PostgreSQL Source Code  git master
levenshtein.c File Reference
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros

#define MAX_LEVENSHTEIN_STRLEN   255
 
#define START_COLUMN   0
 
#define STOP_COLUMN   m
 

Functions

int varstr_levenshtein (const char *source, int slen, const char *target, int tlen, int ins_c, int del_c, int sub_c, bool trusted)
 

Macro Definition Documentation

◆ MAX_LEVENSHTEIN_STRLEN

#define MAX_LEVENSHTEIN_STRLEN   255

Definition at line 26 of file levenshtein.c.

◆ START_COLUMN

#define START_COLUMN   0

◆ STOP_COLUMN

#define STOP_COLUMN   m

Function Documentation

◆ varstr_levenshtein()

int varstr_levenshtein ( const char *  source,
int  slen,
const char *  target,
int  tlen,
int  ins_c,
int  del_c,
int  sub_c,
bool  trusted 
)

Definition at line 73 of file levenshtein.c.

78 {
79  int m,
80  n;
81  int *prev;
82  int *curr;
83  int *s_char_len = NULL;
84  int j;
85  const char *y;
86 
87  /*
88  * For varstr_levenshtein_less_equal, we have real variables called
89  * start_column and stop_column; otherwise it's just short-hand for 0 and
90  * m.
91  */
92 #ifdef LEVENSHTEIN_LESS_EQUAL
93  int start_column,
94  stop_column;
95 
96 #undef START_COLUMN
97 #undef STOP_COLUMN
98 #define START_COLUMN start_column
99 #define STOP_COLUMN stop_column
100 #else
101 #undef START_COLUMN
102 #undef STOP_COLUMN
103 #define START_COLUMN 0
104 #define STOP_COLUMN m
105 #endif
106 
107  /* Convert string lengths (in bytes) to lengths in characters */
108  m = pg_mbstrlen_with_len(source, slen);
109  n = pg_mbstrlen_with_len(target, tlen);
110 
111  /*
112  * We can transform an empty s into t with n insertions, or a non-empty t
113  * into an empty s with m deletions.
114  */
115  if (!m)
116  return n * ins_c;
117  if (!n)
118  return m * del_c;
119 
120  /*
121  * For security concerns, restrict excessive CPU+RAM usage. (This
122  * implementation uses O(m) memory and has O(mn) complexity.) If
123  * "trusted" is true, caller is responsible for not making excessive
124  * requests, typically by using a small max_d along with strings that are
125  * bounded, though not necessarily to MAX_LEVENSHTEIN_STRLEN exactly.
126  */
127  if (!trusted &&
128  (m > MAX_LEVENSHTEIN_STRLEN ||
130  ereport(ERROR,
131  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
132  errmsg("levenshtein argument exceeds maximum length of %d characters",
134 
135 #ifdef LEVENSHTEIN_LESS_EQUAL
136  /* Initialize start and stop columns. */
137  start_column = 0;
138  stop_column = m + 1;
139 
140  /*
141  * If max_d >= 0, determine whether the bound is impossibly tight. If so,
142  * return max_d + 1 immediately. Otherwise, determine whether it's tight
143  * enough to limit the computation we must perform. If so, figure out
144  * initial stop column.
145  */
146  if (max_d >= 0)
147  {
148  int min_theo_d; /* Theoretical minimum distance. */
149  int max_theo_d; /* Theoretical maximum distance. */
150  int net_inserts = n - m;
151 
152  min_theo_d = net_inserts < 0 ?
153  -net_inserts * del_c : net_inserts * ins_c;
154  if (min_theo_d > max_d)
155  return max_d + 1;
156  if (ins_c + del_c < sub_c)
157  sub_c = ins_c + del_c;
158  max_theo_d = min_theo_d + sub_c * Min(m, n);
159  if (max_d >= max_theo_d)
160  max_d = -1;
161  else if (ins_c + del_c > 0)
162  {
163  /*
164  * Figure out how much of the first row of the notional matrix we
165  * need to fill in. If the string is growing, the theoretical
166  * minimum distance already incorporates the cost of deleting the
167  * number of characters necessary to make the two strings equal in
168  * length. Each additional deletion forces another insertion, so
169  * the best-case total cost increases by ins_c + del_c. If the
170  * string is shrinking, the minimum theoretical cost assumes no
171  * excess deletions; that is, we're starting no further right than
172  * column n - m. If we do start further right, the best-case
173  * total cost increases by ins_c + del_c for each move right.
174  */
175  int slack_d = max_d - min_theo_d;
176  int best_column = net_inserts < 0 ? -net_inserts : 0;
177 
178  stop_column = best_column + (slack_d / (ins_c + del_c)) + 1;
179  if (stop_column > m)
180  stop_column = m + 1;
181  }
182  }
183 #endif
184 
185  /*
186  * In order to avoid calling pg_mblen() repeatedly on each character in s,
187  * we cache all the lengths before starting the main loop -- but if all
188  * the characters in both strings are single byte, then we skip this and
189  * use a fast-path in the main loop. If only one string contains
190  * multi-byte characters, we still build the array, so that the fast-path
191  * needn't deal with the case where the array hasn't been initialized.
192  */
193  if (m != slen || n != tlen)
194  {
195  int i;
196  const char *cp = source;
197 
198  s_char_len = (int *) palloc((m + 1) * sizeof(int));
199  for (i = 0; i < m; ++i)
200  {
201  s_char_len[i] = pg_mblen(cp);
202  cp += s_char_len[i];
203  }
204  s_char_len[i] = 0;
205  }
206 
207  /* One more cell for initialization column and row. */
208  ++m;
209  ++n;
210 
211  /* Previous and current rows of notional array. */
212  prev = (int *) palloc(2 * m * sizeof(int));
213  curr = prev + m;
214 
215  /*
216  * To transform the first i characters of s into the first 0 characters of
217  * t, we must perform i deletions.
218  */
219  for (int i = START_COLUMN; i < STOP_COLUMN; i++)
220  prev[i] = i * del_c;
221 
222  /* Loop through rows of the notional array */
223  for (y = target, j = 1; j < n; j++)
224  {
225  int *temp;
226  const char *x = source;
227  int y_char_len = n != tlen + 1 ? pg_mblen(y) : 1;
228  int i;
229 
230 #ifdef LEVENSHTEIN_LESS_EQUAL
231 
232  /*
233  * In the best case, values percolate down the diagonal unchanged, so
234  * we must increment stop_column unless it's already on the right end
235  * of the array. The inner loop will read prev[stop_column], so we
236  * have to initialize it even though it shouldn't affect the result.
237  */
238  if (stop_column < m)
239  {
240  prev[stop_column] = max_d + 1;
241  ++stop_column;
242  }
243 
244  /*
245  * The main loop fills in curr, but curr[0] needs a special case: to
246  * transform the first 0 characters of s into the first j characters
247  * of t, we must perform j insertions. However, if start_column > 0,
248  * this special case does not apply.
249  */
250  if (start_column == 0)
251  {
252  curr[0] = j * ins_c;
253  i = 1;
254  }
255  else
256  i = start_column;
257 #else
258  curr[0] = j * ins_c;
259  i = 1;
260 #endif
261 
262  /*
263  * This inner loop is critical to performance, so we include a
264  * fast-path to handle the (fairly common) case where no multibyte
265  * characters are in the mix. The fast-path is entitled to assume
266  * that if s_char_len is not initialized then BOTH strings contain
267  * only single-byte characters.
268  */
269  if (s_char_len != NULL)
270  {
271  for (; i < STOP_COLUMN; i++)
272  {
273  int ins;
274  int del;
275  int sub;
276  int x_char_len = s_char_len[i - 1];
277 
278  /*
279  * Calculate costs for insertion, deletion, and substitution.
280  *
281  * When calculating cost for substitution, we compare the last
282  * character of each possibly-multibyte character first,
283  * because that's enough to rule out most mis-matches. If we
284  * get past that test, then we compare the lengths and the
285  * remaining bytes.
286  */
287  ins = prev[i] + ins_c;
288  del = curr[i - 1] + del_c;
289  if (x[x_char_len - 1] == y[y_char_len - 1]
290  && x_char_len == y_char_len &&
291  (x_char_len == 1 || rest_of_char_same(x, y, x_char_len)))
292  sub = prev[i - 1];
293  else
294  sub = prev[i - 1] + sub_c;
295 
296  /* Take the one with minimum cost. */
297  curr[i] = Min(ins, del);
298  curr[i] = Min(curr[i], sub);
299 
300  /* Point to next character. */
301  x += x_char_len;
302  }
303  }
304  else
305  {
306  for (; i < STOP_COLUMN; i++)
307  {
308  int ins;
309  int del;
310  int sub;
311 
312  /* Calculate costs for insertion, deletion, and substitution. */
313  ins = prev[i] + ins_c;
314  del = curr[i - 1] + del_c;
315  sub = prev[i - 1] + ((*x == *y) ? 0 : sub_c);
316 
317  /* Take the one with minimum cost. */
318  curr[i] = Min(ins, del);
319  curr[i] = Min(curr[i], sub);
320 
321  /* Point to next character. */
322  x++;
323  }
324  }
325 
326  /* Swap current row with previous row. */
327  temp = curr;
328  curr = prev;
329  prev = temp;
330 
331  /* Point to next character. */
332  y += y_char_len;
333 
334 #ifdef LEVENSHTEIN_LESS_EQUAL
335 
336  /*
337  * This chunk of code represents a significant performance hit if used
338  * in the case where there is no max_d bound. This is probably not
339  * because the max_d >= 0 test itself is expensive, but rather because
340  * the possibility of needing to execute this code prevents tight
341  * optimization of the loop as a whole.
342  */
343  if (max_d >= 0)
344  {
345  /*
346  * The "zero point" is the column of the current row where the
347  * remaining portions of the strings are of equal length. There
348  * are (n - 1) characters in the target string, of which j have
349  * been transformed. There are (m - 1) characters in the source
350  * string, so we want to find the value for zp where (n - 1) - j =
351  * (m - 1) - zp.
352  */
353  int zp = j - (n - m);
354 
355  /* Check whether the stop column can slide left. */
356  while (stop_column > 0)
357  {
358  int ii = stop_column - 1;
359  int net_inserts = ii - zp;
360 
361  if (prev[ii] + (net_inserts > 0 ? net_inserts * ins_c :
362  -net_inserts * del_c) <= max_d)
363  break;
364  stop_column--;
365  }
366 
367  /* Check whether the start column can slide right. */
368  while (start_column < stop_column)
369  {
370  int net_inserts = start_column - zp;
371 
372  if (prev[start_column] +
373  (net_inserts > 0 ? net_inserts * ins_c :
374  -net_inserts * del_c) <= max_d)
375  break;
376 
377  /*
378  * We'll never again update these values, so we must make sure
379  * there's nothing here that could confuse any future
380  * iteration of the outer loop.
381  */
382  prev[start_column] = max_d + 1;
383  curr[start_column] = max_d + 1;
384  if (start_column != 0)
385  source += (s_char_len != NULL) ? s_char_len[start_column - 1] : 1;
386  start_column++;
387  }
388 
389  /* If they cross, we're going to exceed the bound. */
390  if (start_column >= stop_column)
391  return max_d + 1;
392  }
393 #endif
394  }
395 
396  /*
397  * Because the final value was swapped from the previous row to the
398  * current row, that's where we'll find it.
399  */
400  return prev[m - 1];
401 }
#define Min(x, y)
Definition: c.h:1004
int errcode(int sqlerrcode)
Definition: elog.c:859
int errmsg(const char *fmt,...)
Definition: elog.c:1072
#define ERROR
Definition: elog.h:39
#define ereport(elevel,...)
Definition: elog.h:149
int y
Definition: isn.c:72
int x
Definition: isn.c:71
int j
Definition: isn.c:74
int i
Definition: isn.c:73
#define START_COLUMN
#define MAX_LEVENSHTEIN_STRLEN
Definition: levenshtein.c:26
#define STOP_COLUMN
int pg_mbstrlen_with_len(const char *mbstr, int limit)
Definition: mbutils.c:1057
int pg_mblen(const char *mbstr)
Definition: mbutils.c:1023
void * palloc(Size size)
Definition: mcxt.c:1316
static rewind_source * source
Definition: pg_rewind.c:89
static bool rest_of_char_same(const char *s1, const char *s2, int len)
Definition: varlena.c:6152

References ereport, errcode(), errmsg(), ERROR, i, j, MAX_LEVENSHTEIN_STRLEN, Min, palloc(), pg_mblen(), pg_mbstrlen_with_len(), rest_of_char_same(), source, START_COLUMN, STOP_COLUMN, x, and y.

Referenced by levenshtein(), and levenshtein_with_costs().