1 /* Copyright (C) 1991,92,93,94,96,97,98,2000,2004,2007 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2, or (at your option)
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License along
15 with this program; if not, write to the Free Software Foundation,
16 Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. */
29 # define __builtin_expect(expr, val) (expr)
32 /* Knuth-Morris-Pratt algorithm.
33 See http://en.wikipedia.org/wiki/Knuth-Morris-Pratt_algorithm
34 Return a boolean indicating success. */
37 knuth_morris_pratt (const char *haystack, const char *last_haystack,
38 const char *needle, size_t m,
41 /* Allocate the table. */
43 if ((size_t) -1 / sizeof (size_t) < m)
45 table = (size_t *) malloca (m * sizeof (size_t));
50 0 < table[i] <= i is defined such that
51 forall 0 < x < table[i]: needle[x..i-1] != needle[0..i-1-x],
52 and table[i] is as large as possible with this property.
56 needle[table[i]..i-1] = needle[0..i-1-table[i]].
58 rhaystack[0..i-1] == needle[0..i-1]
59 and exists h, i <= h < m: rhaystack[h] != needle[h]
61 forall 0 <= x < table[i]: rhaystack[x..x+m-1] != needle[0..m-1].
62 table[0] remains uninitialized. */
66 /* i = 1: Nothing to verify for x = 0. */
70 for (i = 2; i < m; i++)
72 /* Here: j = i-1 - table[i-1].
73 The inequality needle[x..i-1] != needle[0..i-1-x] is known to hold
74 for x < table[i-1], by induction.
75 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
76 unsigned char b = (unsigned char) needle[i - 1];
80 /* Invariants: The inequality needle[x..i-1] != needle[0..i-1-x]
81 is known to hold for x < i-1-j.
82 Furthermore, if j>0: needle[i-1-j..i-2] = needle[0..j-1]. */
83 if (b == (unsigned char) needle[j])
85 /* Set table[i] := i-1-j. */
89 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
90 for x = i-1-j, because
91 needle[i-1] != needle[j] = needle[i-1-x]. */
94 /* The inequality holds for all possible x. */
98 /* The inequality needle[x..i-1] != needle[0..i-1-x] also holds
99 for i-1-j < x < i-1-j+table[j], because for these x:
101 = needle[x-(i-1-j)..j-1]
102 != needle[0..j-1-(x-(i-1-j))] (by definition of table[j])
104 hence needle[x..i-1] != needle[0..i-1-x].
106 needle[i-1-j+table[j]..i-2]
107 = needle[table[j]..j-1]
108 = needle[0..j-1-table[j]] (by definition of table[j]). */
111 /* Here: j = i - table[i]. */
115 /* Search, using the table to accelerate the processing. */
118 const char *rhaystack;
119 const char *phaystack;
123 rhaystack = haystack;
124 phaystack = haystack;
125 /* Invariant: phaystack = rhaystack + j. */
126 while (phaystack != last_haystack)
127 if ((unsigned char) needle[j] == (unsigned char) *phaystack)
133 /* The entire needle has been found. */
134 *resultp = rhaystack;
140 /* Found a match of needle[0..j-1], mismatch at needle[j]. */
141 rhaystack += table[j];
146 /* Found a mismatch at needle[0] already. */
156 /* Return the first occurrence of NEEDLE in HAYSTACK. Return HAYSTACK
157 if NEEDLE_LEN is 0, otherwise NULL if NEEDLE is not found in
160 memmem (const void *haystack_start, size_t haystack_len,
161 const void *needle_start, size_t needle_len)
163 /* Operating with void * is awkward. */
164 const char *haystack = (const char *) haystack_start;
165 const char *needle = (const char *) needle_start;
166 const char *last_haystack = haystack + haystack_len;
167 const char *last_needle = needle + needle_len;
170 /* The first occurrence of the empty string is deemed to occur at
171 the beginning of the string. */
172 return (void *) haystack;
174 /* Sanity check, otherwise the loop might search through the whole
176 if (__builtin_expect (haystack_len < needle_len, 0))
179 /* Use optimizations in memchr when possible. */
180 if (__builtin_expect (needle_len == 1, 0))
181 return memchr (haystack, (unsigned char) *needle, haystack_len);
183 /* Minimizing the worst-case complexity:
184 Let n = haystack_len, m = needle_len.
185 The naïve algorithm is O(n*m) worst-case.
186 The Knuth-Morris-Pratt algorithm is O(n) worst-case but it needs a
188 To achieve linear complexity and yet amortize the cost of the
189 memory allocation, we activate the Knuth-Morris-Pratt algorithm
190 only once the naïve algorithm has already run for some time; more
192 - the outer loop count is >= 10,
193 - the average number of comparisons per outer loop is >= 5,
194 - the total number of comparisons is >= m.
195 But we try it only once. If the memory allocation attempt failed,
196 we don't retry it. */
199 size_t outer_loop_count = 0;
200 size_t comparison_count = 0;
202 /* Speed up the following searches of needle by caching its first
208 if (haystack == last_haystack)
212 /* See whether it's advisable to use an asymptotically faster
215 && outer_loop_count >= 10
216 && comparison_count >= 5 * outer_loop_count)
218 /* See if needle + comparison_count now reaches the end of
220 if (comparison_count >= needle_len)
222 /* Try the Knuth-Morris-Pratt algorithm. */
224 if (knuth_morris_pratt (haystack, last_haystack,
225 needle - 1, needle_len, &result))
226 return (void *) result;
234 /* The first byte matches. */
236 const char *rhaystack = haystack + 1;
237 const char *rneedle = needle;
239 for (;; rhaystack++, rneedle++)
241 if (rneedle == last_needle)
243 return (void *) haystack;
244 if (rhaystack == last_haystack)
248 if (*rhaystack != *rneedle)
249 /* Nothing in this round. */