1 /* HTML parser for Wget.
2 Copyright (C) 1998-2006 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 Foundation, Inc.,
18 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 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
103 #include "html-parse.h"
109 # define xmalloc malloc
110 # define xrealloc realloc
121 # define ISSPACE(x) isspace (x)
122 # define ISDIGIT(x) isdigit (x)
123 # define ISXDIGIT(x) isxdigit (x)
124 # define ISALPHA(x) isalpha (x)
125 # define ISALNUM(x) isalnum (x)
126 # define TOLOWER(x) tolower (x)
127 # define TOUPPER(x) toupper (x)
133 hash_table_get (const struct hash_table *ht, void *ptr)
137 #else /* not STANDALONE */
141 /* Pool support. A pool is a resizable chunk of memory. It is first
142 allocated on the stack, and moved to the heap if it needs to be
143 larger than originally expected. map_html_tags() uses it to store
144 the zero-terminated names and values of tags and attributes.
146 Thus taginfo->name, and attr->name and attr->value for each
147 attribute, do not point into separately allocated areas, but into
148 different parts of the pool, separated only by terminating zeros.
149 This ensures minimum amount of allocation and, for most tags, no
150 allocation because the entire pool is kept on the stack. */
153 char *contents; /* pointer to the contents. */
154 int size; /* size of the pool. */
155 int tail; /* next available position index. */
156 bool resized; /* whether the pool has been resized
159 char *orig_contents; /* original pool contents, usually
160 stack-allocated. used by POOL_FREE
161 to restore the pool to the initial
166 /* Initialize the pool to hold INITIAL_SIZE bytes of storage. */
168 #define POOL_INIT(p, initial_storage, initial_size) do { \
169 struct pool *P = (p); \
170 P->contents = (initial_storage); \
171 P->size = (initial_size); \
173 P->resized = false; \
174 P->orig_contents = P->contents; \
175 P->orig_size = P->size; \
178 /* Grow the pool to accomodate at least SIZE new bytes. If the pool
179 already has room to accomodate SIZE bytes of data, this is a no-op. */
181 #define POOL_GROW(p, increase) \
182 GROW_ARRAY ((p)->contents, (p)->size, (p)->tail + (increase), \
185 /* Append text in the range [beg, end) to POOL. No zero-termination
188 #define POOL_APPEND(p, beg, end) do { \
189 const char *PA_beg = (beg); \
190 int PA_size = (end) - PA_beg; \
191 POOL_GROW (p, PA_size); \
192 memcpy ((p)->contents + (p)->tail, PA_beg, PA_size); \
193 (p)->tail += PA_size; \
196 /* Append one character to the pool. Can be used to zero-terminate
199 #define POOL_APPEND_CHR(p, ch) do { \
200 char PAC_char = (ch); \
202 (p)->contents[(p)->tail++] = PAC_char; \
205 /* Forget old pool contents. The allocated memory is not freed. */
206 #define POOL_REWIND(p) (p)->tail = 0
208 /* Free heap-allocated memory for contents of POOL. This calls
209 xfree() if the memory was allocated through malloc. It also
210 restores `contents' and `size' to their original, pre-malloc
211 values. That way after POOL_FREE, the pool is fully usable, just
212 as if it were freshly initialized with POOL_INIT. */
214 #define POOL_FREE(p) do { \
215 struct pool *P = p; \
217 xfree (P->contents); \
218 P->contents = P->orig_contents; \
219 P->size = P->orig_size; \
221 P->resized = false; \
224 /* Used for small stack-allocated memory chunks that might grow. Like
225 DO_REALLOC, this macro grows BASEVAR as necessary to take
226 NEEDED_SIZE items of TYPE.
228 The difference is that on the first resize, it will use
229 malloc+memcpy rather than realloc. That way you can stack-allocate
230 the initial chunk, and only resort to heap allocation if you
231 stumble upon large data.
233 After the first resize, subsequent ones are performed with realloc,
234 just like DO_REALLOC. */
236 #define GROW_ARRAY(basevar, sizevar, needed_size, resized, type) do { \
237 long ga_needed_size = (needed_size); \
238 long ga_newsize = (sizevar); \
239 while (ga_newsize < ga_needed_size) \
241 if (ga_newsize != (sizevar)) \
244 basevar = xrealloc (basevar, ga_newsize * sizeof (type)); \
247 void *ga_new = xmalloc (ga_newsize * sizeof (type)); \
248 memcpy (ga_new, basevar, (sizevar) * sizeof (type)); \
249 (basevar) = ga_new; \
252 (sizevar) = ga_newsize; \
256 /* Test whether n+1-sized entity name fits in P. We don't support
257 IE-style non-terminated entities, e.g. "<foo" -> "<foo".
258 However, "<foo" will work, as will "<!foo", "<", etc. In
259 other words an entity needs to be terminated by either a
260 non-alphanumeric or the end of string. */
261 #define FITS(p, n) (p + n == end || (p + n < end && !ISALNUM (p[n])))
263 /* Macros that test entity names by returning true if P is followed by
264 the specified characters. */
265 #define ENT1(p, c0) (FITS (p, 1) && p[0] == c0)
266 #define ENT2(p, c0, c1) (FITS (p, 2) && p[0] == c0 && p[1] == c1)
267 #define ENT3(p, c0, c1, c2) (FITS (p, 3) && p[0]==c0 && p[1]==c1 && p[2]==c2)
269 /* Increment P by INC chars. If P lands at a semicolon, increment it
270 past the semicolon. This ensures that e.g. "<foo" is converted
271 to "<foo", but "<,foo" to "<,foo". */
272 #define SKIP_SEMI(p, inc) (p += inc, p < end && *p == ';' ? ++p : p)
274 struct tagstack_item {
275 const char *tagname_begin;
276 const char *tagname_end;
277 const char *contents_begin;
278 struct tagstack_item *prev;
279 struct tagstack_item *next;
282 struct tagstack_item *
283 tagstack_push (struct tagstack_item **head, struct tagstack_item **tail)
285 struct tagstack_item *ts = xmalloc(sizeof(struct tagstack_item));
289 ts->prev = ts->next = NULL;
302 /* remove ts and everything after it from the stack */
304 tagstack_pop (struct tagstack_item **head, struct tagstack_item **tail,
305 struct tagstack_item *ts)
315 *head = *tail = NULL;
319 ts->prev->next = NULL;
334 ts->prev->next = NULL;
338 struct tagstack_item *p = ts->next;
345 struct tagstack_item *
346 tagstack_find (struct tagstack_item *tail, const char *tagname_begin,
347 const char *tagname_end)
349 int len = tagname_end - tagname_begin;
352 if (len == (tail->tagname_end - tail->tagname_begin))
354 if (0 == strncasecmp (tail->tagname_begin, tagname_begin, len))
362 /* Decode the HTML character entity at *PTR, considering END to be end
363 of buffer. It is assumed that the "&" character that marks the
364 beginning of the entity has been seen at *PTR-1. If a recognized
365 ASCII entity is seen, it is returned, and *PTR is moved to the end
366 of the entity. Otherwise, -1 is returned and *PTR left unmodified.
368 The recognized entities are: <, >, &, &apos, and ". */
371 decode_entity (const char **ptr, const char *end)
373 const char *p = *ptr;
382 /* Process numeric entities "&#DDD;" and "&#xHH;". */
387 for (++p; value < 256 && p < end && ISXDIGIT (*p); p++, digits++)
388 value = (value << 4) + XDIGIT_TO_NUM (*p);
390 for (; value < 256 && p < end && ISDIGIT (*p); p++, digits++)
391 value = (value * 10) + (*p - '0');
394 /* Don't interpret 128+ codes and NUL because we cannot
395 portably reinserted them into HTML. */
396 if (!value || (value & ~0x7f))
398 *ptr = SKIP_SEMI (p, 0);
401 /* Process named ASCII entities. */
404 value = '>', *ptr = SKIP_SEMI (p, 1);
408 value = '<', *ptr = SKIP_SEMI (p, 1);
411 if (ENT2 (p, 'm', 'p'))
412 value = '&', *ptr = SKIP_SEMI (p, 2);
413 else if (ENT3 (p, 'p', 'o', 's'))
414 /* handle &apos for the sake of the XML/XHTML crowd. */
415 value = '\'', *ptr = SKIP_SEMI (p, 3);
418 if (ENT3 (p, 'u', 'o', 't'))
419 value = '\"', *ptr = SKIP_SEMI (p, 3);
432 AP_DECODE_ENTITIES = 2,
436 /* Copy the text in the range [BEG, END) to POOL, optionally
437 performing operations specified by FLAGS. FLAGS may be any
438 combination of AP_DOWNCASE, AP_DECODE_ENTITIES and AP_TRIM_BLANKS
439 with the following meaning:
441 * AP_DOWNCASE -- downcase all the letters;
443 * AP_DECODE_ENTITIES -- decode the named and numeric entities in
444 the ASCII range when copying the string.
446 * AP_TRIM_BLANKS -- ignore blanks at the beginning and at the end
447 of text, as well as embedded newlines. */
450 convert_and_copy (struct pool *pool, const char *beg, const char *end, int flags)
452 int old_tail = pool->tail;
454 /* Skip blanks if required. We must do this before entities are
455 processed, so that blanks can still be inserted as, for instance,
457 if (flags & AP_TRIM_BLANKS)
459 while (beg < end && ISSPACE (*beg))
461 while (end > beg && ISSPACE (end[-1]))
465 if (flags & AP_DECODE_ENTITIES)
467 /* Grow the pool, then copy the text to the pool character by
468 character, processing the encountered entities as we go
471 It's safe (and necessary) to grow the pool in advance because
472 processing the entities can only *shorten* the string, it can
473 never lengthen it. */
474 const char *from = beg;
476 bool squash_newlines = !!(flags & AP_TRIM_BLANKS);
478 POOL_GROW (pool, end - beg);
479 to = pool->contents + pool->tail;
485 int entity = decode_entity (&from, end);
491 else if ((*from == '\n' || *from == '\r') && squash_newlines)
496 /* Verify that we haven't exceeded the original size. (It
497 shouldn't happen, hence the assert.) */
498 assert (to - (pool->contents + pool->tail) <= end - beg);
500 /* Make POOL's tail point to the position following the string
502 pool->tail = to - pool->contents;
503 POOL_APPEND_CHR (pool, '\0');
507 /* Just copy the text to the pool. */
508 POOL_APPEND (pool, beg, end);
509 POOL_APPEND_CHR (pool, '\0');
512 if (flags & AP_DOWNCASE)
514 char *p = pool->contents + old_tail;
520 /* Originally we used to adhere to rfc 1866 here, and allowed only
521 letters, digits, periods, and hyphens as names (of tags or
522 attributes). However, this broke too many pages which used
523 proprietary or strange attributes, e.g. <img src="a.gif"
524 v:shapes="whatever">.
526 So now we allow any character except:
528 * 8-bit and control chars
529 * characters that clearly cannot be part of name:
532 This only affects attribute and tag names; attribute values allow
533 an even greater variety of characters. */
535 #define NAME_CHAR_P(x) ((x) > 32 && (x) < 127 \
536 && (x) != '=' && (x) != '>' && (x) != '/')
539 static int comment_backout_count;
542 /* Advance over an SGML declaration, such as <!DOCTYPE ...>. In
543 strict comments mode, this is used for skipping over comments as
546 To recap: any SGML declaration may have comments associated with
548 <!MY-DECL -- isn't this fun? -- foo bar>
550 An HTML comment is merely an empty declaration (<!>) with a comment
552 <!-- some stuff here -->
554 Several comments may be embedded in one comment declaration:
555 <!-- have -- -- fun -->
557 Whitespace is allowed between and after the comments, but not
558 before the first comment. Additionally, this function attempts to
559 handle double quotes in SGML declarations correctly. */
562 advance_declaration (const char *beg, const char *end)
565 char quote_char = '\0'; /* shut up, gcc! */
588 /* It looked like a good idea to write this as a state machine, but
591 while (state != AC_S_DONE && state != AC_S_BACKOUT)
594 state = AC_S_BACKOUT;
604 state = AC_S_DEFAULT;
607 state = AC_S_BACKOUT;
629 if (NAME_CHAR_P (ch))
630 state = AC_S_DCLNAME;
632 state = AC_S_BACKOUT;
639 else if (NAME_CHAR_P (ch))
642 state = AC_S_DEFAULT;
645 /* We must use 0x22 because broken assert macros choke on
647 assert (ch == '\'' || ch == 0x22);
648 quote_char = ch; /* cheating -- I really don't feel like
649 introducing more different states for
650 different quote characters. */
652 state = AC_S_IN_QUOTE;
655 if (ch == quote_char)
661 assert (ch == quote_char);
663 state = AC_S_DEFAULT;
675 state = AC_S_COMMENT;
678 state = AC_S_BACKOUT;
702 state = AC_S_DEFAULT;
705 state = AC_S_COMMENT;
712 if (state == AC_S_BACKOUT)
715 ++comment_backout_count;
722 /* Find the first occurrence of the substring "-->" in [BEG, END) and
723 return the pointer to the character after the substring. If the
724 substring is not found, return NULL. */
727 find_comment_end (const char *beg, const char *end)
729 /* Open-coded Boyer-Moore search for "-->". Examine the third char;
730 if it's not '>' or '-', advance by three characters. Otherwise,
731 look at the preceding characters and try to find a match. */
733 const char *p = beg - 1;
735 while ((p += 3) < end)
739 if (p[-1] == '-' && p[-2] == '-')
747 if (++p == end) return NULL;
750 case '>': return p + 1;
751 case '-': goto at_dash_dash;
756 if ((p += 2) >= end) return NULL;
771 /* Return true if the string containing of characters inside [b, e) is
772 present in hash table HT. */
775 name_allowed (const struct hash_table *ht, const char *b, const char *e)
780 BOUNDED_TO_ALLOCA (b, e, copy);
781 return hash_table_get (ht, copy) != NULL;
784 /* Advance P (a char pointer), with the explicit intent of being able
785 to read the next character. If this is not possible, go to finish. */
787 #define ADVANCE(p) do { \
793 /* Skip whitespace, if any. */
795 #define SKIP_WS(p) do { \
796 while (ISSPACE (*p)) { \
801 /* Skip non-whitespace, if any. */
803 #define SKIP_NON_WS(p) do { \
804 while (!ISSPACE (*p)) { \
810 static int tag_backout_count;
813 /* Map MAPFUN over HTML tags in TEXT, which is SIZE characters long.
814 MAPFUN will be called with two arguments: pointer to an initialized
815 struct taginfo, and MAPARG.
817 ALLOWED_TAGS and ALLOWED_ATTRIBUTES are hash tables the keys of
818 which are the tags and attribute names that this function should
819 use. If ALLOWED_TAGS is NULL, all tags are processed; if
820 ALLOWED_ATTRIBUTES is NULL, all attributes are returned.
822 (Obviously, the caller can filter out unwanted tags and attributes
823 just as well, but this is just an optimization designed to avoid
824 unnecessary copying of tags/attributes which the caller doesn't
828 map_html_tags (const char *text, int size,
829 void (*mapfun) (struct taginfo *, void *), void *maparg,
831 const struct hash_table *allowed_tags,
832 const struct hash_table *allowed_attributes)
834 /* storage for strings passed to MAPFUN callback; if 256 bytes is
835 too little, POOL_APPEND allocates more with malloc. */
836 char pool_initial_storage[256];
839 const char *p = text;
840 const char *end = text + size;
842 struct attr_pair attr_pair_initial_storage[8];
843 int attr_pair_size = countof (attr_pair_initial_storage);
844 bool attr_pair_resized = false;
845 struct attr_pair *pairs = attr_pair_initial_storage;
847 struct tagstack_item *head = NULL;
848 struct tagstack_item *tail = NULL;
853 POOL_INIT (&pool, pool_initial_storage, countof (pool_initial_storage));
857 const char *tag_name_begin, *tag_name_end;
858 const char *tag_start_position;
859 bool uninteresting_tag;
867 /* Find beginning of tag. We use memchr() instead of the usual
868 looping with ADVANCE() for speed. */
869 p = memchr (p, '<', end - p);
873 tag_start_position = p;
876 /* Establish the type of the tag (start-tag, end-tag or
880 if (!(flags & MHT_STRICT_COMMENTS)
881 && p < end + 3 && p[1] == '-' && p[2] == '-')
883 /* If strict comments are not enforced and if we know
884 we're looking at a comment, simply look for the
885 terminating "-->". Non-strict is the default because
886 it works in other browsers and most HTML writers can't
887 be bothered with getting the comments right. */
888 const char *comment_end = find_comment_end (p + 3, end);
894 /* Either in strict comment mode or looking at a non-empty
895 declaration. Real declarations are much less likely to
896 be misused the way comments are, so advance over them
897 properly regardless of strictness. */
898 p = advance_declaration (p, end);
910 while (NAME_CHAR_P (*p))
912 if (p == tag_name_begin)
919 struct tagstack_item *ts = tagstack_push (&head, &tail);
922 ts->tagname_begin = tag_name_begin;
923 ts->tagname_end = tag_name_end;
924 ts->contents_begin = NULL;
928 if (end_tag && *p != '>')
931 if (!name_allowed (allowed_tags, tag_name_begin, tag_name_end))
932 /* We can't just say "goto look_for_tag" here because we need
933 the loop below to properly advance over the tag's attributes. */
934 uninteresting_tag = true;
937 uninteresting_tag = false;
938 convert_and_copy (&pool, tag_name_begin, tag_name_end, AP_DOWNCASE);
941 /* Find the attributes. */
944 const char *attr_name_begin, *attr_name_end;
945 const char *attr_value_begin, *attr_value_end;
946 const char *attr_raw_value_begin, *attr_raw_value_end;
947 int operation = AP_DOWNCASE; /* stupid compiler. */
953 /* A slash at this point means the tag is about to be
954 closed. This is legal in XML and has been popularized
955 in HTML via XHTML. */
956 /* <foo a=b c=d /> */
964 /* Check for end of tag definition. */
968 /* Establish bounds of attribute name. */
969 attr_name_begin = p; /* <foo bar ...> */
971 while (NAME_CHAR_P (*p))
973 attr_name_end = p; /* <foo bar ...> */
975 if (attr_name_begin == attr_name_end)
978 /* Establish bounds of attribute value. */
980 if (NAME_CHAR_P (*p) || *p == '/' || *p == '>')
982 /* Minimized attribute syntax allows `=' to be omitted.
983 For example, <UL COMPACT> is a valid shorthand for <UL
984 COMPACT="compact">. Even if such attributes are not
985 useful to Wget, we need to support them, so that the
986 tags containing them can be parsed correctly. */
987 attr_raw_value_begin = attr_value_begin = attr_name_begin;
988 attr_raw_value_end = attr_value_end = attr_name_end;
994 if (*p == '\"' || *p == '\'')
996 bool newline_seen = false;
997 char quote_char = *p;
998 attr_raw_value_begin = p;
1000 attr_value_begin = p; /* <foo bar="baz"> */
1002 while (*p != quote_char)
1004 if (!newline_seen && *p == '\n')
1006 /* If a newline is seen within the quotes, it
1007 is most likely that someone forgot to close
1008 the quote. In that case, we back out to
1009 the value beginning, and terminate the tag
1010 at either `>' or the delimiter, whichever
1011 comes first. Such a tag terminated at `>'
1013 p = attr_value_begin;
1014 newline_seen = true;
1017 else if (newline_seen && *p == '>')
1021 attr_value_end = p; /* <foo bar="baz"> */
1023 if (*p == quote_char)
1027 attr_raw_value_end = p; /* <foo bar="baz"> */
1029 operation = AP_DECODE_ENTITIES;
1030 if (flags & MHT_TRIM_VALUES)
1031 operation |= AP_TRIM_BLANKS;
1035 attr_value_begin = p; /* <foo bar=baz> */
1037 /* According to SGML, a name token should consist only
1038 of alphanumerics, . and -. However, this is often
1039 violated by, for instance, `%' in `width=75%'.
1040 We'll be liberal and allow just about anything as
1041 an attribute value. */
1042 while (!ISSPACE (*p) && *p != '>')
1044 attr_value_end = p; /* <foo bar=baz qux=quix> */
1046 if (attr_value_begin == attr_value_end)
1050 attr_raw_value_begin = attr_value_begin;
1051 attr_raw_value_end = attr_value_end;
1052 operation = AP_DECODE_ENTITIES;
1057 /* We skipped the whitespace and found something that is
1058 neither `=' nor the beginning of the next attribute's
1060 goto backout_tag; /* <foo bar [... */
1064 /* If we're not interested in the tag, don't bother with any
1065 of the attributes. */
1066 if (uninteresting_tag)
1069 /* If we aren't interested in the attribute, skip it. We
1070 cannot do this test any sooner, because our text pointer
1071 needs to correctly advance over the attribute. */
1072 if (!name_allowed (allowed_attributes, attr_name_begin, attr_name_end))
1075 GROW_ARRAY (pairs, attr_pair_size, nattrs + 1, attr_pair_resized,
1078 pairs[nattrs].name_pool_index = pool.tail;
1079 convert_and_copy (&pool, attr_name_begin, attr_name_end, AP_DOWNCASE);
1081 pairs[nattrs].value_pool_index = pool.tail;
1082 convert_and_copy (&pool, attr_value_begin, attr_value_end, operation);
1083 pairs[nattrs].value_raw_beginning = attr_raw_value_begin;
1084 pairs[nattrs].value_raw_size = (attr_raw_value_end
1085 - attr_raw_value_begin);
1089 if (!end_tag && tail && (tail->tagname_begin == tag_name_begin))
1091 tail->contents_begin = p+1;
1094 if (uninteresting_tag)
1100 /* By now, we have a valid tag with a name and zero or more
1101 attributes. Fill in the data and call the mapper function. */
1104 struct taginfo taginfo;
1105 struct tagstack_item *ts = NULL;
1107 taginfo.name = pool.contents;
1108 taginfo.end_tag_p = end_tag;
1109 taginfo.nattrs = nattrs;
1110 /* We fill in the char pointers only now, when pool can no
1111 longer get realloc'ed. If we did that above, we could get
1112 hosed by reallocation. Obviously, after this point, the pool
1113 may no longer be grown. */
1114 for (i = 0; i < nattrs; i++)
1116 pairs[i].name = pool.contents + pairs[i].name_pool_index;
1117 pairs[i].value = pool.contents + pairs[i].value_pool_index;
1119 taginfo.attrs = pairs;
1120 taginfo.start_position = tag_start_position;
1121 taginfo.end_position = p + 1;
1122 taginfo.contents_begin = NULL;
1123 taginfo.contents_end = NULL;
1127 ts = tagstack_find (tail, tag_name_begin, tag_name_end);
1130 if (ts->contents_begin)
1132 taginfo.contents_begin = ts->contents_begin;
1133 taginfo.contents_end = tag_start_position;
1135 tagstack_pop (&head, &tail, ts);
1139 mapfun (&taginfo, maparg);
1146 ++tag_backout_count;
1148 /* The tag wasn't really a tag. Treat its contents as ordinary
1150 p = tag_start_position + 1;
1156 if (attr_pair_resized)
1158 /* pop any tag stack that's left */
1159 tagstack_pop (&head, &tail, head);
1168 test_mapper (struct taginfo *taginfo, void *arg)
1172 printf ("%s%s", taginfo->end_tag_p ? "/" : "", taginfo->name);
1173 for (i = 0; i < taginfo->nattrs; i++)
1174 printf (" %s=%s", taginfo->attrs[i].name, taginfo->attrs[i].value);
1182 char *x = xmalloc (size);
1185 int tag_counter = 0;
1187 while ((read_count = fread (x + length, 1, size - length, stdin)))
1189 length += read_count;
1191 x = xrealloc (x, size);
1194 map_html_tags (x, length, test_mapper, &tag_counter, 0, NULL, NULL);
1195 printf ("TAGS: %d\n", tag_counter);
1196 printf ("Tag backouts: %d\n", tag_backout_count);
1197 printf ("Comment backouts: %d\n", comment_backout_count);
1200 #endif /* STANDALONE */