1 files changed, 453 insertions, 48 deletions

diff --git a/contrib/pg_trgm/trgm_op.c b/contrib/pg_trgm/trgm_op.c
index b2c1f6b0612..eaf20e45159 100644
--- a/contrib/pg_trgm/trgm_op.c
+++ b/contrib/pg_trgm/trgm_op.c
@@ -15,7 +15,8 @@
 PG_MODULE_MAGIC;
 
 /* GUC variables */
-double		similarity_threshold = 0.3f;
+double similarity_threshold = 0.3f;
+double word_similarity_threshold = 0.6f;
 
 void		_PG_init(void);
 
@@ -23,8 +24,20 @@ PG_FUNCTION_INFO_V1(set_limit);
 PG_FUNCTION_INFO_V1(show_limit);
 PG_FUNCTION_INFO_V1(show_trgm);
 PG_FUNCTION_INFO_V1(similarity);
+PG_FUNCTION_INFO_V1(word_similarity);
 PG_FUNCTION_INFO_V1(similarity_dist);
 PG_FUNCTION_INFO_V1(similarity_op);
+PG_FUNCTION_INFO_V1(word_similarity_op);
+PG_FUNCTION_INFO_V1(word_similarity_commutator_op);
+PG_FUNCTION_INFO_V1(word_similarity_dist_op);
+PG_FUNCTION_INFO_V1(word_similarity_dist_commutator_op);
+
+/* Trigram with position */
+typedef struct
+{
+	trgm	trg;
+	int		index;
+} pos_trgm;
 
 /*
  * Module load callback
@@ -45,11 +58,23 @@ _PG_init(void)
 							NULL,
 							NULL,
 							NULL);
+	DefineCustomRealVariable("pg_trgm.word_similarity_threshold",
+							"Sets the threshold used by the <%% operator.",
+							"Valid range is 0.0 .. 1.0.",
+							&word_similarity_threshold,
+							0.6,
+							0.0,
+							1.0,
+							PGC_USERSET,
+							0,
+							NULL,
+							NULL,
+							NULL);
 }
 
 /*
  * Deprecated function.
- * Use "pg_trgm.similarity_threshold" GUC variable instead of this function
+ * Use "pg_trgm.similarity_threshold" GUC variable instead of this function.
  */
 Datum
 set_limit(PG_FUNCTION_ARGS)
@@ -59,14 +84,14 @@ set_limit(PG_FUNCTION_ARGS)
 	if (nlimit < 0 || nlimit > 1.0)
 		ereport(ERROR,
 				(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-				 errmsg("wrong limit, should be between 0 and 1")));
+				 errmsg("wrong threshold, should be between 0 and 1")));
 	similarity_threshold = nlimit;
 	PG_RETURN_FLOAT4(similarity_threshold);
 }
 
 /*
  * Deprecated function.
- * Use "pg_trgm.similarity_threshold" GUC variable instead of this function
+ * Use "pg_trgm.similarity_threshold" GUC variable instead of this function.
  */
 Datum
 show_limit(PG_FUNCTION_ARGS)
@@ -199,38 +224,28 @@ make_trigrams(trgm *tptr, char *str, int bytelen, int charlen)
 	return tptr;
 }
 
-TRGM *
-generate_trgm(char *str, int slen)
+/*
+ * Make array of trigrams without sorting and removing duplicate items.
+ *
+ * trg: where to return the array of trigrams.
+ * str: source string, of length slen bytes.
+ *
+ * Returns length of the generated array.
+ */
+static int
+generate_trgm_only(trgm *trg, char *str, int slen)
 {
-	TRGM	   *trg;
-	char	   *buf;
 	trgm	   *tptr;
-	int			len,
-				charlen,
+	char	   *buf;
+	int			charlen,
 				bytelen;
 	char	   *bword,
 			   *eword;
 
-	/*
-	 * Guard against possible overflow in the palloc requests below.  (We
-	 * don't worry about the additive constants, since palloc can detect
-	 * requests that are a little above MaxAllocSize --- we just need to
-	 * prevent integer overflow in the multiplications.)
-	 */
-	if ((Size) (slen / 2) >= (MaxAllocSize / (sizeof(trgm) * 3)) ||
-		(Size) slen >= (MaxAllocSize / pg_database_encoding_max_length()))
-		ereport(ERROR,
-				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("out of memory")));
-
-	trg = (TRGM *) palloc(TRGMHDRSIZE + sizeof(trgm) * (slen / 2 + 1) *3);
-	trg->flag = ARRKEY;
-	SET_VARSIZE(trg, TRGMHDRSIZE);
-
 	if (slen + LPADDING + RPADDING < 3 || slen == 0)
-		return trg;
+		return 0;
 
-	tptr = GETARR(trg);
+	tptr = trg;
 
 	/* Allocate a buffer for case-folded, blank-padded words */
 	buf = (char *) palloc(slen * pg_database_encoding_max_length() + 4);
@@ -270,7 +285,47 @@ generate_trgm(char *str, int slen)
 
 	pfree(buf);
 
-	if ((len = tptr - GETARR(trg)) == 0)
+	return tptr - trg;
+}
+
+/*
+ * Guard against possible overflow in the palloc requests below.  (We
+ * don't worry about the additive constants, since palloc can detect
+ * requests that are a little above MaxAllocSize --- we just need to
+ * prevent integer overflow in the multiplications.)
+ */
+static void
+protect_out_of_mem(int slen)
+{
+	if ((Size) (slen / 2) >= (MaxAllocSize / (sizeof(trgm) * 3)) ||
+		(Size) slen >= (MaxAllocSize / pg_database_encoding_max_length()))
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("out of memory")));
+}
+
+/*
+ * Make array of trigrams with sorting and removing duplicate items.
+ *
+ * str: source string, of length slen bytes.
+ *
+ * Returns the sorted array of unique trigrams.
+ */
+TRGM *
+generate_trgm(char *str, int slen)
+{
+	TRGM	   *trg;
+	int			len;
+
+	protect_out_of_mem(slen);
+
+	trg = (TRGM *) palloc(TRGMHDRSIZE + sizeof(trgm) * (slen / 2 + 1) *3);
+	trg->flag = ARRKEY;
+
+	len = generate_trgm_only(GETARR(trg), str, slen);
+	SET_VARSIZE(trg, CALCGTSIZE(ARRKEY, len));
+
+	if (len == 0)
 		return trg;
 
 	/*
@@ -288,6 +343,285 @@ generate_trgm(char *str, int slen)
 }
 
 /*
+ * Make array of positional trigrams from two trigram arrays trg1 and trg2.
+ *
+ * trg1: trigram array of search pattern, of length len1. trg1 is required
+ *       word which positions don't matter and replaced with -1.
+ * trg2: trigram array of text, of length len2. trg2 is haystack where we
+ *       search and have to store its positions.
+ *
+ * Returns concatenated trigram array.
+ */
+static pos_trgm *
+make_positional_trgm(trgm *trg1, int len1, trgm *trg2, int len2)
+{
+	pos_trgm   *result;
+	int			i, len = len1 + len2;
+
+	result = (pos_trgm *) palloc(sizeof(pos_trgm) * len);
+
+	for (i = 0; i < len1; i++)
+	{
+		memcpy(&result[i].trg, &trg1[i], sizeof(trgm));
+		result[i].index = -1;
+	}
+
+	for (i = 0; i < len2; i++)
+	{
+		memcpy(&result[i + len1].trg, &trg2[i], sizeof(trgm));
+		result[i + len1].index = i;
+	}
+
+	return result;
+}
+
+/*
+ * Compare position trigrams: compare trigrams first and position second.
+ */
+static int
+comp_ptrgm(const void *v1, const void *v2)
+{
+	const pos_trgm *p1 = (const pos_trgm *)v1;
+	const pos_trgm *p2 = (const pos_trgm *)v2;
+	int				cmp;
+
+	cmp = CMPTRGM(p1->trg, p2->trg);
+	if (cmp != 0)
+		return cmp;
+
+	if (p1->index < p2->index)
+		return -1;
+	else if (p1->index == p2->index)
+		return 0;
+	else
+		return 1;
+}
+
+/*
+ * Iterative search function which calculates maximum similarity with word in
+ * the string. But maximum similarity is calculated only if check_only == false.
+ *
+ * trg2indexes: array which stores indexes of the array "found".
+ * found: array which stores true of false values.
+ * ulen1: count of unique trigrams of array "trg1".
+ * len2: length of array "trg2" and array "trg2indexes".
+ * len: length of the array "found".
+ * check_only: if true then only check existaince of similar search pattern in
+ *             text.
+ *
+ * Returns word similarity.
+ */
+static float4
+iterate_word_similarity(int *trg2indexes,
+						bool *found,
+						int ulen1,
+						int len2,
+						int len,
+						bool check_only)
+{
+	int		   *lastpos,
+				i,
+				ulen2 = 0,
+				count = 0,
+				upper = -1,
+				lower = -1;
+	float4		smlr_cur,
+				smlr_max = 0.0f;
+
+	/* Memorise last position of each trigram */
+	lastpos = (int *) palloc(sizeof(int) * len);
+	memset(lastpos, -1, sizeof(int) * len);
+
+	for (i = 0; i < len2; i++)
+	{
+		/* Get index of next trigram */
+		int	trgindex = trg2indexes[i];
+
+		/* Update last position of this trigram */
+		if (lower >= 0 || found[trgindex])
+		{
+			if (lastpos[trgindex] < 0)
+			{
+				ulen2++;
+				if (found[trgindex])
+					count++;
+			}
+			lastpos[trgindex] = i;
+		}
+
+		/* Adjust lower bound if this trigram is present in required substing */
+		if (found[trgindex])
+		{
+			int		prev_lower,
+					tmp_ulen2,
+					tmp_lower,
+					tmp_count;
+
+			upper = i;
+			if (lower == -1)
+			{
+				lower = i;
+				ulen2 = 1;
+			}
+
+			smlr_cur = CALCSML(count, ulen1, ulen2);
+
+			/* Also try to adjust upper bound for greater similarity */
+			tmp_count = count;
+			tmp_ulen2 = ulen2;
+			prev_lower = lower;
+			for (tmp_lower = lower; tmp_lower <= upper; tmp_lower++)
+			{
+				float	smlr_tmp = CALCSML(tmp_count, ulen1, tmp_ulen2);
+				int		tmp_trgindex;
+
+				if (smlr_tmp > smlr_cur)
+				{
+					smlr_cur = smlr_tmp;
+					ulen2 = tmp_ulen2;
+					lower = tmp_lower;
+					count = tmp_count;
+				}
+				/*
+				 * if we only check that word similarity is greater than
+				 * pg_trgm.word_similarity_threshold we do not need to calculate
+				 * a maximum similarity.
+				 */
+				if (check_only && smlr_cur >= word_similarity_threshold)
+					break;
+
+				tmp_trgindex = trg2indexes[tmp_lower];
+				if (lastpos[tmp_trgindex] == tmp_lower)
+				{
+					tmp_ulen2--;
+					if (found[tmp_trgindex])
+						tmp_count--;
+				}
+			}
+
+			smlr_max = Max(smlr_max, smlr_cur);
+			/*
+			 * if we only check that word similarity is greater than
+			 * pg_trgm.word_similarity_threshold we do not need to calculate a
+			 * maximum similarity
+			 */
+			if (check_only && smlr_max >= word_similarity_threshold)
+				break;
+
+			for (tmp_lower = prev_lower; tmp_lower < lower; tmp_lower++)
+			{
+				int		tmp_trgindex;
+				tmp_trgindex = trg2indexes[tmp_lower];
+				if (lastpos[tmp_trgindex] == tmp_lower)
+					lastpos[tmp_trgindex] = -1;
+			}
+		}
+	}
+
+	pfree(lastpos);
+
+	return smlr_max;
+}
+
+/*
+ * Calculate word similarity.
+ * This function prepare two arrays: "trg2indexes" and "found". Then this arrays
+ * are used to calculate word similarity using iterate_word_similarity().
+ *
+ * "trg2indexes" is array which stores indexes of the array "found".
+ * In other words:
+ * trg2indexes[j] = i;
+ * found[i] = true (or false);
+ * If found[i] == true then there is trigram trg2[j] in array "trg1".
+ * If found[i] == false then there is not trigram trg2[j] in array "trg1".
+ *
+ * str1: search pattern string, of length slen1 bytes.
+ * str2: text in which we are looking for a word, of length slen2 bytes.
+ * check_only: if true then only check existaince of similar search pattern in
+ *             text.
+ *
+ * Returns word similarity.
+ */
+static float4
+calc_word_similarity(char *str1, int slen1, char *str2, int slen2,
+						  bool check_only)
+{
+	bool	   *found;
+	pos_trgm   *ptrg;
+	trgm	   *trg1;
+	trgm	   *trg2;
+	int			len1,
+				len2,
+				len,
+				i,
+				j,
+				ulen1;
+	int		   *trg2indexes;
+	float4		result;
+
+	protect_out_of_mem(slen1 + slen2);
+
+	/* Make positional trigrams */
+	trg1 = (trgm *) palloc(sizeof(trgm) * (slen1 / 2 + 1) * 3);
+	trg2 = (trgm *) palloc(sizeof(trgm) * (slen2 / 2 + 1) * 3);
+
+	len1 = generate_trgm_only(trg1, str1, slen1);
+	len2 = generate_trgm_only(trg2, str2, slen2);
+
+	ptrg = make_positional_trgm(trg1, len1, trg2, len2);
+	len = len1 + len2;
+	qsort(ptrg, len, sizeof(pos_trgm), comp_ptrgm);
+
+	pfree(trg1);
+	pfree(trg2);
+
+	/*
+	 * Merge positional trigrams array: enumerate each trigram and find its
+	 * presence in required word.
+	 */
+	trg2indexes = (int *) palloc(sizeof(int) * len2);
+	found = (bool *) palloc0(sizeof(bool) * len);
+
+	ulen1 = 0;
+	j = 0;
+	for (i = 0; i < len; i++)
+	{
+		if (i > 0)
+		{
+			int cmp = CMPTRGM(ptrg[i - 1].trg, ptrg[i].trg);
+			if (cmp != 0)
+			{
+				if (found[j])
+					ulen1++;
+				j++;
+			}
+		}
+
+		if (ptrg[i].index >= 0)
+		{
+			trg2indexes[ptrg[i].index] = j;
+		}
+		else
+		{
+			found[j] = true;
+		}
+	}
+	if (found[j])
+		ulen1++;
+
+	/* Run iterative procedure to find maximum similarity with word */
+	result = iterate_word_similarity(trg2indexes, found, ulen1, len2, len,
+										  check_only);
+
+	pfree(trg2indexes);
+	pfree(found);
+	pfree(ptrg);
+
+	return result;
+}
+
+
+/*
  * Extract the next non-wildcard part of a search string, ie, a word bounded
  * by '_' or '%' meta-characters, non-word characters or string end.
  *
@@ -459,17 +793,7 @@ generate_wildcard_trgm(const char *str, int slen)
 				bytelen;
 	const char *eword;
 
-	/*
-	 * Guard against possible overflow in the palloc requests below.  (We
-	 * don't worry about the additive constants, since palloc can detect
-	 * requests that are a little above MaxAllocSize --- we just need to
-	 * prevent integer overflow in the multiplications.)
-	 */
-	if ((Size) (slen / 2) >= (MaxAllocSize / (sizeof(trgm) * 3)) ||
-		(Size) slen >= (MaxAllocSize / pg_database_encoding_max_length()))
-		ereport(ERROR,
-				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("out of memory")));
+	protect_out_of_mem(slen);
 
 	trg = (TRGM *) palloc(TRGMHDRSIZE + sizeof(trgm) * (slen / 2 + 1) *3);
 	trg->flag = ARRKEY;
@@ -590,7 +914,7 @@ show_trgm(PG_FUNCTION_ARGS)
 }
 
 float4
-cnt_sml(TRGM *trg1, TRGM *trg2)
+cnt_sml(TRGM *trg1, TRGM *trg2, bool inexact)
 {
 	trgm	   *ptr1,
 			   *ptr2;
@@ -624,14 +948,15 @@ cnt_sml(TRGM *trg1, TRGM *trg2)
 		}
 	}
 
-#ifdef DIVUNION
-	return ((float4) count) / ((float4) (len1 + len2 - count));
-#else
-	return ((float4) count) / ((float4) ((len1 > len2) ? len1 : len2));
-#endif
-
+	/*
+	 * If inexact then len2 is equal to count, because we don't know actual
+	 * length of second string in inexact search and we can assume that count
+	 * is a lower bound of len2.
+	 */
+	return CALCSML(count, len1, inexact ? count : len2);
 }
 
+
 /*
  * Returns whether trg2 contains all trigrams in trg1.
  * This relies on the trigram arrays being sorted.
@@ -726,7 +1051,7 @@ similarity(PG_FUNCTION_ARGS)
 	trg1 = generate_trgm(VARDATA(in1), VARSIZE(in1) - VARHDRSZ);
 	trg2 = generate_trgm(VARDATA(in2), VARSIZE(in2) - VARHDRSZ);
 
-	res = cnt_sml(trg1, trg2);
+	res = cnt_sml(trg1, trg2, false);
 
 	pfree(trg1);
 	pfree(trg2);
@@ -737,6 +1062,22 @@ similarity(PG_FUNCTION_ARGS)
 }
 
 Datum
+word_similarity(PG_FUNCTION_ARGS)
+{
+	text	   *in1 = PG_GETARG_TEXT_PP(0);
+	text	   *in2 = PG_GETARG_TEXT_PP(1);
+	float4		res;
+
+	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
+								VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
+								false);
+
+	PG_FREE_IF_COPY(in1, 0);
+	PG_FREE_IF_COPY(in2, 1);
+	PG_RETURN_FLOAT4(res);
+}
+
+Datum
 similarity_dist(PG_FUNCTION_ARGS)
 {
 	float4		res = DatumGetFloat4(DirectFunctionCall2(similarity,
@@ -755,3 +1096,67 @@ similarity_op(PG_FUNCTION_ARGS)
 
 	PG_RETURN_BOOL(res >= similarity_threshold);
 }
+
+Datum
+word_similarity_op(PG_FUNCTION_ARGS)
+{
+	text	   *in1 = PG_GETARG_TEXT_PP(0);
+	text	   *in2 = PG_GETARG_TEXT_PP(1);
+	float4		res;
+
+	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
+								VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
+								true);
+
+	PG_FREE_IF_COPY(in1, 0);
+	PG_FREE_IF_COPY(in2, 1);
+	PG_RETURN_BOOL(res >= word_similarity_threshold);
+}
+
+Datum
+word_similarity_commutator_op(PG_FUNCTION_ARGS)
+{
+	text	   *in1 = PG_GETARG_TEXT_PP(0);
+	text	   *in2 = PG_GETARG_TEXT_PP(1);
+	float4		res;
+
+	res = calc_word_similarity(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
+								VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
+								true);
+
+	PG_FREE_IF_COPY(in1, 0);
+	PG_FREE_IF_COPY(in2, 1);
+	PG_RETURN_BOOL(res >= word_similarity_threshold);
+}
+
+Datum
+word_similarity_dist_op(PG_FUNCTION_ARGS)
+{
+	text	   *in1 = PG_GETARG_TEXT_PP(0);
+	text	   *in2 = PG_GETARG_TEXT_PP(1);
+	float4		res;
+
+	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
+								VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
+								false);
+
+	PG_FREE_IF_COPY(in1, 0);
+	PG_FREE_IF_COPY(in2, 1);
+	PG_RETURN_FLOAT4(1.0 - res);
+}
+
+Datum
+word_similarity_dist_commutator_op(PG_FUNCTION_ARGS)
+{
+	text	   *in1 = PG_GETARG_TEXT_PP(0);
+	text	   *in2 = PG_GETARG_TEXT_PP(1);
+	float4		res;
+
+	res = calc_word_similarity(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
+								VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
+								false);
+
+	PG_FREE_IF_COPY(in1, 0);
+	PG_FREE_IF_COPY(in2, 1);
+	PG_RETURN_FLOAT4(1.0 - res);
+}