1 /* Searching in a string.
2 Copyright (C) 2005-2008 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:
39 Return true and set *RESULTP if the search was completed.
40 Return false if it was aborted because not enough memory was available. */
42 knuth_morris_pratt_multibyte (const char *haystack, const char *needle,
45 size_t m = mbslen (needle);
46 mbchar_t *needle_mbchars;
49 /* Allocate room for needle_mbchars and the table. */
50 char *memory = (char *) nmalloca (m, sizeof (mbchar_t) + sizeof (size_t));
53 needle_mbchars = (mbchar_t *) memory;
54 table = (size_t *) (memory + m * sizeof (mbchar_t));
56 /* Fill needle_mbchars. */
62 for (mbui_init (iter, needle); mbui_avail (iter); mbui_advance (iter), j++)
63 mb_copy (&needle_mbchars[j], &mbui_cur (iter));
68 0 < table[i] <= i is defined such that
69 forall 0 < x < table[i]: needle[x..i-1] != needle[0..i-1-x],
70 and table[i] is as large as possible with this property.
74 needle[table[i]..i-1] = needle[0..i-1-table[i]].
76 rhaystack[0..i-1] == needle[0..i-1]
77 and exists h, i <= h < m: rhaystack[h] != needle[h]
79 forall 0 <= x < table[i]: rhaystack[x..x+m-1] != needle[0..m-1].
80 table[0] remains uninitialized. */
84 /* i = 1: Nothing to verify for x = 0. */
88 for (i = 2; i < m; i++)
90 /* Here: j = i-1 - table[i-1].
91 The inequality needle[x..i-1] != needle[0..i-1-x] is known to hold
92 for x < table[i-1], by induction.
93 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
94 mbchar_t *b = &needle_mbchars[i - 1];
98 /* Invariants: The inequality needle[x..i-1] != needle[0..i-1-x]
99 is known to hold for x < i-1-j.
100 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
101 if (mb_equal (*b, needle_mbchars[j]))
103 /* Set table[i] := i-1-j. */
107 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
108 for x = i-1-j, because
109 needle[i-1] != needle[j] = needle[i-1-x]. */
112 /* The inequality holds for all possible x. */
116 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
117 for i-1-j < x < i-1-j+table[j], because for these x:
119 = needle[x-(i-1-j)..j-1]
120 != needle[0..j-1-(x-(i-1-j))] (by definition of table[j])
122 hence needle[x..i-1] != needle[0..i-1-x].
124 needle[i-1-j+table[j]..i-2]
125 = needle[table[j]..j-1]
126 = needle[0..j-1-table[j]] (by definition of table[j]). */
129 /* Here: j = i - table[i]. */
133 /* Search, using the table to accelerate the processing. */
136 mbui_iterator_t rhaystack;
137 mbui_iterator_t phaystack;
141 mbui_init (rhaystack, haystack);
142 mbui_init (phaystack, haystack);
143 /* Invariant: phaystack = rhaystack + j. */
144 while (mbui_avail (phaystack))
145 if (mb_equal (needle_mbchars[j], mbui_cur (phaystack)))
148 mbui_advance (phaystack);
151 /* The entire needle has been found. */
152 *resultp = mbui_cur_ptr (rhaystack);
158 /* Found a match of needle[0..j-1], mismatch at needle[j]. */
159 size_t count = table[j];
161 for (; count > 0; count--)
163 if (!mbui_avail (rhaystack))
165 mbui_advance (rhaystack);
170 /* Found a mismatch at needle[0] already. */
171 if (!mbui_avail (rhaystack))
173 mbui_advance (rhaystack);
174 mbui_advance (phaystack);
183 /* Find the first occurrence of the character string NEEDLE in the character
184 string HAYSTACK. Return NULL if NEEDLE is not found in HAYSTACK. */
186 mbsstr (const char *haystack, const char *needle)
188 /* Be careful not to look at the entire extent of haystack or needle
189 until needed. This is useful because of these two cases:
190 - haystack may be very long, and a match of needle found early,
191 - needle may be very long, and not even a short initial segment of
192 needle may be found in haystack. */
196 mbui_iterator_t iter_needle;
198 mbui_init (iter_needle, needle);
199 if (mbui_avail (iter_needle))
201 /* Minimizing the worst-case complexity:
202 Let n = mbslen(haystack), m = mbslen(needle).
203 The naïve algorithm is O(n*m) worst-case.
204 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
206 To achieve linear complexity and yet amortize the cost of the
207 memory allocation, we activate the Knuth-Morris-Pratt algorithm
208 only once the naïve algorithm has already run for some time; more
210 - the outer loop count is >= 10,
211 - the average number of comparisons per outer loop is >= 5,
212 - the total number of comparisons is >= m.
213 But we try it only once. If the memory allocation attempt failed,
214 we don't retry it. */
216 size_t outer_loop_count = 0;
217 size_t comparison_count = 0;
218 size_t last_ccount = 0; /* last comparison count */
219 mbui_iterator_t iter_needle_last_ccount; /* = needle + last_ccount */
221 mbui_iterator_t iter_haystack;
223 mbui_init (iter_needle_last_ccount, needle);
224 mbui_init (iter_haystack, haystack);
225 for (;; mbui_advance (iter_haystack))
227 if (!mbui_avail (iter_haystack))
231 /* See whether it's advisable to use an asymptotically faster
234 && outer_loop_count >= 10
235 && comparison_count >= 5 * outer_loop_count)
237 /* See if needle + comparison_count now reaches the end of
239 size_t count = comparison_count - last_ccount;
241 count > 0 && mbui_avail (iter_needle_last_ccount);
243 mbui_advance (iter_needle_last_ccount);
244 last_ccount = comparison_count;
245 if (!mbui_avail (iter_needle_last_ccount))
247 /* Try the Knuth-Morris-Pratt algorithm. */
250 knuth_morris_pratt_multibyte (haystack, needle,
253 return (char *) result;
260 if (mb_equal (mbui_cur (iter_haystack), mbui_cur (iter_needle)))
261 /* The first character matches. */
263 mbui_iterator_t rhaystack;
264 mbui_iterator_t rneedle;
266 memcpy (&rhaystack, &iter_haystack, sizeof (mbui_iterator_t));
267 mbui_advance (rhaystack);
269 mbui_init (rneedle, needle);
270 if (!mbui_avail (rneedle))
272 mbui_advance (rneedle);
274 for (;; mbui_advance (rhaystack), mbui_advance (rneedle))
276 if (!mbui_avail (rneedle))
278 return (char *) mbui_cur_ptr (iter_haystack);
279 if (!mbui_avail (rhaystack))
283 if (!mb_equal (mbui_cur (rhaystack), mbui_cur (rneedle)))
284 /* Nothing in this round. */
291 return (char *) haystack;
298 /* Minimizing the worst-case complexity:
299 Let n = strlen(haystack), m = strlen(needle).
300 The naïve algorithm is O(n*m) worst-case.
301 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
303 To achieve linear complexity and yet amortize the cost of the
304 memory allocation, we activate the Knuth-Morris-Pratt algorithm
305 only once the naïve algorithm has already run for some time; more
307 - the outer loop count is >= 10,
308 - the average number of comparisons per outer loop is >= 5,
309 - the total number of comparisons is >= m.
310 But we try it only once. If the memory allocation attempt failed,
311 we don't retry it. */
313 size_t outer_loop_count = 0;
314 size_t comparison_count = 0;
315 size_t last_ccount = 0; /* last comparison count */
316 const char *needle_last_ccount = needle; /* = needle + last_ccount */
318 /* Speed up the following searches of needle by caching its first
324 if (*haystack == '\0')
328 /* See whether it's advisable to use an asymptotically faster
331 && outer_loop_count >= 10
332 && comparison_count >= 5 * outer_loop_count)
334 /* See if needle + comparison_count now reaches the end of
336 if (needle_last_ccount != NULL)
338 needle_last_ccount +=
339 strnlen (needle_last_ccount,
340 comparison_count - last_ccount);
341 if (*needle_last_ccount == '\0')
342 needle_last_ccount = NULL;
343 last_ccount = comparison_count;
345 if (needle_last_ccount == NULL)
347 /* Try the Knuth-Morris-Pratt algorithm. */
350 knuth_morris_pratt_unibyte (haystack, needle - 1,
353 return (char *) result;
361 /* The first character matches. */
363 const char *rhaystack = haystack + 1;
364 const char *rneedle = needle;
366 for (;; rhaystack++, rneedle++)
368 if (*rneedle == '\0')
370 return (char *) haystack;
371 if (*rhaystack == '\0')
375 if (*rhaystack != *rneedle)
376 /* Nothing in this round. */
383 return (char *) haystack;