1 /* Searching in a string.
2 Copyright (C) 2005-2013 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 */
29 /* Knuth-Morris-Pratt algorithm. */
30 #define UNIT unsigned char
31 #define CANON_ELEMENT(c) c
34 /* Knuth-Morris-Pratt algorithm.
35 See http://en.wikipedia.org/wiki/Knuth-Morris-Pratt_algorithm
36 Return a boolean indicating success:
37 Return true and set *RESULTP if the search was completed.
38 Return false if it was aborted because not enough memory was available. */
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 void *memory = nmalloca (m, sizeof (mbchar_t) + sizeof (size_t));
52 needle_mbchars = memory;
53 table = table_memory = needle_mbchars + m;
55 /* Fill needle_mbchars. */
61 for (mbui_init (iter, needle); mbui_avail (iter); mbui_advance (iter), j++)
62 mb_copy (&needle_mbchars[j], &mbui_cur (iter));
67 0 < table[i] <= i is defined such that
68 forall 0 < x < table[i]: needle[x..i-1] != needle[0..i-1-x],
69 and table[i] is as large as possible with this property.
73 needle[table[i]..i-1] = needle[0..i-1-table[i]].
75 rhaystack[0..i-1] == needle[0..i-1]
76 and exists h, i <= h < m: rhaystack[h] != needle[h]
78 forall 0 <= x < table[i]: rhaystack[x..x+m-1] != needle[0..m-1].
79 table[0] remains uninitialized. */
83 /* i = 1: Nothing to verify for x = 0. */
87 for (i = 2; i < m; i++)
89 /* Here: j = i-1 - table[i-1].
90 The inequality needle[x..i-1] != needle[0..i-1-x] is known to hold
91 for x < table[i-1], by induction.
92 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
93 mbchar_t *b = &needle_mbchars[i - 1];
97 /* Invariants: The inequality needle[x..i-1] != needle[0..i-1-x]
98 is known to hold for x < i-1-j.
99 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
100 if (mb_equal (*b, needle_mbchars[j]))
102 /* Set table[i] := i-1-j. */
106 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
107 for x = i-1-j, because
108 needle[i-1] != needle[j] = needle[i-1-x]. */
111 /* The inequality holds for all possible x. */
115 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
116 for i-1-j < x < i-1-j+table[j], because for these x:
118 = needle[x-(i-1-j)..j-1]
119 != needle[0..j-1-(x-(i-1-j))] (by definition of table[j])
121 hence needle[x..i-1] != needle[0..i-1-x].
123 needle[i-1-j+table[j]..i-2]
124 = needle[table[j]..j-1]
125 = needle[0..j-1-table[j]] (by definition of table[j]). */
128 /* Here: j = i - table[i]. */
132 /* Search, using the table to accelerate the processing. */
135 mbui_iterator_t rhaystack;
136 mbui_iterator_t phaystack;
140 mbui_init (rhaystack, haystack);
141 mbui_init (phaystack, haystack);
142 /* Invariant: phaystack = rhaystack + j. */
143 while (mbui_avail (phaystack))
144 if (mb_equal (needle_mbchars[j], mbui_cur (phaystack)))
147 mbui_advance (phaystack);
150 /* The entire needle has been found. */
151 *resultp = mbui_cur_ptr (rhaystack);
157 /* Found a match of needle[0..j-1], mismatch at needle[j]. */
158 size_t count = table[j];
160 for (; count > 0; count--)
162 if (!mbui_avail (rhaystack))
164 mbui_advance (rhaystack);
169 /* Found a mismatch at needle[0] already. */
170 if (!mbui_avail (rhaystack))
172 mbui_advance (rhaystack);
173 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. */
193 mbui_iterator_t iter_needle;
195 mbui_init (iter_needle, needle);
196 if (mbui_avail (iter_needle))
198 /* Minimizing the worst-case complexity:
199 Let n = mbslen(haystack), m = mbslen(needle).
200 The naïve algorithm is O(n*m) worst-case.
201 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
203 To achieve linear complexity and yet amortize the cost of the
204 memory allocation, we activate the Knuth-Morris-Pratt algorithm
205 only once the naïve algorithm has already run for some time; more
207 - the outer loop count is >= 10,
208 - the average number of comparisons per outer loop is >= 5,
209 - the total number of comparisons is >= m.
210 But we try it only once. If the memory allocation attempt failed,
211 we don't retry it. */
213 size_t outer_loop_count = 0;
214 size_t comparison_count = 0;
215 size_t last_ccount = 0; /* last comparison count */
216 mbui_iterator_t iter_needle_last_ccount; /* = needle + last_ccount */
218 mbui_iterator_t iter_haystack;
220 mbui_init (iter_needle_last_ccount, needle);
221 mbui_init (iter_haystack, haystack);
222 for (;; mbui_advance (iter_haystack))
224 if (!mbui_avail (iter_haystack))
228 /* See whether it's advisable to use an asymptotically faster
231 && outer_loop_count >= 10
232 && comparison_count >= 5 * outer_loop_count)
234 /* See if needle + comparison_count now reaches the end of
236 size_t count = comparison_count - last_ccount;
238 count > 0 && mbui_avail (iter_needle_last_ccount);
240 mbui_advance (iter_needle_last_ccount);
241 last_ccount = comparison_count;
242 if (!mbui_avail (iter_needle_last_ccount))
244 /* Try the Knuth-Morris-Pratt algorithm. */
247 knuth_morris_pratt_multibyte (haystack, needle,
250 return (char *) result;
257 if (mb_equal (mbui_cur (iter_haystack), mbui_cur (iter_needle)))
258 /* The first character matches. */
260 mbui_iterator_t rhaystack;
261 mbui_iterator_t rneedle;
263 memcpy (&rhaystack, &iter_haystack, sizeof (mbui_iterator_t));
264 mbui_advance (rhaystack);
266 mbui_init (rneedle, needle);
267 if (!mbui_avail (rneedle))
269 mbui_advance (rneedle);
271 for (;; mbui_advance (rhaystack), mbui_advance (rneedle))
273 if (!mbui_avail (rneedle))
275 return (char *) mbui_cur_ptr (iter_haystack);
276 if (!mbui_avail (rhaystack))
280 if (!mb_equal (mbui_cur (rhaystack), mbui_cur (rneedle)))
281 /* Nothing in this round. */
288 return (char *) haystack;
294 /* Minimizing the worst-case complexity:
295 Let n = strlen(haystack), m = strlen(needle).
296 The naïve algorithm is O(n*m) worst-case.
297 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
299 To achieve linear complexity and yet amortize the cost of the
300 memory allocation, we activate the Knuth-Morris-Pratt algorithm
301 only once the naïve algorithm has already run for some time; more
303 - the outer loop count is >= 10,
304 - the average number of comparisons per outer loop is >= 5,
305 - the total number of comparisons is >= m.
306 But we try it only once. If the memory allocation attempt failed,
307 we don't retry it. */
309 size_t outer_loop_count = 0;
310 size_t comparison_count = 0;
311 size_t last_ccount = 0; /* last comparison count */
312 const char *needle_last_ccount = needle; /* = needle + last_ccount */
314 /* Speed up the following searches of needle by caching its first
320 if (*haystack == '\0')
324 /* See whether it's advisable to use an asymptotically faster
327 && outer_loop_count >= 10
328 && comparison_count >= 5 * outer_loop_count)
330 /* See if needle + comparison_count now reaches the end of
332 if (needle_last_ccount != NULL)
334 needle_last_ccount +=
335 strnlen (needle_last_ccount,
336 comparison_count - last_ccount);
337 if (*needle_last_ccount == '\0')
338 needle_last_ccount = NULL;
339 last_ccount = comparison_count;
341 if (needle_last_ccount == NULL)
343 /* Try the Knuth-Morris-Pratt algorithm. */
344 const unsigned char *result;
346 knuth_morris_pratt ((const unsigned char *) haystack,
347 (const unsigned char *) (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;