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 /* Test whether n+1-sized entity name fits in P. We don't support
261 IE-style non-terminated entities, e.g. "<foo" -> "<foo".
262 However, "<foo" will work, as will "<!foo", "<", etc. In
263 other words an entity needs to be terminated by either a
264 non-alphanumeric or the end of string. */
265 #define FITS(p, n) (p + n == end || (p + n < end && !ISALNUM (p[n])))
267 /* Macros that test entity names by returning true if P is followed by
268 the specified characters. */
269 #define ENT1(p, c0) (FITS (p, 1) && p[0] == c0)
270 #define ENT2(p, c0, c1) (FITS (p, 2) && p[0] == c0 && p[1] == c1)
271 #define ENT3(p, c0, c1, c2) (FITS (p, 3) && p[0]==c0 && p[1]==c1 && p[2]==c2)
273 /* Increment P by INC chars. If P lands at a semicolon, increment it
274 past the semicolon. This ensures that e.g. "<foo" is converted
275 to "<foo", but "<,foo" to "<,foo". */
276 #define SKIP_SEMI(p, inc) (p += inc, p < end && *p == ';' ? ++p : p)
278 /* Decode the HTML character entity at *PTR, considering END to be end
279 of buffer. It is assumed that the "&" character that marks the
280 beginning of the entity has been seen at *PTR-1. If a recognized
281 ASCII entity is seen, it is returned, and *PTR is moved to the end
282 of the entity. Otherwise, -1 is returned and *PTR left unmodified.
284 The recognized entities are: <, >, &, &apos, and ". */
287 decode_entity (const char **ptr, const char *end)
289 const char *p = *ptr;
298 /* Process numeric entities "&#DDD;" and "&#xHH;". */
303 for (++p; value < 256 && p < end && ISXDIGIT (*p); p++, digits++)
304 value = (value << 4) + XDIGIT_TO_NUM (*p);
306 for (; value < 256 && p < end && ISDIGIT (*p); p++, digits++)
307 value = (value * 10) + (*p - '0');
310 /* Don't interpret 128+ codes and NUL because we cannot
311 portably reinserted them into HTML. */
312 if (!value || (value & ~0x7f))
314 *ptr = SKIP_SEMI (p, 0);
317 /* Process named ASCII entities. */
320 value = '>', *ptr = SKIP_SEMI (p, 1);
324 value = '<', *ptr = SKIP_SEMI (p, 1);
327 if (ENT2 (p, 'm', 'p'))
328 value = '&', *ptr = SKIP_SEMI (p, 2);
329 else if (ENT3 (p, 'p', 'o', 's'))
330 /* handle &apos for the sake of the XML/XHTML crowd. */
331 value = '\'', *ptr = SKIP_SEMI (p, 3);
334 if (ENT3 (p, 'u', 'o', 't'))
335 value = '\"', *ptr = SKIP_SEMI (p, 3);
346 AP_DECODE_ENTITIES = 2,
350 /* Copy the text in the range [BEG, END) to POOL, optionally
351 performing operations specified by FLAGS. FLAGS may be any
352 combination of AP_DOWNCASE, AP_DECODE_ENTITIES and AP_TRIM_BLANKS
353 with the following meaning:
355 * AP_DOWNCASE -- downcase all the letters;
357 * AP_DECODE_ENTITIES -- decode the named and numeric entities in
358 the ASCII range when copying the string.
360 * AP_TRIM_BLANKS -- ignore blanks at the beginning and at the end
364 convert_and_copy (struct pool *pool, const char *beg, const char *end, int flags)
366 int old_tail = pool->tail;
369 /* First, skip blanks if required. We must do this before entities
370 are processed, so that blanks can still be inserted as, for
371 instance, ` '. */
372 if (flags & AP_TRIM_BLANKS)
374 while (beg < end && ISSPACE (*beg))
376 while (end > beg && ISSPACE (end[-1]))
381 if (flags & AP_DECODE_ENTITIES)
383 /* Grow the pool, then copy the text to the pool character by
384 character, processing the encountered entities as we go
387 It's safe (and necessary) to grow the pool in advance because
388 processing the entities can only *shorten* the string, it can
389 never lengthen it. */
390 const char *from = beg;
393 POOL_GROW (pool, end - beg);
394 to = pool->contents + pool->tail;
402 int entity = decode_entity (&from, end);
409 /* Verify that we haven't exceeded the original size. (It
410 shouldn't happen, hence the assert.) */
411 assert (to - (pool->contents + pool->tail) <= end - beg);
413 /* Make POOL's tail point to the position following the string
415 pool->tail = to - pool->contents;
416 POOL_APPEND_CHR (pool, '\0');
420 /* Just copy the text to the pool. */
421 POOL_APPEND (pool, beg, end);
422 POOL_APPEND_CHR (pool, '\0');
425 if (flags & AP_DOWNCASE)
427 char *p = pool->contents + old_tail;
433 /* Originally we used to adhere to rfc 1866 here, and allowed only
434 letters, digits, periods, and hyphens as names (of tags or
435 attributes). However, this broke too many pages which used
436 proprietary or strange attributes, e.g. <img src="a.gif"
437 v:shapes="whatever">.
439 So now we allow any character except:
441 * 8-bit and control chars
442 * characters that clearly cannot be part of name:
445 This only affects attribute and tag names; attribute values allow
446 an even greater variety of characters. */
448 #define NAME_CHAR_P(x) ((x) > 32 && (x) < 127 \
449 && (x) != '=' && (x) != '>' && (x) != '/')
452 static int comment_backout_count;
455 /* Advance over an SGML declaration, such as <!DOCTYPE ...>. In
456 strict comments mode, this is used for skipping over comments as
459 To recap: any SGML declaration may have comments associated with
461 <!MY-DECL -- isn't this fun? -- foo bar>
463 An HTML comment is merely an empty declaration (<!>) with a comment
465 <!-- some stuff here -->
467 Several comments may be embedded in one comment declaration:
468 <!-- have -- -- fun -->
470 Whitespace is allowed between and after the comments, but not
471 before the first comment. Additionally, this function attempts to
472 handle double quotes in SGML declarations correctly. */
475 advance_declaration (const char *beg, const char *end)
478 char quote_char = '\0'; /* shut up, gcc! */
501 /* It looked like a good idea to write this as a state machine, but
504 while (state != AC_S_DONE && state != AC_S_BACKOUT)
507 state = AC_S_BACKOUT;
517 state = AC_S_DEFAULT;
520 state = AC_S_BACKOUT;
542 if (NAME_CHAR_P (ch))
543 state = AC_S_DCLNAME;
545 state = AC_S_BACKOUT;
552 else if (NAME_CHAR_P (ch))
555 state = AC_S_DEFAULT;
558 /* We must use 0x22 because broken assert macros choke on
560 assert (ch == '\'' || ch == 0x22);
561 quote_char = ch; /* cheating -- I really don't feel like
562 introducing more different states for
563 different quote characters. */
565 state = AC_S_IN_QUOTE;
568 if (ch == quote_char)
574 assert (ch == quote_char);
576 state = AC_S_DEFAULT;
588 state = AC_S_COMMENT;
591 state = AC_S_BACKOUT;
615 state = AC_S_DEFAULT;
618 state = AC_S_COMMENT;
625 if (state == AC_S_BACKOUT)
628 ++comment_backout_count;
635 /* Find the first occurrence of the substring "-->" in [BEG, END) and
636 return the pointer to the character after the substring. If the
637 substring is not found, return NULL. */
640 find_comment_end (const char *beg, const char *end)
642 /* Open-coded Boyer-Moore search for "-->". Examine the third char;
643 if it's not '>' or '-', advance by three characters. Otherwise,
644 look at the preceding characters and try to find a match. */
646 const char *p = beg - 1;
648 while ((p += 3) < end)
652 if (p[-1] == '-' && p[-2] == '-')
660 if (++p == end) return NULL;
663 case '>': return p + 1;
664 case '-': goto at_dash_dash;
669 if ((p += 2) >= end) return NULL;
684 /* Return non-zero of the string inside [b, e) are present in hash
688 name_allowed (const struct hash_table *ht, const char *b, const char *e)
693 BOUNDED_TO_ALLOCA (b, e, copy);
694 return hash_table_get (ht, copy) != NULL;
697 /* Advance P (a char pointer), with the explicit intent of being able
698 to read the next character. If this is not possible, go to finish. */
700 #define ADVANCE(p) do { \
706 /* Skip whitespace, if any. */
708 #define SKIP_WS(p) do { \
709 while (ISSPACE (*p)) { \
714 /* Skip non-whitespace, if any. */
716 #define SKIP_NON_WS(p) do { \
717 while (!ISSPACE (*p)) { \
723 static int tag_backout_count;
726 /* Map MAPFUN over HTML tags in TEXT, which is SIZE characters long.
727 MAPFUN will be called with two arguments: pointer to an initialized
728 struct taginfo, and MAPARG.
730 ALLOWED_TAG_NAMES should be a NULL-terminated array of tag names to
731 be processed by this function. If it is NULL, all the tags are
732 allowed. The same goes for attributes and ALLOWED_ATTRIBUTE_NAMES.
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 for tags/attributes which the caller doesn't
737 want to know about. These lists are searched linearly; therefore,
738 if you're interested in a large number of tags or attributes, you'd
739 better set these to NULL and filter them out yourself with a
740 hashing process most appropriate for your application.) */
743 map_html_tags (const char *text, int size,
744 void (*mapfun) (struct taginfo *, void *), void *maparg,
746 const struct hash_table *allowed_tags,
747 const struct hash_table *allowed_attributes)
749 /* storage for strings passed to MAPFUN callback; if 256 bytes is
750 too little, POOL_APPEND allocates more with malloc. */
751 char pool_initial_storage[256];
754 const char *p = text;
755 const char *end = text + size;
757 struct attr_pair attr_pair_initial_storage[8];
758 int attr_pair_size = countof (attr_pair_initial_storage);
759 int attr_pair_resized = 0;
760 struct attr_pair *pairs = attr_pair_initial_storage;
765 POOL_INIT (&pool, pool_initial_storage, countof (pool_initial_storage));
769 const char *tag_name_begin, *tag_name_end;
770 const char *tag_start_position;
771 int uninteresting_tag;
779 /* Find beginning of tag. We use memchr() instead of the usual
780 looping with ADVANCE() for speed. */
781 p = memchr (p, '<', end - p);
785 tag_start_position = p;
788 /* Establish the type of the tag (start-tag, end-tag or
792 if (!(flags & MHT_STRICT_COMMENTS)
793 && p < end + 3 && p[1] == '-' && p[2] == '-')
795 /* If strict comments are not enforced and if we know
796 we're looking at a comment, simply look for the
797 terminating "-->". Non-strict is the default because
798 it works in other browsers and most HTML writers can't
799 be bothered with getting the comments right. */
800 const char *comment_end = find_comment_end (p + 3, end);
806 /* Either in strict comment mode or looking at a non-empty
807 declaration. Real declarations are much less likely to
808 be misused the way comments are, so advance over them
809 properly regardless of strictness. */
810 p = advance_declaration (p, end);
822 while (NAME_CHAR_P (*p))
824 if (p == tag_name_begin)
828 if (end_tag && *p != '>')
831 if (!name_allowed (allowed_tags, tag_name_begin, tag_name_end))
832 /* We can't just say "goto look_for_tag" here because we need
833 the loop below to properly advance over the tag's attributes. */
834 uninteresting_tag = 1;
837 uninteresting_tag = 0;
838 convert_and_copy (&pool, tag_name_begin, tag_name_end, AP_DOWNCASE);
841 /* Find the attributes. */
844 const char *attr_name_begin, *attr_name_end;
845 const char *attr_value_begin, *attr_value_end;
846 const char *attr_raw_value_begin, *attr_raw_value_end;
847 int operation = AP_DOWNCASE; /* stupid compiler. */
853 /* A slash at this point means the tag is about to be
854 closed. This is legal in XML and has been popularized
855 in HTML via XHTML. */
856 /* <foo a=b c=d /> */
864 /* Check for end of tag definition. */
868 /* Establish bounds of attribute name. */
869 attr_name_begin = p; /* <foo bar ...> */
871 while (NAME_CHAR_P (*p))
873 attr_name_end = p; /* <foo bar ...> */
875 if (attr_name_begin == attr_name_end)
878 /* Establish bounds of attribute value. */
880 if (NAME_CHAR_P (*p) || *p == '/' || *p == '>')
882 /* Minimized attribute syntax allows `=' to be omitted.
883 For example, <UL COMPACT> is a valid shorthand for <UL
884 COMPACT="compact">. Even if such attributes are not
885 useful to Wget, we need to support them, so that the
886 tags containing them can be parsed correctly. */
887 attr_raw_value_begin = attr_value_begin = attr_name_begin;
888 attr_raw_value_end = attr_value_end = attr_name_end;
894 if (*p == '\"' || *p == '\'')
896 int newline_seen = 0;
897 char quote_char = *p;
898 attr_raw_value_begin = p;
900 attr_value_begin = p; /* <foo bar="baz"> */
902 while (*p != quote_char)
904 if (!newline_seen && *p == '\n')
906 /* If a newline is seen within the quotes, it
907 is most likely that someone forgot to close
908 the quote. In that case, we back out to
909 the value beginning, and terminate the tag
910 at either `>' or the delimiter, whichever
911 comes first. Such a tag terminated at `>'
913 p = attr_value_begin;
917 else if (newline_seen && *p == '>')
921 attr_value_end = p; /* <foo bar="baz"> */
923 if (*p == quote_char)
927 attr_raw_value_end = p; /* <foo bar="baz"> */
929 operation = AP_DECODE_ENTITIES;
930 if (flags & MHT_TRIM_VALUES)
931 operation |= AP_TRIM_BLANKS;
935 attr_value_begin = p; /* <foo bar=baz> */
937 /* According to SGML, a name token should consist only
938 of alphanumerics, . and -. However, this is often
939 violated by, for instance, `%' in `width=75%'.
940 We'll be liberal and allow just about anything as
941 an attribute value. */
942 while (!ISSPACE (*p) && *p != '>')
944 attr_value_end = p; /* <foo bar=baz qux=quix> */
946 if (attr_value_begin == attr_value_end)
950 attr_raw_value_begin = attr_value_begin;
951 attr_raw_value_end = attr_value_end;
952 operation = AP_DECODE_ENTITIES;
957 /* We skipped the whitespace and found something that is
958 neither `=' nor the beginning of the next attribute's
960 goto backout_tag; /* <foo bar [... */
964 /* If we're not interested in the tag, don't bother with any
965 of the attributes. */
966 if (uninteresting_tag)
969 /* If we aren't interested in the attribute, skip it. We
970 cannot do this test any sooner, because our text pointer
971 needs to correctly advance over the attribute. */
972 if (!name_allowed (allowed_attributes, attr_name_begin, attr_name_end))
975 GROW_ARRAY (pairs, attr_pair_size, nattrs + 1, attr_pair_resized,
978 pairs[nattrs].name_pool_index = pool.tail;
979 convert_and_copy (&pool, attr_name_begin, attr_name_end, AP_DOWNCASE);
981 pairs[nattrs].value_pool_index = pool.tail;
982 convert_and_copy (&pool, attr_value_begin, attr_value_end, operation);
983 pairs[nattrs].value_raw_beginning = attr_raw_value_begin;
984 pairs[nattrs].value_raw_size = (attr_raw_value_end
985 - attr_raw_value_begin);
989 if (uninteresting_tag)
995 /* By now, we have a valid tag with a name and zero or more
996 attributes. Fill in the data and call the mapper function. */
999 struct taginfo taginfo;
1001 taginfo.name = pool.contents;
1002 taginfo.end_tag_p = end_tag;
1003 taginfo.nattrs = nattrs;
1004 /* We fill in the char pointers only now, when pool can no
1005 longer get realloc'ed. If we did that above, we could get
1006 hosed by reallocation. Obviously, after this point, the pool
1007 may no longer be grown. */
1008 for (i = 0; i < nattrs; i++)
1010 pairs[i].name = pool.contents + pairs[i].name_pool_index;
1011 pairs[i].value = pool.contents + pairs[i].value_pool_index;
1013 taginfo.attrs = pairs;
1014 taginfo.start_position = tag_start_position;
1015 taginfo.end_position = p + 1;
1017 (*mapfun) (&taginfo, maparg);
1024 ++tag_backout_count;
1026 /* The tag wasn't really a tag. Treat its contents as ordinary
1028 p = tag_start_position + 1;
1034 if (attr_pair_resized)
1044 test_mapper (struct taginfo *taginfo, void *arg)
1048 printf ("%s%s", taginfo->end_tag_p ? "/" : "", taginfo->name);
1049 for (i = 0; i < taginfo->nattrs; i++)
1050 printf (" %s=%s", taginfo->attrs[i].name, taginfo->attrs[i].value);
1058 char *x = (char *)xmalloc (size);
1061 int tag_counter = 0;
1063 while ((read_count = fread (x + length, 1, size - length, stdin)))
1065 length += read_count;
1067 x = (char *)xrealloc (x, size);
1070 map_html_tags (x, length, test_mapper, &tag_counter, 0, NULL, NULL);
1071 printf ("TAGS: %d\n", tag_counter);
1072 printf ("Tag backouts: %d\n", tag_backout_count);
1073 printf ("Comment backouts: %d\n", comment_backout_count);
1076 #endif /* STANDALONE */