Add word_similarity to pg_trgm contrib module.

Patch introduces a concept of similarity over string and just a word from
another string.

Version of extension is not changed because 1.2 was already introduced in 9.6
release cycle, so, there wasn't a public version.

Author: Alexander Korotkov, Artur Zakirov

contrib/pg_trgm/Makefile

@@ -7,7 +7,7 @@ EXTENSION = pg_trgm
 DATA = pg_trgm--1.2.sql pg_trgm--1.0--1.1.sql pg_trgm--1.1--1.2.sql pg_trgm--unpackaged--1.0.sql
 PGFILEDESC = "pg_trgm - trigram matching"
 
-REGRESS = pg_trgm
+REGRESS = pg_trgm pg_word_trgm
 
 ifdef USE_PGXS
 PG_CONFIG = pg_config

contrib/pg_trgm/expected/pg_trgm.out

@@ -59,7 +59,7 @@ select similarity('---', '####---');
           0
 (1 row)
 
-CREATE TABLE test_trgm(t text);
+CREATE TABLE test_trgm(t text COLLATE "C");
 \copy test_trgm from 'data/trgm.data'
 select t,similarity(t,'qwertyu0988') as sml from test_trgm where t % 'qwertyu0988' order by sml desc, t;
       t      |   sml    
@@ -3467,7 +3467,7 @@ select t,similarity(t,'gwertyu1988') as sml from test_trgm where t % 'gwertyu198
  qwertyu0988 | 0.333333
 (1 row)
 
-create table test2(t text);
+create table test2(t text COLLATE "C");
 insert into test2 values ('abcdef');
 insert into test2 values ('quark');
 insert into test2 values ('  z foo bar');

contrib/pg_trgm/pg_trgm--1.1--1.2.sql

@@ -3,10 +3,72 @@
 -- complain if script is sourced in psql, rather than via ALTER EXTENSION
 \echo Use "ALTER EXTENSION pg_trgm UPDATE TO '1.2'" to load this file. \quit
 
+CREATE FUNCTION word_similarity(text,text)
+RETURNS float4
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT IMMUTABLE;
+
+CREATE FUNCTION word_similarity_op(text,text)
+RETURNS bool
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT STABLE;  -- stable because depends on pg_trgm.word_similarity_threshold
+
+CREATE FUNCTION word_similarity_commutator_op(text,text)
+RETURNS bool
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT STABLE;  -- stable because depends on pg_trgm.word_similarity_threshold
+
+CREATE FUNCTION word_similarity_dist_op(text,text)
+RETURNS float4
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT IMMUTABLE;
+
+CREATE FUNCTION word_similarity_dist_commutator_op(text,text)
+RETURNS float4
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT IMMUTABLE;
+
+CREATE OPERATOR <% (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_op,
+        COMMUTATOR = '%>',
+        RESTRICT = contsel,
+        JOIN = contjoinsel
+);
+
+CREATE OPERATOR %> (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_commutator_op,
+        COMMUTATOR = '<%',
+        RESTRICT = contsel,
+        JOIN = contjoinsel
+);
+
+CREATE OPERATOR <<-> (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_dist_op,
+        COMMUTATOR = '<->>'
+);
+
+CREATE OPERATOR <->> (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_dist_commutator_op,
+        COMMUTATOR = '<<->'
+);
+
 CREATE FUNCTION gin_trgm_triconsistent(internal, int2, text, int4, internal, internal, internal)
 RETURNS "char"
 AS 'MODULE_PATHNAME'
 LANGUAGE C IMMUTABLE STRICT;
 
+ALTER OPERATOR FAMILY gist_trgm_ops USING gist ADD
+        OPERATOR        7       %> (text, text),
+        OPERATOR        8       <->> (text, text) FOR ORDER BY pg_catalog.float_ops;
+
 ALTER OPERATOR FAMILY gin_trgm_ops USING gin ADD
+        OPERATOR        7       %> (text, text),
         FUNCTION        6      (text, text)   gin_trgm_triconsistent (internal, int2, text, int4, internal, internal, internal);

contrib/pg_trgm/pg_trgm--1.2.sql

@@ -39,6 +39,39 @@ CREATE OPERATOR % (
         JOIN = contjoinsel
 );
 
+CREATE FUNCTION word_similarity(text,text)
+RETURNS float4
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT IMMUTABLE;
+
+CREATE FUNCTION word_similarity_op(text,text)
+RETURNS bool
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT STABLE;  -- stable because depends on pg_trgm.word_similarity_threshold
+
+CREATE FUNCTION word_similarity_commutator_op(text,text)
+RETURNS bool
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT STABLE;  -- stable because depends on pg_trgm.word_similarity_threshold
+
+CREATE OPERATOR <% (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_op,
+        COMMUTATOR = '%>',
+        RESTRICT = contsel,
+        JOIN = contjoinsel
+);
+
+CREATE OPERATOR %> (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_commutator_op,
+        COMMUTATOR = '<%',
+        RESTRICT = contsel,
+        JOIN = contjoinsel
+);
+
 CREATE FUNCTION similarity_dist(text,text)
 RETURNS float4
 AS 'MODULE_PATHNAME'
@@ -51,6 +84,30 @@ CREATE OPERATOR <-> (
         COMMUTATOR = '<->'
 );
 
+CREATE FUNCTION word_similarity_dist_op(text,text)
+RETURNS float4
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT IMMUTABLE;
+
+CREATE FUNCTION word_similarity_dist_commutator_op(text,text)
+RETURNS float4
+AS 'MODULE_PATHNAME'
+LANGUAGE C STRICT IMMUTABLE;
+
+CREATE OPERATOR <<-> (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_dist_op,
+        COMMUTATOR = '<->>'
+);
+
+CREATE OPERATOR <->> (
+        LEFTARG = text,
+        RIGHTARG = text,
+        PROCEDURE = word_similarity_dist_commutator_op,
+        COMMUTATOR = '<<->'
+);
+
 -- gist key
 CREATE FUNCTION gtrgm_in(cstring)
 RETURNS gtrgm
@@ -140,6 +197,12 @@ ALTER OPERATOR FAMILY gist_trgm_ops USING gist ADD
         OPERATOR        5       pg_catalog.~ (text, text),
         OPERATOR        6       pg_catalog.~* (text, text);
 
+-- Add operators that are new in 9.6 (pg_trgm 1.2).
+
+ALTER OPERATOR FAMILY gist_trgm_ops USING gist ADD
+        OPERATOR        7       %> (text, text),
+        OPERATOR        8       <->> (text, text) FOR ORDER BY pg_catalog.float_ops;
+
 -- support functions for gin
 CREATE FUNCTION gin_extract_value_trgm(text, internal)
 RETURNS internal
@@ -187,4 +250,5 @@ AS 'MODULE_PATHNAME'
 LANGUAGE C IMMUTABLE STRICT;
 
 ALTER OPERATOR FAMILY gin_trgm_ops USING gin ADD
+        OPERATOR        7       %> (text, text),
         FUNCTION        6      (text,text) gin_trgm_triconsistent (internal, int2, text, int4, internal, internal, internal);

contrib/pg_trgm/sql/pg_trgm.sql

@@ -13,7 +13,7 @@ select similarity('wow',' WOW ');
 
 select similarity('---', '####---');
 
-CREATE TABLE test_trgm(t text);
+CREATE TABLE test_trgm(t text COLLATE "C");
 
 \copy test_trgm from 'data/trgm.data'
 
@@ -40,7 +40,7 @@ select t,similarity(t,'qwertyu0988') as sml from test_trgm where t % 'qwertyu098
 select t,similarity(t,'gwertyu0988') as sml from test_trgm where t % 'gwertyu0988' order by sml desc, t;
 select t,similarity(t,'gwertyu1988') as sml from test_trgm where t % 'gwertyu1988' order by sml desc, t;
 
-create table test2(t text);
+create table test2(t text COLLATE "C");
 insert into test2 values ('abcdef');
 insert into test2 values ('quark');
 insert into test2 values ('  z foo bar');

contrib/pg_trgm/trgm.h

@@ -32,7 +32,8 @@
 #define ILikeStrategyNumber				4
 #define RegExpStrategyNumber			5
 #define RegExpICaseStrategyNumber		6
-
+#define WordSimilarityStrategyNumber	7
+#define WordDistanceStrategyNumber		8
 
 typedef char trgm[3];
 
@@ -103,15 +104,28 @@ typedef char *BITVECP;
 #define GETARR(x)		( (trgm*)( (char*)x+TRGMHDRSIZE ) )
 #define ARRNELEM(x) ( ( VARSIZE(x) - TRGMHDRSIZE )/sizeof(trgm) )
 
+/*
+ * If DIVUNION is defined then similarity formula is:
+ * count / (len1 + len2 - count)
+ * else if DIVUNION is not defined then similarity formula is:
+ * count / max(len1, len2)
+ */
+#ifdef DIVUNION
+#define CALCSML(count, len1, len2) ((float4) (count)) / ((float4) ((len1) + (len2) - (count)))
+#else
+#define CALCSML(count, len1, len2) ((float4) (count)) / ((float4) (((len1) > (len2)) ? (len1) : (len2)))
+#endif
+
 typedef struct TrgmPackedGraph TrgmPackedGraph;
 
 extern double similarity_threshold;
+extern double word_similarity_threshold;
 
 extern uint32 trgm2int(trgm *ptr);
 extern void compact_trigram(trgm *tptr, char *str, int bytelen);
 extern TRGM *generate_trgm(char *str, int slen);
 extern TRGM *generate_wildcard_trgm(const char *str, int slen);
-extern float4 cnt_sml(TRGM *trg1, TRGM *trg2);
+extern float4 cnt_sml(TRGM *trg1, TRGM *trg2, bool inexact);
 extern bool trgm_contained_by(TRGM *trg1, TRGM *trg2);
 extern bool *trgm_presence_map(TRGM *query, TRGM *key);
 extern TRGM *createTrgmNFA(text *text_re, Oid collation,

contrib/pg_trgm/trgm_gin.c

@@ -89,6 +89,7 @@ gin_extract_query_trgm(PG_FUNCTION_ARGS)
 	switch (strategy)
 	{
 		case SimilarityStrategyNumber:
+		case WordSimilarityStrategyNumber:
 			trg = generate_trgm(VARDATA(val), VARSIZE(val) - VARHDRSZ);
 			break;
 		case ILikeStrategyNumber:
@@ -176,6 +177,7 @@ gin_trgm_consistent(PG_FUNCTION_ARGS)
 	bool		res;
 	int32		i,
 				ntrue;
+	double		nlimit;
 
 	/* All cases served by this function are inexact */
 	*recheck = true;
@@ -183,6 +185,10 @@ gin_trgm_consistent(PG_FUNCTION_ARGS)
 	switch (strategy)
 	{
 		case SimilarityStrategyNumber:
+		case WordSimilarityStrategyNumber:
+			nlimit = (strategy == SimilarityStrategyNumber) ?
+				similarity_threshold : word_similarity_threshold;
+
 			/* Count the matches */
 			ntrue = 0;
 			for (i = 0; i < nkeys; i++)
@@ -207,8 +213,7 @@ gin_trgm_consistent(PG_FUNCTION_ARGS)
 			 * So, independly on DIVUNION the upper bound formula is the same.
 			 */
 			res = (nkeys == 0) ? false :
-				((((((float4) ntrue) / ((float4) nkeys))) >= similarity_threshold)
-					? true : false);
+				(((((float4) ntrue) / ((float4) nkeys))) >= nlimit);
 			break;
 		case ILikeStrategyNumber:
 #ifndef IGNORECASE
@@ -270,10 +275,15 @@ gin_trgm_triconsistent(PG_FUNCTION_ARGS)
 	int32		i,
 				ntrue;
 	bool	   *boolcheck;
+	double		nlimit;
 
 	switch (strategy)
 	{
 		case SimilarityStrategyNumber:
+		case WordSimilarityStrategyNumber:
+			nlimit = (strategy == SimilarityStrategyNumber) ?
+				similarity_threshold : word_similarity_threshold;
+
 			/* Count the matches */
 			ntrue = 0;
 			for (i = 0; i < nkeys; i++)
@@ -285,8 +295,8 @@ gin_trgm_triconsistent(PG_FUNCTION_ARGS)
 			/*
 			 * See comment in gin_trgm_consistent() about * upper bound formula
 			 */
-			res = (nkeys == 0) ? GIN_FALSE :
-				(((((float4) ntrue) / ((float4) nkeys)) >= similarity_threshold)
+			res = (nkeys == 0)
+				? GIN_FALSE : (((((float4) ntrue) / ((float4) nkeys)) >= nlimit)
 							? GIN_MAYBE : GIN_FALSE);
 			break;
 		case ILikeStrategyNumber:

contrib/pg_trgm/trgm_gist.c

@@ -191,6 +191,7 @@ gtrgm_consistent(PG_FUNCTION_ARGS)
 	bool		res;
 	Size		querysize = VARSIZE(query);
 	gtrgm_consistent_cache *cache;
+	double		nlimit;
 
 	/*
 	 * We keep the extracted trigrams in cache, because trigram extraction is
@@ -218,6 +219,7 @@ gtrgm_consistent(PG_FUNCTION_ARGS)
 		switch (strategy)
 		{
 			case SimilarityStrategyNumber:
+			case WordSimilarityStrategyNumber:
 				qtrg = generate_trgm(VARDATA(query),
 									 querysize - VARHDRSZ);
 				break;
@@ -286,16 +288,23 @@ gtrgm_consistent(PG_FUNCTION_ARGS)
 	switch (strategy)
 	{
 		case SimilarityStrategyNumber:
-			/* Similarity search is exact */
-			*recheck = false;
+		case WordSimilarityStrategyNumber:
+			/* Similarity search is exact. Word similarity search is inexact */
+			*recheck = (strategy == WordSimilarityStrategyNumber);
+			nlimit = (strategy == SimilarityStrategyNumber) ?
+				similarity_threshold : word_similarity_threshold;
 
 			if (GIST_LEAF(entry))
 			{					/* all leafs contains orig trgm */
-				float4		tmpsml = cnt_sml(key, qtrg);
+				/*
+				 * Prevent gcc optimizing the tmpsml variable using volatile
+				 * keyword. Otherwise comparison of nlimit and tmpsml may give
+				 * wrong results.
+				 */
+				float4 volatile tmpsml = cnt_sml(qtrg, key, *recheck);
 
 				/* strange bug at freebsd 5.2.1 and gcc 3.3.3 */
-				res = (*(int *) &tmpsml == *(int *) &similarity_threshold
-						|| tmpsml > similarity_threshold) ? true : false;
+				res = (*(int *) &tmpsml == *(int *) &nlimit || tmpsml > nlimit);
 			}
 			else if (ISALLTRUE(key))
 			{					/* non-leaf contains signature */
@@ -309,8 +318,7 @@ gtrgm_consistent(PG_FUNCTION_ARGS)
 				if (len == 0)
 					res = false;
 				else
-					res = (((((float8) count) / ((float8) len))) >= similarity_threshold)
-							? true : false;
+					res = (((((float8) count) / ((float8) len))) >= nlimit);
 			}
 			break;
 		case ILikeStrategyNumber:
@@ -428,6 +436,7 @@ gtrgm_distance(PG_FUNCTION_ARGS)
 	StrategyNumber strategy = (StrategyNumber) PG_GETARG_UINT16(2);
 
 	/* Oid		subtype = PG_GETARG_OID(3); */
+	bool	   *recheck = (bool *) PG_GETARG_POINTER(4);
 	TRGM	   *key = (TRGM *) DatumGetPointer(entry->key);
 	TRGM	   *qtrg;
 	float8		res;
@@ -463,9 +472,17 @@ gtrgm_distance(PG_FUNCTION_ARGS)
 	switch (strategy)
 	{
 		case DistanceStrategyNumber:
+		case WordDistanceStrategyNumber:
+			*recheck = strategy == WordDistanceStrategyNumber;
 			if (GIST_LEAF(entry))
 			{					/* all leafs contains orig trgm */
-				res = 1.0 - cnt_sml(key, qtrg);
+				/*
+				 * Prevent gcc optimizing the sml variable using volatile
+				 * keyword. Otherwise res can differ from the
+				 * word_similarity_dist_op() function.
+				 */
+				float4 volatile sml = cnt_sml(qtrg, key, *recheck);
+				res = 1.0 - sml;
 			}
 			else if (ISALLTRUE(key))
 			{					/* all leafs contains orig trgm */

doc/src/sgml/pgtrgm.sgml

@@ -92,6 +92,21 @@
        (In practice this is seldom useful except for debugging.)
       </entry>
      </row>
+     <row>
+      <entry>
+       <function>word_similarity(text, text)</function>
+       <indexterm><primary>word_similarity</primary></indexterm>
+      </entry>
+      <entry><type>real</type></entry>
+      <entry>
+       Returns a number that indicates how similar the first string
+       to the most similar word of the second string. The function searches in
+       the second string a most similar word not a most similar substring.  The
+       range of the result is zero (indicating that the two strings are
+       completely dissimilar) to one (indicating that the first string is
+       identical to one of the word of the second string).
+      </entry>
+     </row>
      <row>
       <entry><function>show_limit()</function><indexterm><primary>show_limit</primary></indexterm></entry>
       <entry><type>real</type></entry>
@@ -137,6 +152,16 @@
        <varname>pg_trgm.similarity_threshold</>.
       </entry>
      </row>
+     <row>
+       <entry><type>text</> <literal>%&gt;</literal> <type>text</></entry>
+       <entry><type>boolean</type></entry>
+       <entry>
+        Returns <literal>true</> if its first argument has the similar word in
+        the second argument and they have a similarity that is greater than the
+        current word similarity threshold set by
+        <varname>pg_trgm.word_similarity_threshold</> parameter.
+       </entry>
+      </row>
      <row>
       <entry><type>text</> <literal>&lt;-&gt;</literal> <type>text</></entry>
       <entry><type>real</type></entry>
@@ -145,6 +170,16 @@
        one minus the <function>similarity()</> value.
       </entry>
      </row>
+     <row>
+       <entry>
+        <type>text</> <literal>&lt;-&gt;&gt;</literal> <type>text</>
+       </entry>
+       <entry><type>real</type></entry>
+       <entry>
+        Returns the <quote>distance</> between the arguments, that is
+        one minus the <function>word_similarity()</> value.
+       </entry>
+     </row>
     </tbody>
    </tgroup>
   </table>
@@ -168,6 +203,23 @@
      </para>
     </listitem>
    </varlistentry>
+    <varlistentry id="guc-pgtrgm-word-similarity-threshold" xreflabel="pg_trgm.word_similarity_threshold">
+     <term>
+      <varname>pg_trgm.word_similarity_threshold</> (<type>real</type>)
+      <indexterm>
+       <primary>
+        <varname>pg_trgm.word_similarity_threshold</> configuration parameter
+       </primary>
+      </indexterm>
+     </term>
+     <listitem>
+      <para>
+       Sets the current word similarity threshold that is used by
+       the <literal>%&gt;</> operator.  The threshold must be between
+       0 and 1 (default is 0.6).
+      </para>
+     </listitem>
+    </varlistentry>
   </variablelist>
  </sect2>
 
@@ -225,6 +277,33 @@ SELECT t, t &lt;-&gt; '<replaceable>word</>' AS dist
    a small number of the closest matches is wanted.
   </para>
 
+  <para>
+   Also you can use an index on the <structfield>t</> column for word
+   similarity.  For example:
+<programlisting>
+SELECT t, word_similarity('<replaceable>word</>', t) AS sml
+  FROM test_trgm
+  WHERE t %&gt; '<replaceable>word</>'
+  ORDER BY sml DESC, t;
+</programlisting>
+   This will return all values in the text column that have a word
+   which sufficiently similar to <replaceable>word</>, sorted from best
+   match to worst.  The index will be used to make this a fast operation
+   even over very large data sets.
+  </para>
+
+  <para>
+   A variant of the above query is
+<programlisting>
+SELECT t, t &lt;-&gt;&gt; '<replaceable>word</>' AS dist
+  FROM test_trgm
+  ORDER BY dist LIMIT 10;
+</programlisting>
+   This can be implemented quite efficiently by GiST indexes, but not
+   by GIN indexes.
+  </para>
+
+
   <para>
    Beginning in <productname>PostgreSQL</> 9.1, these index types also support
    index searches for <literal>LIKE</> and <literal>ILIKE</>, for example