 1866 here, and allowed only
405 letters, digits, periods, and hyphens as names (of tags or
406 attributes). However, this broke too many pages which used
407 proprietary or strange attributes, e.g. <img src="a.gif"
408 v:shapes="whatever">.
410 So now we allow any character except:
412 * 8-bit and control chars
413 * characters that clearly cannot be part of name:
416 This only affects attribute and tag names; attribute values allow
417 an even greater variety of characters. */
419 #define NAME_CHAR_P(x) ((x) > 32 && (x) < 127 \
420 && (x) != '=' && (x) != '>' && (x) != '/')
423 static int comment_backout_count;
426 /* Advance over an SGML declaration, such as <!DOCTYPE ...>. In
427 strict comments mode, this is used for skipping over comments as
430 To recap: any SGML declaration may have comments associated with
432 <!MY-DECL -- isn't this fun? -- foo bar>
434 An HTML comment is merely an empty declaration (<!>) with a comment
436 <!-- some stuff here -->
438 Several comments may be embedded in one comment declaration:
439 <!-- have -- -- fun -->
441 Whitespace is allowed between and after the comments, but not
442 before the first comment. Additionally, this function attempts to
443 handle double quotes in SGML declarations correctly. */
446 advance_declaration (const char *beg, const char *end)
449 char quote_char = '\0'; /* shut up, gcc! */
472 /* It looked like a good idea to write this as a state machine, but
475 while (state != AC_S_DONE && state != AC_S_BACKOUT)
478 state = AC_S_BACKOUT;
488 state = AC_S_DEFAULT;
491 state = AC_S_BACKOUT;
513 if (NAME_CHAR_P (ch))
514 state = AC_S_DCLNAME;
516 state = AC_S_BACKOUT;
523 else if (NAME_CHAR_P (ch))
526 state = AC_S_DEFAULT;
529 /* We must use 0x22 because broken assert macros choke on
531 assert (ch == '\'' || ch == 0x22);
532 quote_char = ch; /* cheating -- I really don't feel like
533 introducing more different states for
534 different quote characters. */
536 state = AC_S_IN_QUOTE;
539 if (ch == quote_char)
545 assert (ch == quote_char);
547 state = AC_S_DEFAULT;
559 state = AC_S_COMMENT;
562 state = AC_S_BACKOUT;
586 state = AC_S_DEFAULT;
589 state = AC_S_COMMENT;
596 if (state == AC_S_BACKOUT)
599 ++comment_backout_count;
606 /* Find the first occurrence of the substring "-->" in [BEG, END) and
607 return the pointer to the character after the substring. If the
608 substring is not found, return NULL. */
611 find_comment_end (const char *beg, const char *end)
613 /* Open-coded Boyer-Moore search for "-->". Examine the third char;
614 if it's not '>' or '-', advance by three characters. Otherwise,
615 look at the preceding characters and try to find a match. */
617 const char *p = beg - 1;
619 while ((p += 3) < end)
623 if (p[-1] == '-' && p[-2] == '-')
631 if (++p == end) return NULL;
634 case '>': return p + 1;
635 case '-': goto at_dash_dash;
640 if ((p += 2) >= end) return NULL;
655 /* Return non-zero of the string inside [b, e) are present in hash
659 name_allowed (const struct hash_table *ht, const char *b, const char *e)
664 BOUNDED_TO_ALLOCA (b, e, copy);
665 return hash_table_get (ht, copy) != NULL;
668 /* Advance P (a char pointer), with the explicit intent of being able
669 to read the next character. If this is not possible, go to finish. */
671 #define ADVANCE(p) do { \
677 /* Skip whitespace, if any. */
679 #define SKIP_WS(p) do { \
680 while (ISSPACE (*p)) { \
685 /* Skip non-whitespace, if any. */
687 #define SKIP_NON_WS(p) do { \
688 while (!ISSPACE (*p)) { \
694 static int tag_backout_count;
697 /* Map MAPFUN over HTML tags in TEXT, which is SIZE characters long.
698 MAPFUN will be called with two arguments: pointer to an initialized
699 struct taginfo, and MAPARG.
701 ALLOWED_TAG_NAMES should be a NULL-terminated array of tag names to
702 be processed by this function. If it is NULL, all the tags are
703 allowed. The same goes for attributes and ALLOWED_ATTRIBUTE_NAMES.
705 (Obviously, the caller can filter out unwanted tags and attributes
706 just as well, but this is just an optimization designed to avoid
707 unnecessary copying for tags/attributes which the caller doesn't
708 want to know about. These lists are searched linearly; therefore,
709 if you're interested in a large number of tags or attributes, you'd
710 better set these to NULL and filter them out yourself with a
711 hashing process most appropriate for your application.) */
714 map_html_tags (const char *text, int size,
715 void (*mapfun) (struct taginfo *, void *), void *maparg,
717 const struct hash_table *allowed_tags,
718 const struct hash_table *allowed_attributes)
720 /* storage for strings passed to MAPFUN callback; if 256 bytes is
721 too little, POOL_APPEND allocates more with malloc. */
722 char pool_initial_storage[256];
725 const char *p = text;
726 const char *end = text + size;
728 struct attr_pair attr_pair_initial_storage[8];
729 int attr_pair_size = countof (attr_pair_initial_storage);
730 int attr_pair_resized = 0;
731 struct attr_pair *pairs = attr_pair_initial_storage;
736 POOL_INIT (&pool, pool_initial_storage, countof (pool_initial_storage));
740 const char *tag_name_begin, *tag_name_end;
741 const char *tag_start_position;
742 int uninteresting_tag;
750 /* Find beginning of tag. We use memchr() instead of the usual
751 looping with ADVANCE() for speed. */
752 p = memchr (p, '<', end - p);
756 tag_start_position = p;
759 /* Establish the type of the tag (start-tag, end-tag or
763 if (!(flags & MHT_STRICT_COMMENTS)
764 && p < end + 3 && p[1] == '-' && p[2] == '-')
766 /* If strict comments are not enforced and if we know
767 we're looking at a comment, simply look for the
768 terminating "-->". Non-strict is the default because
769 it works in other browsers and most HTML writers can't
770 be bothered with getting the comments right. */
771 const char *comment_end = find_comment_end (p + 3, end);
777 /* Either in strict comment mode or looking at a non-empty
778 declaration. Real declarations are much less likely to
779 be misused the way comments are, so advance over them
780 properly regardless of strictness. */
781 p = advance_declaration (p, end);
793 while (NAME_CHAR_P (*p))
795 if (p == tag_name_begin)
799 if (end_tag && *p != '>')
802 if (!name_allowed (allowed_tags, tag_name_begin, tag_name_end))
803 /* We can't just say "goto look_for_tag" here because we need
804 the loop below to properly advance over the tag's attributes. */
805 uninteresting_tag = 1;
808 uninteresting_tag = 0;
809 convert_and_copy (&pool, tag_name_begin, tag_name_end, AP_DOWNCASE);
812 /* Find the attributes. */
815 const char *attr_name_begin, *attr_name_end;
816 const char *attr_value_begin, *attr_value_end;
817 const char *attr_raw_value_begin, *attr_raw_value_end;
818 int operation = AP_DOWNCASE; /* stupid compiler. */
824 /* A slash at this point means the tag is about to be
825 closed. This is legal in XML and has been popularized
826 in HTML via XHTML. */
827 /* <foo a=b c=d /> */
835 /* Check for end of tag definition. */
839 /* Establish bounds of attribute name. */
840 attr_name_begin = p; /* <foo bar ...> */
842 while (NAME_CHAR_P (*p))
844 attr_name_end = p; /* <foo bar ...> */
846 if (attr_name_begin == attr_name_end)
849 /* Establish bounds of attribute value. */
851 if (NAME_CHAR_P (*p) || *p == '/' || *p == '>')
853 /* Minimized attribute syntax allows `=' to be omitted.
854 For example, <UL COMPACT> is a valid shorthand for <UL
855 COMPACT="compact">. Even if such attributes are not
856 useful to Wget, we need to support them, so that the
857 tags containing them can be parsed correctly. */
858 attr_raw_value_begin = attr_value_begin = attr_name_begin;
859 attr_raw_value_end = attr_value_end = attr_name_end;
865 if (*p == '\"' || *p == '\'')
867 int newline_seen = 0;
868 char quote_char = *p;
869 attr_raw_value_begin = p;
871 attr_value_begin = p; /* <foo bar="baz"> */
873 while (*p != quote_char)
875 if (!newline_seen && *p == '\n')
877 /* If a newline is seen within the quotes, it
878 is most likely that someone forgot to close
879 the quote. In that case, we back out to
880 the value beginning, and terminate the tag
881 at either `>' or the delimiter, whichever
882 comes first. Such a tag terminated at `>'
884 p = attr_value_begin;
888 else if (newline_seen && *p == '>')
892 attr_value_end = p; /* <foo bar="baz"> */
894 if (*p == quote_char)
898 attr_raw_value_end = p; /* <foo bar="baz"> */
900 operation = AP_PROCESS_ENTITIES;
901 if (flags & MHT_TRIM_VALUES)
902 operation |= AP_TRIM_BLANKS;
906 attr_value_begin = p; /* <foo bar=baz> */
908 /* According to SGML, a name token should consist only
909 of alphanumerics, . and -. However, this is often
910 violated by, for instance, `%' in `width=75%'.
911 We'll be liberal and allow just about anything as
912 an attribute value. */
913 while (!ISSPACE (*p) && *p != '>')
915 attr_value_end = p; /* <foo bar=baz qux=quix> */
917 if (attr_value_begin == attr_value_end)
921 attr_raw_value_begin = attr_value_begin;
922 attr_raw_value_end = attr_value_end;
923 operation = AP_PROCESS_ENTITIES;
928 /* We skipped the whitespace and found something that is
929 neither `=' nor the beginning of the next attribute's
931 goto backout_tag; /* <foo bar [... */
935 /* If we're not interested in the tag, don't bother with any
936 of the attributes. */
937 if (uninteresting_tag)
940 /* If we aren't interested in the attribute, skip it. We
941 cannot do this test any sooner, because our text pointer
942 needs to correctly advance over the attribute. */
943 if (!name_allowed (allowed_attributes, attr_name_begin, attr_name_end))
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, maparg);
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, test_mapper, &tag_counter, 0, NULL, NULL);
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 */