Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * levenshtein.c
4 : * Levenshtein distance implementation.
5 : *
6 : * Original author: Joe Conway <mail@joeconway.com>
7 : *
8 : * This file is included by varlena.c twice, to provide matching code for (1)
9 : * Levenshtein distance with custom costings, and (2) Levenshtein distance with
10 : * custom costings and a "max" value above which exact distances are not
11 : * interesting. Before the inclusion, we rely on the presence of the inline
12 : * function rest_of_char_same().
13 : *
14 : * Written based on a description of the algorithm by Michael Gilleland found
15 : * at http://www.merriampark.com/ld.htm. Also looked at levenshtein.c in the
16 : * PHP 4.0.6 distribution for inspiration. Configurable penalty costs
17 : * extension is introduced by Volkan YAZICI <volkan.yazici@gmail.com.
18 : *
19 : * Copyright (c) 2001-2026, PostgreSQL Global Development Group
20 : *
21 : * IDENTIFICATION
22 : * src/backend/utils/adt/levenshtein.c
23 : *
24 : *-------------------------------------------------------------------------
25 : */
26 : #define MAX_LEVENSHTEIN_STRLEN 255
27 :
28 : /*
29 : * Calculates Levenshtein distance metric between supplied strings, which are
30 : * not necessarily null-terminated.
31 : *
32 : * source: source string, of length slen bytes.
33 : * target: target string, of length tlen bytes.
34 : * ins_c, del_c, sub_c: costs to charge for character insertion, deletion,
35 : * and substitution respectively; (1, 1, 1) costs suffice for common
36 : * cases, but your mileage may vary.
37 : * max_d: if provided and >= 0, maximum distance we care about; see below.
38 : * trusted: caller is trusted and need not obey MAX_LEVENSHTEIN_STRLEN.
39 : *
40 : * One way to compute Levenshtein distance is to incrementally construct
41 : * an (m+1)x(n+1) matrix where cell (i, j) represents the minimum number
42 : * of operations required to transform the first i characters of s into
43 : * the first j characters of t. The last column of the final row is the
44 : * answer.
45 : *
46 : * We use that algorithm here with some modification. In lieu of holding
47 : * the entire array in memory at once, we'll just use two arrays of size
48 : * m+1 for storing accumulated values. At each step one array represents
49 : * the "previous" row and one is the "current" row of the notional large
50 : * array.
51 : *
52 : * If max_d >= 0, we only need to provide an accurate answer when that answer
53 : * is less than or equal to max_d. From any cell in the matrix, there is
54 : * theoretical "minimum residual distance" from that cell to the last column
55 : * of the final row. This minimum residual distance is zero when the
56 : * untransformed portions of the strings are of equal length (because we might
57 : * get lucky and find all the remaining characters matching) and is otherwise
58 : * based on the minimum number of insertions or deletions needed to make them
59 : * equal length. The residual distance grows as we move toward the upper
60 : * right or lower left corners of the matrix. When the max_d bound is
61 : * usefully tight, we can use this property to avoid computing the entirety
62 : * of each row; instead, we maintain a start_column and stop_column that
63 : * identify the portion of the matrix close to the diagonal which can still
64 : * affect the final answer.
65 : */
66 : int
67 : #ifdef LEVENSHTEIN_LESS_EQUAL
68 1520 : varstr_levenshtein_less_equal(const char *source, int slen,
69 : const char *target, int tlen,
70 : int ins_c, int del_c, int sub_c,
71 : int max_d, bool trusted)
72 : #else
73 2 : varstr_levenshtein(const char *source, int slen,
74 : const char *target, int tlen,
75 : int ins_c, int del_c, int sub_c,
76 : bool trusted)
77 : #endif
78 : {
79 : int m,
80 : n;
81 : int *prev;
82 : int *curr;
83 1522 : int *s_char_len = NULL;
84 : int j;
85 : const char *y;
86 1522 : const char *send = source + slen;
87 1522 : const char *tend = target + tlen;
88 :
89 : /*
90 : * For varstr_levenshtein_less_equal, we have real variables called
91 : * start_column and stop_column; otherwise it's just short-hand for 0 and
92 : * m.
93 : */
94 : #ifdef LEVENSHTEIN_LESS_EQUAL
95 : int start_column,
96 : stop_column;
97 :
98 : #undef START_COLUMN
99 : #undef STOP_COLUMN
100 : #define START_COLUMN start_column
101 : #define STOP_COLUMN stop_column
102 : #else
103 : #undef START_COLUMN
104 : #undef STOP_COLUMN
105 : #define START_COLUMN 0
106 : #define STOP_COLUMN m
107 : #endif
108 :
109 : /* Convert string lengths (in bytes) to lengths in characters */
110 1522 : m = pg_mbstrlen_with_len(source, slen);
111 1522 : n = pg_mbstrlen_with_len(target, tlen);
112 :
113 : /*
114 : * We can transform an empty s into t with n insertions, or a non-empty t
115 : * into an empty s with m deletions.
116 : */
117 1522 : if (!m)
118 0 : return n * ins_c;
119 1522 : if (!n)
120 0 : return m * del_c;
121 :
122 : /*
123 : * For security concerns, restrict excessive CPU+RAM usage. (This
124 : * implementation uses O(m) memory and has O(mn) complexity.) If
125 : * "trusted" is true, caller is responsible for not making excessive
126 : * requests, typically by using a small max_d along with strings that are
127 : * bounded, though not necessarily to MAX_LEVENSHTEIN_STRLEN exactly.
128 : */
129 1522 : if (!trusted &&
130 4 : (m > MAX_LEVENSHTEIN_STRLEN ||
131 : n > MAX_LEVENSHTEIN_STRLEN))
132 0 : ereport(ERROR,
133 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
134 : errmsg("levenshtein argument exceeds maximum length of %d characters",
135 : MAX_LEVENSHTEIN_STRLEN)));
136 :
137 : #ifdef LEVENSHTEIN_LESS_EQUAL
138 : /* Initialize start and stop columns. */
139 1520 : start_column = 0;
140 1520 : stop_column = m + 1;
141 :
142 : /*
143 : * If max_d >= 0, determine whether the bound is impossibly tight. If so,
144 : * return max_d + 1 immediately. Otherwise, determine whether it's tight
145 : * enough to limit the computation we must perform. If so, figure out
146 : * initial stop column.
147 : */
148 1520 : if (max_d >= 0)
149 : {
150 : int min_theo_d; /* Theoretical minimum distance. */
151 : int max_theo_d; /* Theoretical maximum distance. */
152 1520 : int net_inserts = n - m;
153 :
154 1520 : min_theo_d = net_inserts < 0 ?
155 1520 : -net_inserts * del_c : net_inserts * ins_c;
156 1520 : if (min_theo_d > max_d)
157 558 : return max_d + 1;
158 962 : if (ins_c + del_c < sub_c)
159 0 : sub_c = ins_c + del_c;
160 962 : max_theo_d = min_theo_d + sub_c * Min(m, n);
161 962 : if (max_d >= max_theo_d)
162 286 : max_d = -1;
163 676 : else if (ins_c + del_c > 0)
164 : {
165 : /*
166 : * Figure out how much of the first row of the notional matrix we
167 : * need to fill in. If the string is growing, the theoretical
168 : * minimum distance already incorporates the cost of deleting the
169 : * number of characters necessary to make the two strings equal in
170 : * length. Each additional deletion forces another insertion, so
171 : * the best-case total cost increases by ins_c + del_c. If the
172 : * string is shrinking, the minimum theoretical cost assumes no
173 : * excess deletions; that is, we're starting no further right than
174 : * column n - m. If we do start further right, the best-case
175 : * total cost increases by ins_c + del_c for each move right.
176 : */
177 676 : int slack_d = max_d - min_theo_d;
178 676 : int best_column = net_inserts < 0 ? -net_inserts : 0;
179 :
180 676 : stop_column = best_column + (slack_d / (ins_c + del_c)) + 1;
181 676 : if (stop_column > m)
182 0 : stop_column = m + 1;
183 : }
184 : }
185 : #endif
186 :
187 : /*
188 : * In order to avoid calling pg_mblen_range() repeatedly on each character
189 : * in s, we cache all the lengths before starting the main loop -- but if
190 : * all the characters in both strings are single byte, then we skip this
191 : * and use a fast-path in the main loop. If only one string contains
192 : * multi-byte characters, we still build the array, so that the fast-path
193 : * needn't deal with the case where the array hasn't been initialized.
194 : */
195 964 : if (m != slen || n != tlen)
196 : {
197 : int i;
198 3 : const char *cp = source;
199 :
200 3 : s_char_len = (int *) palloc((m + 1) * sizeof(int));
201 30 : for (i = 0; i < m; ++i)
202 : {
203 27 : s_char_len[i] = pg_mblen_range(cp, send);
204 27 : cp += s_char_len[i];
205 : }
206 3 : s_char_len[i] = 0;
207 : }
208 :
209 : /* One more cell for initialization column and row. */
210 964 : ++m;
211 964 : ++n;
212 :
213 : /* Previous and current rows of notional array. */
214 964 : prev = (int *) palloc(2 * m * sizeof(int));
215 964 : curr = prev + m;
216 :
217 : /*
218 : * To transform the first i characters of s into the first 0 characters of
219 : * t, we must perform i deletions.
220 : */
221 3748 : for (int i = START_COLUMN; i < STOP_COLUMN; i++)
222 2784 : prev[i] = i * del_c;
223 :
224 : /* Loop through rows of the notional array */
225 3785 : for (y = target, j = 1; j < n; j++)
226 : {
227 : int *temp;
228 3397 : const char *x = source;
229 3397 : int y_char_len = n != tlen + 1 ? pg_mblen_range(y, tend) : 1;
230 : int i;
231 :
232 : #ifdef LEVENSHTEIN_LESS_EQUAL
233 :
234 : /*
235 : * In the best case, values percolate down the diagonal unchanged, so
236 : * we must increment stop_column unless it's already on the right end
237 : * of the array. The inner loop will read prev[stop_column], so we
238 : * have to initialize it even though it shouldn't affect the result.
239 : */
240 3385 : if (stop_column < m)
241 : {
242 2736 : prev[stop_column] = max_d + 1;
243 2736 : ++stop_column;
244 : }
245 :
246 : /*
247 : * The main loop fills in curr, but curr[0] needs a special case: to
248 : * transform the first 0 characters of s into the first j characters
249 : * of t, we must perform j insertions. However, if start_column > 0,
250 : * this special case does not apply.
251 : */
252 3385 : if (start_column == 0)
253 : {
254 2167 : curr[0] = j * ins_c;
255 2167 : i = 1;
256 : }
257 : else
258 1218 : i = start_column;
259 : #else
260 12 : curr[0] = j * ins_c;
261 12 : i = 1;
262 : #endif
263 :
264 : /*
265 : * This inner loop is critical to performance, so we include a
266 : * fast-path to handle the (fairly common) case where no multibyte
267 : * characters are in the mix. The fast-path is entitled to assume
268 : * that if s_char_len is not initialized then BOTH strings contain
269 : * only single-byte characters.
270 : */
271 3397 : if (s_char_len != NULL)
272 : {
273 186 : for (; i < STOP_COLUMN; i++)
274 : {
275 : int ins;
276 : int del;
277 : int sub;
278 156 : int x_char_len = s_char_len[i - 1];
279 :
280 : /*
281 : * Calculate costs for insertion, deletion, and substitution.
282 : *
283 : * When calculating cost for substitution, we compare the last
284 : * character of each possibly-multibyte character first,
285 : * because that's enough to rule out most mis-matches. If we
286 : * get past that test, then we compare the lengths and the
287 : * remaining bytes.
288 : */
289 156 : ins = prev[i] + ins_c;
290 156 : del = curr[i - 1] + del_c;
291 156 : if (x[x_char_len - 1] == y[y_char_len - 1]
292 27 : && x_char_len == y_char_len &&
293 0 : (x_char_len == 1 || rest_of_char_same(x, y, x_char_len)))
294 27 : sub = prev[i - 1];
295 : else
296 129 : sub = prev[i - 1] + sub_c;
297 :
298 : /* Take the one with minimum cost. */
299 156 : curr[i] = Min(ins, del);
300 156 : curr[i] = Min(curr[i], sub);
301 :
302 : /* Point to next character. */
303 156 : x += x_char_len;
304 : }
305 : }
306 : else
307 : {
308 13604 : for (; i < STOP_COLUMN; i++)
309 : {
310 : int ins;
311 : int del;
312 : int sub;
313 :
314 : /* Calculate costs for insertion, deletion, and substitution. */
315 10237 : ins = prev[i] + ins_c;
316 10237 : del = curr[i - 1] + del_c;
317 10237 : sub = prev[i - 1] + ((*x == *y) ? 0 : sub_c);
318 :
319 : /* Take the one with minimum cost. */
320 10237 : curr[i] = Min(ins, del);
321 10237 : curr[i] = Min(curr[i], sub);
322 :
323 : /* Point to next character. */
324 10237 : x++;
325 : }
326 : }
327 :
328 : /* Swap current row with previous row. */
329 3397 : temp = curr;
330 3397 : curr = prev;
331 3397 : prev = temp;
332 :
333 : /* Point to next character. */
334 12 : y += y_char_len;
335 :
336 : #ifdef LEVENSHTEIN_LESS_EQUAL
337 :
338 : /*
339 : * This chunk of code represents a significant performance hit if used
340 : * in the case where there is no max_d bound. This is probably not
341 : * because the max_d >= 0 test itself is expensive, but rather because
342 : * the possibility of needing to execute this code prevents tight
343 : * optimization of the loop as a whole.
344 : */
345 3385 : if (max_d >= 0)
346 : {
347 : /*
348 : * The "zero point" is the column of the current row where the
349 : * remaining portions of the strings are of equal length. There
350 : * are (n - 1) characters in the target string, of which j have
351 : * been transformed. There are (m - 1) characters in the source
352 : * string, so we want to find the value for zp where (n - 1) - j =
353 : * (m - 1) - zp.
354 : */
355 2817 : int zp = j - (n - m);
356 :
357 : /* Check whether the stop column can slide left. */
358 6774 : while (stop_column > 0)
359 : {
360 6198 : int ii = stop_column - 1;
361 6198 : int net_inserts = ii - zp;
362 :
363 10451 : if (prev[ii] + (net_inserts > 0 ? net_inserts * ins_c :
364 4253 : -net_inserts * del_c) <= max_d)
365 2241 : break;
366 3957 : stop_column--;
367 : }
368 :
369 : /* Check whether the start column can slide right. */
370 4639 : while (start_column < stop_column)
371 : {
372 4063 : int net_inserts = start_column - zp;
373 :
374 4063 : if (prev[start_column] +
375 4063 : (net_inserts > 0 ? net_inserts * ins_c :
376 3804 : -net_inserts * del_c) <= max_d)
377 2241 : break;
378 :
379 : /*
380 : * We'll never again update these values, so we must make sure
381 : * there's nothing here that could confuse any future
382 : * iteration of the outer loop.
383 : */
384 1822 : prev[start_column] = max_d + 1;
385 1822 : curr[start_column] = max_d + 1;
386 1822 : if (start_column != 0)
387 1251 : source += (s_char_len != NULL) ? s_char_len[start_column - 1] : 1;
388 1822 : start_column++;
389 : }
390 :
391 : /* If they cross, we're going to exceed the bound. */
392 2817 : if (start_column >= stop_column)
393 576 : return max_d + 1;
394 : }
395 : #endif
396 : }
397 :
398 : /*
399 : * Because the final value was swapped from the previous row to the
400 : * current row, that's where we'll find it.
401 : */
402 388 : return prev[m - 1];
403 : }
|
