1 /* Searching in a string. -*- coding: utf-8 -*-
2 Copyright (C) 2005-2018 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 <https://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 https://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_memory = needle_mbchars + m;
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);
182 /* Find the first occurrence of the character string NEEDLE in the character
183 string HAYSTACK. Return NULL if NEEDLE is not found in HAYSTACK. */
185 mbsstr (const char *haystack, const char *needle)
187 /* Be careful not to look at the entire extent of haystack or needle
188 until needed. This is useful because of these two cases:
189 - haystack may be very long, and a match of needle found early,
190 - needle may be very long, and not even a short initial segment of
191 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;
295 /* Minimizing the worst-case complexity:
296 Let n = strlen(haystack), m = strlen(needle).
297 The naïve algorithm is O(n*m) worst-case.
298 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
300 To achieve linear complexity and yet amortize the cost of the
301 memory allocation, we activate the Knuth-Morris-Pratt algorithm
302 only once the naïve algorithm has already run for some time; more
304 - the outer loop count is >= 10,
305 - the average number of comparisons per outer loop is >= 5,
306 - the total number of comparisons is >= m.
307 But we try it only once. If the memory allocation attempt failed,
308 we don't retry it. */
310 size_t outer_loop_count = 0;
311 size_t comparison_count = 0;
312 size_t last_ccount = 0; /* last comparison count */
313 const char *needle_last_ccount = needle; /* = needle + last_ccount */
315 /* Speed up the following searches of needle by caching its first
321 if (*haystack == '\0')
325 /* See whether it's advisable to use an asymptotically faster
328 && outer_loop_count >= 10
329 && comparison_count >= 5 * outer_loop_count)
331 /* See if needle + comparison_count now reaches the end of
333 if (needle_last_ccount != NULL)
335 needle_last_ccount +=
336 strnlen (needle_last_ccount,
337 comparison_count - last_ccount);
338 if (*needle_last_ccount == '\0')
339 needle_last_ccount = NULL;
340 last_ccount = comparison_count;
342 if (needle_last_ccount == NULL)
344 /* Try the Knuth-Morris-Pratt algorithm. */
345 const unsigned char *result;
347 knuth_morris_pratt ((const unsigned char *) haystack,
348 (const unsigned char *) (needle - 1),
352 return (char *) result;
360 /* The first character matches. */
362 const char *rhaystack = haystack + 1;
363 const char *rneedle = needle;
365 for (;; rhaystack++, rneedle++)
367 if (*rneedle == '\0')
369 return (char *) haystack;
370 if (*rhaystack == '\0')
374 if (*rhaystack != *rneedle)
375 /* Nothing in this round. */
382 return (char *) haystack;