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
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 /* Decode the HTML character entity at *PTR, considering END to be end
275 of buffer. It is assumed that the "&" character that marks the
276 beginning of the entity has been seen at *PTR-1. If a recognized
277 ASCII entity is seen, it is returned, and *PTR is moved to the end
278 of the entity. Otherwise, -1 is returned and *PTR left unmodified.
280 The recognized entities are: <, >, &, &apos, and ". */
283 decode_entity (const char **ptr, const char *end)
285 const char *p = *ptr;
294 /* Process numeric entities "&#DDD;" and "&#xHH;". */
299 for (++p; value < 256 && p < end && ISXDIGIT (*p); p++, digits++)
300 value = (value << 4) + XDIGIT_TO_NUM (*p);
302 for (; value < 256 && p < end && ISDIGIT (*p); p++, digits++)
303 value = (value * 10) + (*p - '0');
306 /* Don't interpret 128+ codes and NUL because we cannot
307 portably reinserted them into HTML. */
308 if (!value || (value & ~0x7f))
310 *ptr = SKIP_SEMI (p, 0);
313 /* Process named ASCII entities. */
316 value = '>', *ptr = SKIP_SEMI (p, 1);
320 value = '<', *ptr = SKIP_SEMI (p, 1);
323 if (ENT2 (p, 'm', 'p'))
324 value = '&', *ptr = SKIP_SEMI (p, 2);
325 else if (ENT3 (p, 'p', 'o', 's'))
326 /* handle &apos for the sake of the XML/XHTML crowd. */
327 value = '\'', *ptr = SKIP_SEMI (p, 3);
330 if (ENT3 (p, 'u', 'o', 't'))
331 value = '\"', *ptr = SKIP_SEMI (p, 3);
344 AP_DECODE_ENTITIES = 2,
348 /* Copy the text in the range [BEG, END) to POOL, optionally
349 performing operations specified by FLAGS. FLAGS may be any
350 combination of AP_DOWNCASE, AP_DECODE_ENTITIES and AP_TRIM_BLANKS
351 with the following meaning:
353 * AP_DOWNCASE -- downcase all the letters;
355 * AP_DECODE_ENTITIES -- decode the named and numeric entities in
356 the ASCII range when copying the string.
358 * AP_TRIM_BLANKS -- ignore blanks at the beginning and at the end
359 of text, as well as embedded newlines. */
362 convert_and_copy (struct pool *pool, const char *beg, const char *end, int flags)
364 int old_tail = pool->tail;
366 /* Skip blanks if required. We must do this before entities are
367 processed, so that blanks can still be inserted as, for instance,
369 if (flags & AP_TRIM_BLANKS)
371 while (beg < end && ISSPACE (*beg))
373 while (end > beg && ISSPACE (end[-1]))
377 if (flags & AP_DECODE_ENTITIES)
379 /* Grow the pool, then copy the text to the pool character by
380 character, processing the encountered entities as we go
383 It's safe (and necessary) to grow the pool in advance because
384 processing the entities can only *shorten* the string, it can
385 never lengthen it. */
386 const char *from = beg;
388 bool squash_newlines = !!(flags & AP_TRIM_BLANKS);
390 POOL_GROW (pool, end - beg);
391 to = pool->contents + pool->tail;
397 int entity = decode_entity (&from, end);
403 else if ((*from == '\n' || *from == '\r') && squash_newlines)
408 /* Verify that we haven't exceeded the original size. (It
409 shouldn't happen, hence the assert.) */
410 assert (to - (pool->contents + pool->tail) <= end - beg);
412 /* Make POOL's tail point to the position following the string
414 pool->tail = to - pool->contents;
415 POOL_APPEND_CHR (pool, '\0');
419 /* Just copy the text to the pool. */
420 POOL_APPEND (pool, beg, end);
421 POOL_APPEND_CHR (pool, '\0');
424 if (flags & AP_DOWNCASE)
426 char *p = pool->contents + old_tail;
432 /* Originally we used to adhere to rfc 1866 here, and allowed only
433 letters, digits, periods, and hyphens as names (of tags or
434 attributes). However, this broke too many pages which used
435 proprietary or strange attributes, e.g. <img src="a.gif"
436 v:shapes="whatever">.
438 So now we allow any character except:
440 * 8-bit and control chars
441 * characters that clearly cannot be part of name:
444 This only affects attribute and tag names; attribute values allow
445 an even greater variety of characters. */
447 #define NAME_CHAR_P(x) ((x) > 32 && (x) < 127 \
448 && (x) != '=' && (x) != '>' && (x) != '/')
451 static int comment_backout_count;
454 /* Advance over an SGML declaration, such as <!DOCTYPE ...>. In
455 strict comments mode, this is used for skipping over comments as
458 To recap: any SGML declaration may have comments associated with
460 <!MY-DECL -- isn't this fun? -- foo bar>
462 An HTML comment is merely an empty declaration (<!>) with a comment
464 <!-- some stuff here -->
466 Several comments may be embedded in one comment declaration:
467 <!-- have -- -- fun -->
469 Whitespace is allowed between and after the comments, but not
470 before the first comment. Additionally, this function attempts to
471 handle double quotes in SGML declarations correctly. */
474 advance_declaration (const char *beg, const char *end)
477 char quote_char = '\0'; /* shut up, gcc! */
500 /* It looked like a good idea to write this as a state machine, but
503 while (state != AC_S_DONE && state != AC_S_BACKOUT)
506 state = AC_S_BACKOUT;
516 state = AC_S_DEFAULT;
519 state = AC_S_BACKOUT;
541 if (NAME_CHAR_P (ch))
542 state = AC_S_DCLNAME;
544 state = AC_S_BACKOUT;
551 else if (NAME_CHAR_P (ch))
554 state = AC_S_DEFAULT;
557 /* We must use 0x22 because broken assert macros choke on
559 assert (ch == '\'' || ch == 0x22);
560 quote_char = ch; /* cheating -- I really don't feel like
561 introducing more different states for
562 different quote characters. */
564 state = AC_S_IN_QUOTE;
567 if (ch == quote_char)
573 assert (ch == quote_char);
575 state = AC_S_DEFAULT;
587 state = AC_S_COMMENT;
590 state = AC_S_BACKOUT;
614 state = AC_S_DEFAULT;
617 state = AC_S_COMMENT;
624 if (state == AC_S_BACKOUT)
627 ++comment_backout_count;
634 /* Find the first occurrence of the substring "-->" in [BEG, END) and
635 return the pointer to the character after the substring. If the
636 substring is not found, return NULL. */
639 find_comment_end (const char *beg, const char *end)
641 /* Open-coded Boyer-Moore search for "-->". Examine the third char;
642 if it's not '>' or '-', advance by three characters. Otherwise,
643 look at the preceding characters and try to find a match. */
645 const char *p = beg - 1;
647 while ((p += 3) < end)
651 if (p[-1] == '-' && p[-2] == '-')
659 if (++p == end) return NULL;
662 case '>': return p + 1;
663 case '-': goto at_dash_dash;
668 if ((p += 2) >= end) return NULL;
683 /* Return true if the string containing of characters inside [b, e) is
684 present in hash table HT. */
687 name_allowed (const struct hash_table *ht, const char *b, const char *e)
692 BOUNDED_TO_ALLOCA (b, e, copy);
693 return hash_table_get (ht, copy) != NULL;
696 /* Advance P (a char pointer), with the explicit intent of being able
697 to read the next character. If this is not possible, go to finish. */
699 #define ADVANCE(p) do { \
705 /* Skip whitespace, if any. */
707 #define SKIP_WS(p) do { \
708 while (ISSPACE (*p)) { \
713 /* Skip non-whitespace, if any. */
715 #define SKIP_NON_WS(p) do { \
716 while (!ISSPACE (*p)) { \
722 static int tag_backout_count;
725 /* Map MAPFUN over HTML tags in TEXT, which is SIZE characters long.
726 MAPFUN will be called with two arguments: pointer to an initialized
727 struct taginfo, and MAPARG.
729 ALLOWED_TAGS and ALLOWED_ATTRIBUTES are hash tables the keys of
730 which are the tags and attribute names that this function should
731 use. If ALLOWED_TAGS is NULL, all tags are processed; if
732 ALLOWED_ATTRIBUTES is NULL, all attributes are returned.
734 (Obviously, the caller can filter out unwanted tags and attributes
735 just as well, but this is just an optimization designed to avoid
736 unnecessary copying of tags/attributes which the caller doesn't
740 map_html_tags (const char *text, int size,
741 void (*mapfun) (struct taginfo *, void *), void *maparg,
743 const struct hash_table *allowed_tags,
744 const struct hash_table *allowed_attributes)
746 /* storage for strings passed to MAPFUN callback; if 256 bytes is
747 too little, POOL_APPEND allocates more with malloc. */
748 char pool_initial_storage[256];
751 const char *p = text;
752 const char *end = text + size;
754 struct attr_pair attr_pair_initial_storage[8];
755 int attr_pair_size = countof (attr_pair_initial_storage);
756 bool attr_pair_resized = false;
757 struct attr_pair *pairs = attr_pair_initial_storage;
762 POOL_INIT (&pool, pool_initial_storage, countof (pool_initial_storage));
766 const char *tag_name_begin, *tag_name_end;
767 const char *tag_start_position;
768 bool uninteresting_tag;
776 /* Find beginning of tag. We use memchr() instead of the usual
777 looping with ADVANCE() for speed. */
778 p = memchr (p, '<', end - p);
782 tag_start_position = p;
785 /* Establish the type of the tag (start-tag, end-tag or
789 if (!(flags & MHT_STRICT_COMMENTS)
790 && p < end + 3 && p[1] == '-' && p[2] == '-')
792 /* If strict comments are not enforced and if we know
793 we're looking at a comment, simply look for the
794 terminating "-->". Non-strict is the default because
795 it works in other browsers and most HTML writers can't
796 be bothered with getting the comments right. */
797 const char *comment_end = find_comment_end (p + 3, end);
803 /* Either in strict comment mode or looking at a non-empty
804 declaration. Real declarations are much less likely to
805 be misused the way comments are, so advance over them
806 properly regardless of strictness. */
807 p = advance_declaration (p, end);
819 while (NAME_CHAR_P (*p))
821 if (p == tag_name_begin)
825 if (end_tag && *p != '>')
828 if (!name_allowed (allowed_tags, tag_name_begin, tag_name_end))
829 /* We can't just say "goto look_for_tag" here because we need
830 the loop below to properly advance over the tag's attributes. */
831 uninteresting_tag = true;
834 uninteresting_tag = false;
835 convert_and_copy (&pool, tag_name_begin, tag_name_end, AP_DOWNCASE);
838 /* Find the attributes. */
841 const char *attr_name_begin, *attr_name_end;
842 const char *attr_value_begin, *attr_value_end;
843 const char *attr_raw_value_begin, *attr_raw_value_end;
844 int operation = AP_DOWNCASE; /* stupid compiler. */
850 /* A slash at this point means the tag is about to be
851 closed. This is legal in XML and has been popularized
852 in HTML via XHTML. */
853 /* <foo a=b c=d /> */
861 /* Check for end of tag definition. */
865 /* Establish bounds of attribute name. */
866 attr_name_begin = p; /* <foo bar ...> */
868 while (NAME_CHAR_P (*p))
870 attr_name_end = p; /* <foo bar ...> */
872 if (attr_name_begin == attr_name_end)
875 /* Establish bounds of attribute value. */
877 if (NAME_CHAR_P (*p) || *p == '/' || *p == '>')
879 /* Minimized attribute syntax allows `=' to be omitted.
880 For example, <UL COMPACT> is a valid shorthand for <UL
881 COMPACT="compact">. Even if such attributes are not
882 useful to Wget, we need to support them, so that the
883 tags containing them can be parsed correctly. */
884 attr_raw_value_begin = attr_value_begin = attr_name_begin;
885 attr_raw_value_end = attr_value_end = attr_name_end;
891 if (*p == '\"' || *p == '\'')
893 bool newline_seen = false;
894 char quote_char = *p;
895 attr_raw_value_begin = p;
897 attr_value_begin = p; /* <foo bar="baz"> */
899 while (*p != quote_char)
901 if (!newline_seen && *p == '\n')
903 /* If a newline is seen within the quotes, it
904 is most likely that someone forgot to close
905 the quote. In that case, we back out to
906 the value beginning, and terminate the tag
907 at either `>' or the delimiter, whichever
908 comes first. Such a tag terminated at `>'
910 p = attr_value_begin;
914 else if (newline_seen && *p == '>')
918 attr_value_end = p; /* <foo bar="baz"> */
920 if (*p == quote_char)
924 attr_raw_value_end = p; /* <foo bar="baz"> */
926 operation = AP_DECODE_ENTITIES;
927 if (flags & MHT_TRIM_VALUES)
928 operation |= AP_TRIM_BLANKS;
932 attr_value_begin = p; /* <foo bar=baz> */
934 /* According to SGML, a name token should consist only
935 of alphanumerics, . and -. However, this is often
936 violated by, for instance, `%' in `width=75%'.
937 We'll be liberal and allow just about anything as
938 an attribute value. */
939 while (!ISSPACE (*p) && *p != '>')
941 attr_value_end = p; /* <foo bar=baz qux=quix> */
943 if (attr_value_begin == attr_value_end)
947 attr_raw_value_begin = attr_value_begin;
948 attr_raw_value_end = attr_value_end;
949 operation = AP_DECODE_ENTITIES;
954 /* We skipped the whitespace and found something that is
955 neither `=' nor the beginning of the next attribute's
957 goto backout_tag; /* <foo bar [... */
961 /* If we're not interested in the tag, don't bother with any
962 of the attributes. */
963 if (uninteresting_tag)
966 /* If we aren't interested in the attribute, skip it. We
967 cannot do this test any sooner, because our text pointer
968 needs to correctly advance over the attribute. */
969 if (!name_allowed (allowed_attributes, attr_name_begin, attr_name_end))
972 GROW_ARRAY (pairs, attr_pair_size, nattrs + 1, attr_pair_resized,
975 pairs[nattrs].name_pool_index = pool.tail;
976 convert_and_copy (&pool, attr_name_begin, attr_name_end, AP_DOWNCASE);
978 pairs[nattrs].value_pool_index = pool.tail;
979 convert_and_copy (&pool, attr_value_begin, attr_value_end, operation);
980 pairs[nattrs].value_raw_beginning = attr_raw_value_begin;
981 pairs[nattrs].value_raw_size = (attr_raw_value_end
982 - attr_raw_value_begin);
986 if (uninteresting_tag)
992 /* By now, we have a valid tag with a name and zero or more
993 attributes. Fill in the data and call the mapper function. */
996 struct taginfo taginfo;
998 taginfo.name = pool.contents;
999 taginfo.end_tag_p = end_tag;
1000 taginfo.nattrs = nattrs;
1001 /* We fill in the char pointers only now, when pool can no
1002 longer get realloc'ed. If we did that above, we could get
1003 hosed by reallocation. Obviously, after this point, the pool
1004 may no longer be grown. */
1005 for (i = 0; i < nattrs; i++)
1007 pairs[i].name = pool.contents + pairs[i].name_pool_index;
1008 pairs[i].value = pool.contents + pairs[i].value_pool_index;
1010 taginfo.attrs = pairs;
1011 taginfo.start_position = tag_start_position;
1012 taginfo.end_position = p + 1;
1013 mapfun (&taginfo, maparg);
1020 ++tag_backout_count;
1022 /* The tag wasn't really a tag. Treat its contents as ordinary
1024 p = tag_start_position + 1;
1030 if (attr_pair_resized)
1040 test_mapper (struct taginfo *taginfo, void *arg)
1044 printf ("%s%s", taginfo->end_tag_p ? "/" : "", taginfo->name);
1045 for (i = 0; i < taginfo->nattrs; i++)
1046 printf (" %s=%s", taginfo->attrs[i].name, taginfo->attrs[i].value);
1054 char *x = xmalloc (size);
1057 int tag_counter = 0;
1059 while ((read_count = fread (x + length, 1, size - length, stdin)))
1061 length += read_count;
1063 x = xrealloc (x, size);
1066 map_html_tags (x, length, test_mapper, &tag_counter, 0, NULL, NULL);
1067 printf ("TAGS: %d\n", tag_counter);
1068 printf ("Tag backouts: %d\n", tag_backout_count);
1069 printf ("Comment backouts: %d\n", comment_backout_count);
1072 #endif /* STANDALONE */