1 /* Searching in a string.
2 Copyright (C) 2005-2007 Free Software Foundation, Inc.
3 Written by Bruno Haible <bruno@clisp.org>, 2005.
5 This program is free software: you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 3 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program. If not, see <http://www.gnu.org/licenses/>. */
24 #include <stddef.h> /* for NULL, in case a nonstandard string.h lacks it */
31 /* Knuth-Morris-Pratt algorithm. */
32 #define CANON_ELEMENT(c) c
36 /* Knuth-Morris-Pratt algorithm.
37 See http://en.wikipedia.org/wiki/Knuth-Morris-Pratt_algorithm
38 Return a boolean indicating success. */
40 knuth_morris_pratt_multibyte (const char *haystack, const char *needle,
43 size_t m = mbslen (needle);
44 mbchar_t *needle_mbchars;
47 /* Allocate room for needle_mbchars and the table. */
48 char *memory = (char *) nmalloca (m, sizeof (mbchar_t) + sizeof (size_t));
51 needle_mbchars = (mbchar_t *) memory;
52 table = (size_t *) (memory + m * sizeof (mbchar_t));
54 /* Fill needle_mbchars. */
60 for (mbui_init (iter, needle); mbui_avail (iter); mbui_advance (iter), j++)
61 mb_copy (&needle_mbchars[j], &mbui_cur (iter));
66 0 < table[i] <= i is defined such that
67 forall 0 < x < table[i]: needle[x..i-1] != needle[0..i-1-x],
68 and table[i] is as large as possible with this property.
72 needle[table[i]..i-1] = needle[0..i-1-table[i]].
74 rhaystack[0..i-1] == needle[0..i-1]
75 and exists h, i <= h < m: rhaystack[h] != needle[h]
77 forall 0 <= x < table[i]: rhaystack[x..x+m-1] != needle[0..m-1].
78 table[0] remains uninitialized. */
82 /* i = 1: Nothing to verify for x = 0. */
86 for (i = 2; i < m; i++)
88 /* Here: j = i-1 - table[i-1].
89 The inequality needle[x..i-1] != needle[0..i-1-x] is known to hold
90 for x < table[i-1], by induction.
91 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
92 mbchar_t *b = &needle_mbchars[i - 1];
96 /* Invariants: The inequality needle[x..i-1] != needle[0..i-1-x]
97 is known to hold for x < i-1-j.
98 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
99 if (mb_equal (*b, needle_mbchars[j]))
101 /* Set table[i] := i-1-j. */
105 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
106 for x = i-1-j, because
107 needle[i-1] != needle[j] = needle[i-1-x]. */
110 /* The inequality holds for all possible x. */
114 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
115 for i-1-j < x < i-1-j+table[j], because for these x:
117 = needle[x-(i-1-j)..j-1]
118 != needle[0..j-1-(x-(i-1-j))] (by definition of table[j])
120 hence needle[x..i-1] != needle[0..i-1-x].
122 needle[i-1-j+table[j]..i-2]
123 = needle[table[j]..j-1]
124 = needle[0..j-1-table[j]] (by definition of table[j]). */
127 /* Here: j = i - table[i]. */
131 /* Search, using the table to accelerate the processing. */
134 mbui_iterator_t rhaystack;
135 mbui_iterator_t phaystack;
139 mbui_init (rhaystack, haystack);
140 mbui_init (phaystack, haystack);
141 /* Invariant: phaystack = rhaystack + j. */
142 while (mbui_avail (phaystack))
143 if (mb_equal (needle_mbchars[j], mbui_cur (phaystack)))
146 mbui_advance (phaystack);
149 /* The entire needle has been found. */
150 *resultp = mbui_cur_ptr (rhaystack);
156 /* Found a match of needle[0..j-1], mismatch at needle[j]. */
157 size_t count = table[j];
159 for (; count > 0; count--)
161 if (!mbui_avail (rhaystack))
163 mbui_advance (rhaystack);
168 /* Found a mismatch at needle[0] already. */
169 if (!mbui_avail (rhaystack))
171 mbui_advance (rhaystack);
172 mbui_advance (phaystack);
181 /* Find the first occurrence of the character string NEEDLE in the character
182 string HAYSTACK. Return NULL if NEEDLE is not found in HAYSTACK. */
184 mbsstr (const char *haystack, const char *needle)
186 /* Be careful not to look at the entire extent of haystack or needle
187 until needed. This is useful because of these two cases:
188 - haystack may be very long, and a match of needle found early,
189 - needle may be very long, and not even a short initial segment of
190 needle may be found in haystack. */
194 mbui_iterator_t iter_needle;
196 mbui_init (iter_needle, needle);
197 if (mbui_avail (iter_needle))
199 /* Minimizing the worst-case complexity:
200 Let n = mbslen(haystack), m = mbslen(needle).
201 The naïve algorithm is O(n*m) worst-case.
202 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
204 To achieve linear complexity and yet amortize the cost of the
205 memory allocation, we activate the Knuth-Morris-Pratt algorithm
206 only once the naïve algorithm has already run for some time; more
208 - the outer loop count is >= 10,
209 - the average number of comparisons per outer loop is >= 5,
210 - the total number of comparisons is >= m.
211 But we try it only once. If the memory allocation attempt failed,
212 we don't retry it. */
214 size_t outer_loop_count = 0;
215 size_t comparison_count = 0;
216 size_t last_ccount = 0; /* last comparison count */
217 mbui_iterator_t iter_needle_last_ccount; /* = needle + last_ccount */
219 mbui_iterator_t iter_haystack;
221 mbui_init (iter_needle_last_ccount, needle);
222 mbui_init (iter_haystack, haystack);
223 for (;; mbui_advance (iter_haystack))
225 if (!mbui_avail (iter_haystack))
229 /* See whether it's advisable to use an asymptotically faster
232 && outer_loop_count >= 10
233 && comparison_count >= 5 * outer_loop_count)
235 /* See if needle + comparison_count now reaches the end of
237 size_t count = comparison_count - last_ccount;
239 count > 0 && mbui_avail (iter_needle_last_ccount);
241 mbui_advance (iter_needle_last_ccount);
242 last_ccount = comparison_count;
243 if (!mbui_avail (iter_needle_last_ccount))
245 /* Try the Knuth-Morris-Pratt algorithm. */
248 knuth_morris_pratt_multibyte (haystack, needle,
251 return (char *) result;
258 if (mb_equal (mbui_cur (iter_haystack), mbui_cur (iter_needle)))
259 /* The first character matches. */
261 mbui_iterator_t rhaystack;
262 mbui_iterator_t rneedle;
264 memcpy (&rhaystack, &iter_haystack, sizeof (mbui_iterator_t));
265 mbui_advance (rhaystack);
267 mbui_init (rneedle, needle);
268 if (!mbui_avail (rneedle))
270 mbui_advance (rneedle);
272 for (;; mbui_advance (rhaystack), mbui_advance (rneedle))
274 if (!mbui_avail (rneedle))
276 return (char *) mbui_cur_ptr (iter_haystack);
277 if (!mbui_avail (rhaystack))
281 if (!mb_equal (mbui_cur (rhaystack), mbui_cur (rneedle)))
282 /* Nothing in this round. */
289 return (char *) haystack;
296 /* Minimizing the worst-case complexity:
297 Let n = strlen(haystack), m = strlen(needle).
298 The naïve algorithm is O(n*m) worst-case.
299 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
301 To achieve linear complexity and yet amortize the cost of the
302 memory allocation, we activate the Knuth-Morris-Pratt algorithm
303 only once the naïve algorithm has already run for some time; more
305 - the outer loop count is >= 10,
306 - the average number of comparisons per outer loop is >= 5,
307 - the total number of comparisons is >= m.
308 But we try it only once. If the memory allocation attempt failed,
309 we don't retry it. */
311 size_t outer_loop_count = 0;
312 size_t comparison_count = 0;
313 size_t last_ccount = 0; /* last comparison count */
314 const char *needle_last_ccount = needle; /* = needle + last_ccount */
316 /* Speed up the following searches of needle by caching its first
322 if (*haystack == '\0')
326 /* See whether it's advisable to use an asymptotically faster
329 && outer_loop_count >= 10
330 && comparison_count >= 5 * outer_loop_count)
332 /* See if needle + comparison_count now reaches the end of
334 if (needle_last_ccount != NULL)
336 needle_last_ccount +=
337 strnlen (needle_last_ccount,
338 comparison_count - last_ccount);
339 if (*needle_last_ccount == '\0')
340 needle_last_ccount = NULL;
341 last_ccount = comparison_count;
343 if (needle_last_ccount == NULL)
345 /* Try the Knuth-Morris-Pratt algorithm. */
348 knuth_morris_pratt_unibyte (haystack, needle - 1,
351 return (char *) result;
359 /* The first character matches. */
361 const char *rhaystack = haystack + 1;
362 const char *rneedle = needle;
364 for (;; rhaystack++, rneedle++)
366 if (*rneedle == '\0')
368 return (char *) haystack;
369 if (*rhaystack == '\0')
373 if (*rhaystack != *rneedle)
374 /* Nothing in this round. */
381 return (char *) haystack;