This graph shows which files directly or indirectly include 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.
 {
     int         m,
                 n;
     int        *prev;
     int        *curr;
     int        *s_char_len = NULL;
     int         j;
     const char *y;
  
     /*
      * For varstr_levenshtein_less_equal, we have real variables called
      * start_column and stop_column; otherwise it's just short-hand for 0 and
      * m.
      */
 #ifdef LEVENSHTEIN_LESS_EQUAL
     int         start_column,
                 stop_column;
  
 #undef START_COLUMN
 #undef STOP_COLUMN
 #define START_COLUMN start_column
 #define STOP_COLUMN stop_column
 #else
 #undef START_COLUMN
 #undef STOP_COLUMN
 #define START_COLUMN 0
 #define STOP_COLUMN m
 #endif
  
     /* Convert string lengths (in bytes) to lengths in characters */
     m = pg_mbstrlen_with_len(source, slen);
     n = pg_mbstrlen_with_len(target, tlen);
  
     /*
      * We can transform an empty s into t with n insertions, or a non-empty t
      * into an empty s with m deletions.
      */
     if (!m)
         return n * ins_c;
     if (!n)
         return m * del_c;
  
     /*
      * For security concerns, restrict excessive CPU+RAM usage. (This
      * implementation uses O(m) memory and has O(mn) complexity.)  If
      * "trusted" is true, caller is responsible for not making excessive
      * requests, typically by using a small max_d along with strings that are
      * bounded, though not necessarily to MAX_LEVENSHTEIN_STRLEN exactly.
      */
     if (!trusted &&
         (m > MAX_LEVENSHTEIN_STRLEN ||
          n > MAX_LEVENSHTEIN_STRLEN))
         ereport(ERROR,
                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
                  errmsg("levenshtein argument exceeds maximum length of %d characters",
                         MAX_LEVENSHTEIN_STRLEN)));
  
 #ifdef LEVENSHTEIN_LESS_EQUAL
     /* Initialize start and stop columns. */
     start_column = 0;
     stop_column = m + 1;
  
     /*
      * If max_d >= 0, determine whether the bound is impossibly tight.  If so,
      * return max_d + 1 immediately.  Otherwise, determine whether it's tight
      * enough to limit the computation we must perform.  If so, figure out
      * initial stop column.
      */
     if (max_d >= 0)
     {
         int         min_theo_d; /* Theoretical minimum distance. */
         int         max_theo_d; /* Theoretical maximum distance. */
         int         net_inserts = n - m;
  
         min_theo_d = net_inserts < 0 ?
             -net_inserts * del_c : net_inserts * ins_c;
         if (min_theo_d > max_d)
             return max_d + 1;
         if (ins_c + del_c < sub_c)
             sub_c = ins_c + del_c;
         max_theo_d = min_theo_d + sub_c * Min(m, n);
         if (max_d >= max_theo_d)
             max_d = -1;
         else if (ins_c + del_c > 0)
         {
             /*
              * Figure out how much of the first row of the notional matrix we
              * need to fill in.  If the string is growing, the theoretical
              * minimum distance already incorporates the cost of deleting the
              * number of characters necessary to make the two strings equal in
              * length.  Each additional deletion forces another insertion, so
              * the best-case total cost increases by ins_c + del_c. If the
              * string is shrinking, the minimum theoretical cost assumes no
              * excess deletions; that is, we're starting no further right than
              * column n - m.  If we do start further right, the best-case
              * total cost increases by ins_c + del_c for each move right.
              */
             int         slack_d = max_d - min_theo_d;
             int         best_column = net_inserts < 0 ? -net_inserts : 0;
  
             stop_column = best_column + (slack_d / (ins_c + del_c)) + 1;
             if (stop_column > m)
                 stop_column = m + 1;
         }
     }
 #endif
  
     /*
      * In order to avoid calling pg_mblen() repeatedly on each character in s,
      * we cache all the lengths before starting the main loop -- but if all
      * the characters in both strings are single byte, then we skip this and
      * use a fast-path in the main loop.  If only one string contains
      * multi-byte characters, we still build the array, so that the fast-path
      * needn't deal with the case where the array hasn't been initialized.
      */
     if (m != slen || n != tlen)
     {
         int         i;
         const char *cp = source;
  
         s_char_len = (int *) palloc((m + 1) * sizeof(int));
         for (i = 0; i < m; ++i)
         {
             s_char_len[i] = pg_mblen(cp);
             cp += s_char_len[i];
         }
         s_char_len[i] = 0;
     }
  
     /* One more cell for initialization column and row. */
     ++m;
     ++n;
  
     /* Previous and current rows of notional array. */
     prev = (int *) palloc(2 * m * sizeof(int));
     curr = prev + m;
  
     /*
      * To transform the first i characters of s into the first 0 characters of
      * t, we must perform i deletions.
      */
     for (int i = START_COLUMN; i < STOP_COLUMN; i++)
         prev[i] = i * del_c;
  
     /* Loop through rows of the notional array */
     for (y = target, j = 1; j < n; j++)
     {
         int        *temp;
         const char *x = source;
         int         y_char_len = n != tlen + 1 ? pg_mblen(y) : 1;
         int         i;
  
 #ifdef LEVENSHTEIN_LESS_EQUAL
  
         /*
          * In the best case, values percolate down the diagonal unchanged, so
          * we must increment stop_column unless it's already on the right end
          * of the array.  The inner loop will read prev[stop_column], so we
          * have to initialize it even though it shouldn't affect the result.
          */
         if (stop_column < m)
         {
             prev[stop_column] = max_d + 1;
             ++stop_column;
         }
  
         /*
          * The main loop fills in curr, but curr[0] needs a special case: to
          * transform the first 0 characters of s into the first j characters
          * of t, we must perform j insertions.  However, if start_column > 0,
          * this special case does not apply.
          */
         if (start_column == 0)
         {
             curr[0] = j * ins_c;
             i = 1;
         }
         else
             i = start_column;
 #else
         curr[0] = j * ins_c;
         i = 1;
 #endif
  
         /*
          * This inner loop is critical to performance, so we include a
          * fast-path to handle the (fairly common) case where no multibyte
          * characters are in the mix.  The fast-path is entitled to assume
          * that if s_char_len is not initialized then BOTH strings contain
          * only single-byte characters.
          */
         if (s_char_len != NULL)
         {
             for (; i < STOP_COLUMN; i++)
             {
                 int         ins;
                 int         del;
                 int         sub;
                 int         x_char_len = s_char_len[i - 1];
  
                 /*
                  * Calculate costs for insertion, deletion, and substitution.
                  *
                  * When calculating cost for substitution, we compare the last
                  * character of each possibly-multibyte character first,
                  * because that's enough to rule out most mis-matches.  If we
                  * get past that test, then we compare the lengths and the
                  * remaining bytes.
                  */
                 ins = prev[i] + ins_c;
                 del = curr[i - 1] + del_c;
                 if (x[x_char_len - 1] == y[y_char_len - 1]
                     && x_char_len == y_char_len &&
                     (x_char_len == 1 || rest_of_char_same(x, y, x_char_len)))
                     sub = prev[i - 1];
                 else
                     sub = prev[i - 1] + sub_c;
  
                 /* Take the one with minimum cost. */
                 curr[i] = Min(ins, del);
                 curr[i] = Min(curr[i], sub);
  
                 /* Point to next character. */
                 x += x_char_len;
             }
         }
         else
         {
             for (; i < STOP_COLUMN; i++)
             {
                 int         ins;
                 int         del;
                 int         sub;
  
                 /* Calculate costs for insertion, deletion, and substitution. */
                 ins = prev[i] + ins_c;
                 del = curr[i - 1] + del_c;
                 sub = prev[i - 1] + ((*x == *y) ? 0 : sub_c);
  
                 /* Take the one with minimum cost. */
                 curr[i] = Min(ins, del);
                 curr[i] = Min(curr[i], sub);
  
                 /* Point to next character. */
                 x++;
             }
         }
  
         /* Swap current row with previous row. */
         temp = curr;
         curr = prev;
         prev = temp;
  
         /* Point to next character. */
         y += y_char_len;
  
 #ifdef LEVENSHTEIN_LESS_EQUAL
  
         /*
          * This chunk of code represents a significant performance hit if used
          * in the case where there is no max_d bound.  This is probably not
          * because the max_d >= 0 test itself is expensive, but rather because
          * the possibility of needing to execute this code prevents tight
          * optimization of the loop as a whole.
          */
         if (max_d >= 0)
         {
             /*
              * The "zero point" is the column of the current row where the
              * remaining portions of the strings are of equal length.  There
              * are (n - 1) characters in the target string, of which j have
              * been transformed.  There are (m - 1) characters in the source
              * string, so we want to find the value for zp where (n - 1) - j =
              * (m - 1) - zp.
              */
             int         zp = j - (n - m);
  
             /* Check whether the stop column can slide left. */
             while (stop_column > 0)
             {
                 int         ii = stop_column - 1;
                 int         net_inserts = ii - zp;
  
                 if (prev[ii] + (net_inserts > 0 ? net_inserts * ins_c :
                                 -net_inserts * del_c) <= max_d)
                     break;
                 stop_column--;
             }
  
             /* Check whether the start column can slide right. */
             while (start_column < stop_column)
             {
                 int         net_inserts = start_column - zp;
  
                 if (prev[start_column] +
                     (net_inserts > 0 ? net_inserts * ins_c :
                      -net_inserts * del_c) <= max_d)
                     break;
  
                 /*
                  * We'll never again update these values, so we must make sure
                  * there's nothing here that could confuse any future
                  * iteration of the outer loop.
                  */
                 prev[start_column] = max_d + 1;
                 curr[start_column] = max_d + 1;
                 if (start_column != 0)
                     source += (s_char_len != NULL) ? s_char_len[start_column - 1] : 1;
                 start_column++;
             }
  
             /* If they cross, we're going to exceed the bound. */
             if (start_column >= stop_column)
                 return max_d + 1;
         }
 #endif
     }
  
     /*
      * Because the final value was swapped from the previous row to the
      * current row, that's where we'll find it.
      */
     return prev[m - 1];
 }
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().