v1-0001-Use-correct-collation-for-lowercasing.patch

From dee7d76a76608b51bad4d98399e4d6a7484a2714 Mon Sep 17 00:00:00 2001
From: David Geier <geidav.pg@gmail.com>
Date: Wed, 21 Jan 2026 14:54:28 +0100
Subject: [PATCH v1 1/3] Use correct collation for lowercasing

pg_trgm converts the input words to lowercase before extracting the
trigrams. The lowercase conversion depends on the collation. Previously,
pg_trgm always used the default collation. Now, the specified collation
is used instead.
---
 contrib/pg_trgm/Makefile                      |  2 +-
 .../pg_trgm/expected/pg_trgm_collation.out    | 43 +++++++++++++
 .../pg_trgm/expected/pg_trgm_collation_1.out  |  9 +++
 contrib/pg_trgm/meson.build                   |  1 +
 contrib/pg_trgm/sql/pg_trgm_collation.sql     | 24 ++++++++
 contrib/pg_trgm/trgm.h                        |  4 +-
 contrib/pg_trgm/trgm_gin.c                    |  7 ++-
 contrib/pg_trgm/trgm_gist.c                   | 10 ++--
 contrib/pg_trgm/trgm_op.c                     | 60 ++++++++++---------
 contrib/pg_trgm/trgm_regexp.c                 | 20 +++----
 10 files changed, 132 insertions(+), 48 deletions(-)
 create mode 100644 contrib/pg_trgm/expected/pg_trgm_collation.out
 create mode 100644 contrib/pg_trgm/expected/pg_trgm_collation_1.out
 create mode 100644 contrib/pg_trgm/sql/pg_trgm_collation.sql

diff --git a/contrib/pg_trgm/Makefile b/contrib/pg_trgm/Makefile
index c1756993ec7..26b3028b75e 100644
--- a/contrib/pg_trgm/Makefile
+++ b/contrib/pg_trgm/Makefile
@@ -14,7 +14,7 @@ DATA = pg_trgm--1.5--1.6.sql pg_trgm--1.4--1.5.sql pg_trgm--1.3--1.4.sql \
 	pg_trgm--1.0--1.1.sql
 PGFILEDESC = "pg_trgm - trigram matching"
 
-REGRESS = pg_trgm pg_utf8_trgm pg_word_trgm pg_strict_word_trgm
+REGRESS = pg_trgm pg_utf8_trgm pg_word_trgm pg_strict_word_trgm pg_trgm_collation
 
 ifdef USE_PGXS
 PG_CONFIG = pg_config
diff --git a/contrib/pg_trgm/expected/pg_trgm_collation.out b/contrib/pg_trgm/expected/pg_trgm_collation.out
new file mode 100644
index 00000000000..472ce867665
--- /dev/null
+++ b/contrib/pg_trgm/expected/pg_trgm_collation.out
@@ -0,0 +1,43 @@
+/*
+ * This test is for ICU collations.
+ */
+/* skip test if not UTF8 server encoding or no ICU collations installed */
+SELECT getdatabaseencoding() <> 'UTF8' OR
+       (SELECT count(*) FROM pg_collation WHERE collprovider = 'i' AND collname <> 'unicode') = 0
+       AS skip_test \gset
+\if :skip_test
+\quit
+\endif
+-- Test that lowercase conversion of trigrams uses specified collation
+CREATE TABLE test(col TEXT COLLATE "tr-x-icu");
+INSERT INTO test VALUES ('ISTANBUL');
+SELECT show_trgm(col) FROM test;
+                       show_trgm                        
+--------------------------------------------------------
+ {0xf31e1a,0xfe581d,0x3efd30,anb,bul,nbu,sta,tan,"ul "}
+(1 row)
+
+SELECT show_trgm('ISTANBUL' COLLATE "tr-x-icu");
+                       show_trgm                        
+--------------------------------------------------------
+ {0xf31e1a,0xfe581d,0x3efd30,anb,bul,nbu,sta,tan,"ul "}
+(1 row)
+
+SELECT show_trgm('ISTANBUL' COLLATE "C");
+                  show_trgm                  
+---------------------------------------------
+ {"  i"," is",anb,bul,ist,nbu,sta,tan,"ul "}
+(1 row)
+
+SELECT similarity('ıstanbul' COLLATE "tr-x-icu", 'ISTANBUL' COLLATE "tr-x-icu");
+ similarity 
+------------
+          1
+(1 row)
+
+SELECT similarity('ıstanbul' COLLATE "C", 'ISTANBUL' COLLATE "C");
+ similarity 
+------------
+ 0.54545456
+(1 row)
+
diff --git a/contrib/pg_trgm/expected/pg_trgm_collation_1.out b/contrib/pg_trgm/expected/pg_trgm_collation_1.out
new file mode 100644
index 00000000000..25c99c4abf0
--- /dev/null
+++ b/contrib/pg_trgm/expected/pg_trgm_collation_1.out
@@ -0,0 +1,9 @@
+/*
+ * This test is for ICU collations.
+ */
+/* skip test if not UTF8 server encoding or no ICU collations installed */
+SELECT getdatabaseencoding() <> 'UTF8' OR
+       (SELECT count(*) FROM pg_collation WHERE collprovider = 'i' AND collname <> 'unicode') = 0
+       AS skip_test \gset
+\if :skip_test
+\quit
diff --git a/contrib/pg_trgm/meson.build b/contrib/pg_trgm/meson.build
index 3ecf95ba862..5eafa774435 100644
--- a/contrib/pg_trgm/meson.build
+++ b/contrib/pg_trgm/meson.build
@@ -42,6 +42,7 @@ tests += {
       'pg_utf8_trgm',
       'pg_word_trgm',
       'pg_strict_word_trgm',
+      'pg_trgm_collation',
     ],
   },
 }
diff --git a/contrib/pg_trgm/sql/pg_trgm_collation.sql b/contrib/pg_trgm/sql/pg_trgm_collation.sql
new file mode 100644
index 00000000000..afb3973a8b3
--- /dev/null
+++ b/contrib/pg_trgm/sql/pg_trgm_collation.sql
@@ -0,0 +1,24 @@
+/*
+ * This test is for ICU collations.
+ */
+
+/* skip test if not UTF8 server encoding or no ICU collations installed */
+SELECT getdatabaseencoding() <> 'UTF8' OR
+       (SELECT count(*) FROM pg_collation WHERE collprovider = 'i' AND collname <> 'unicode') = 0
+       AS skip_test \gset
+\if :skip_test
+\quit
+\endif
+
+-- Test that lowercase conversion of trigrams uses specified collation
+
+CREATE TABLE test(col TEXT COLLATE "tr-x-icu");
+INSERT INTO test VALUES ('ISTANBUL');
+SELECT show_trgm(col) FROM test;
+SELECT show_trgm('ISTANBUL' COLLATE "tr-x-icu");
+
+SELECT show_trgm('ISTANBUL' COLLATE "C");
+
+SELECT similarity('ıstanbul' COLLATE "tr-x-icu", 'ISTANBUL' COLLATE "tr-x-icu");
+SELECT similarity('ıstanbul' COLLATE "C", 'ISTANBUL' COLLATE "C");
+
diff --git a/contrib/pg_trgm/trgm.h b/contrib/pg_trgm/trgm.h
index ca23aad4dd9..b6911e91458 100644
--- a/contrib/pg_trgm/trgm.h
+++ b/contrib/pg_trgm/trgm.h
@@ -119,8 +119,8 @@ extern double strict_word_similarity_threshold;
 extern double index_strategy_get_limit(StrategyNumber strategy);
 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 TRGM *generate_trgm(char *str, int slen, Oid collation);
+extern TRGM *generate_wildcard_trgm(const char *str, int slen, Oid collation);
 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);
diff --git a/contrib/pg_trgm/trgm_gin.c b/contrib/pg_trgm/trgm_gin.c
index 5766b3e9955..14a892c657d 100644
--- a/contrib/pg_trgm/trgm_gin.c
+++ b/contrib/pg_trgm/trgm_gin.c
@@ -42,7 +42,7 @@ gin_extract_value_trgm(PG_FUNCTION_ARGS)
 
 	*nentries = 0;
 
-	trg = generate_trgm(VARDATA_ANY(val), VARSIZE_ANY_EXHDR(val));
+	trg = generate_trgm(VARDATA_ANY(val), VARSIZE_ANY_EXHDR(val), PG_GET_COLLATION());
 	trglen = ARRNELEM(trg);
 
 	if (trglen > 0)
@@ -93,7 +93,7 @@ gin_extract_query_trgm(PG_FUNCTION_ARGS)
 		case WordSimilarityStrategyNumber:
 		case StrictWordSimilarityStrategyNumber:
 		case EqualStrategyNumber:
-			trg = generate_trgm(VARDATA_ANY(val), VARSIZE_ANY_EXHDR(val));
+			trg = generate_trgm(VARDATA_ANY(val), VARSIZE_ANY_EXHDR(val), PG_GET_COLLATION());
 			break;
 		case ILikeStrategyNumber:
 #ifndef IGNORECASE
@@ -107,7 +107,8 @@ gin_extract_query_trgm(PG_FUNCTION_ARGS)
 			 * potentially-matching string must include.
 			 */
 			trg = generate_wildcard_trgm(VARDATA_ANY(val),
-										 VARSIZE_ANY_EXHDR(val));
+										 VARSIZE_ANY_EXHDR(val),
+										 PG_GET_COLLATION());
 			break;
 		case RegExpICaseStrategyNumber:
 #ifndef IGNORECASE
diff --git a/contrib/pg_trgm/trgm_gist.c b/contrib/pg_trgm/trgm_gist.c
index 11812b2984e..d9102400442 100644
--- a/contrib/pg_trgm/trgm_gist.c
+++ b/contrib/pg_trgm/trgm_gist.c
@@ -123,7 +123,7 @@ gtrgm_compress(PG_FUNCTION_ARGS)
 		TRGM	   *res;
 		text	   *val = DatumGetTextPP(entry->key);
 
-		res = generate_trgm(VARDATA_ANY(val), VARSIZE_ANY_EXHDR(val));
+		res = generate_trgm(VARDATA_ANY(val), VARSIZE_ANY_EXHDR(val), PG_GET_COLLATION());
 		retval = palloc_object(GISTENTRY);
 		gistentryinit(*retval, PointerGetDatum(res),
 					  entry->rel, entry->page,
@@ -242,7 +242,8 @@ gtrgm_consistent(PG_FUNCTION_ARGS)
 			case StrictWordSimilarityStrategyNumber:
 			case EqualStrategyNumber:
 				qtrg = generate_trgm(VARDATA(query),
-									 querysize - VARHDRSZ);
+									 querysize - VARHDRSZ,
+									 PG_GET_COLLATION());
 				break;
 			case ILikeStrategyNumber:
 #ifndef IGNORECASE
@@ -251,7 +252,8 @@ gtrgm_consistent(PG_FUNCTION_ARGS)
 				pg_fallthrough;
 			case LikeStrategyNumber:
 				qtrg = generate_wildcard_trgm(VARDATA(query),
-											  querysize - VARHDRSZ);
+											  querysize - VARHDRSZ,
+											  PG_GET_COLLATION());
 				break;
 			case RegExpICaseStrategyNumber:
 #ifndef IGNORECASE
@@ -475,7 +477,7 @@ gtrgm_distance(PG_FUNCTION_ARGS)
 	{
 		char	   *newcache;
 
-		qtrg = generate_trgm(VARDATA(query), querysize - VARHDRSZ);
+		qtrg = generate_trgm(VARDATA(query), querysize - VARHDRSZ, PG_GET_COLLATION());
 
 		newcache = MemoryContextAlloc(fcinfo->flinfo->fn_mcxt,
 									  MAXALIGN(querysize) +
diff --git a/contrib/pg_trgm/trgm_op.c b/contrib/pg_trgm/trgm_op.c
index 22bcc3c3361..c04d7265153 100644
--- a/contrib/pg_trgm/trgm_op.c
+++ b/contrib/pg_trgm/trgm_op.c
@@ -490,7 +490,7 @@ done:
  * bounds_p: where to return bounds of trigrams (if needed).
  */
 static void
-generate_trgm_only(growable_trgm_array *dst, char *str, int slen, TrgmBound **bounds_p)
+generate_trgm_only(growable_trgm_array *dst, char *str, int slen, Oid collation, TrgmBound **bounds_p)
 {
 	size_t		buflen;
 	char	   *buf;
@@ -542,7 +542,7 @@ generate_trgm_only(growable_trgm_array *dst, char *str, int slen, TrgmBound **bo
 		{
 			char	   *lowered;
 
-			lowered = str_tolower(bword, eword - bword, DEFAULT_COLLATION_OID);
+			lowered = str_tolower(bword, eword - bword, collation);
 			bytelen = strlen(lowered);
 
 			/* grow the buffer if necessary */
@@ -596,13 +596,13 @@ generate_trgm_only(growable_trgm_array *dst, char *str, int slen, TrgmBound **bo
  * Returns the sorted array of unique trigrams.
  */
 TRGM *
-generate_trgm(char *str, int slen)
+generate_trgm(char *str, int slen, Oid collation)
 {
 	TRGM	   *trg;
 	growable_trgm_array arr;
 	int			len;
 
-	generate_trgm_only(&arr, str, slen, NULL);
+	generate_trgm_only(&arr, str, slen, collation, NULL);
 	len = arr.length;
 	trg = arr.datum;
 	trg->flag = ARRKEY;
@@ -857,7 +857,7 @@ iterate_word_similarity(int *trg2indexes,
  */
 static float4
 calc_word_similarity(char *str1, int slen1, char *str2, int slen2,
-					 uint8 flags)
+					 uint8 flags, Oid collation)
 {
 	bool	   *found;
 	pos_trgm   *ptrg;
@@ -875,9 +875,9 @@ calc_word_similarity(char *str1, int slen1, char *str2, int slen2,
 
 	/* Make positional trigrams */
 
-	generate_trgm_only(&trg1, str1, slen1, NULL);
+	generate_trgm_only(&trg1, str1, slen1, collation, NULL);
 	len1 = trg1.length;
-	generate_trgm_only(&trg2, str2, slen2, (flags & WORD_SIMILARITY_STRICT) ? &bounds : NULL);
+	generate_trgm_only(&trg2, str2, slen2, collation, (flags & WORD_SIMILARITY_STRICT) ? &bounds : NULL);
 	len2 = trg2.length;
 
 	ptrg = make_positional_trgm(GETARR(trg1.datum), len1, GETARR(trg2.datum), len2);
@@ -1086,7 +1086,7 @@ get_wildcard_part(const char *str, int lenstr,
  * " a", "bcd" would be extracted.
  */
 TRGM *
-generate_wildcard_trgm(const char *str, int slen)
+generate_wildcard_trgm(const char *str, int slen, Oid collation)
 {
 	TRGM	   *trg;
 	growable_trgm_array arr;
@@ -1118,7 +1118,7 @@ generate_wildcard_trgm(const char *str, int slen)
 		char	   *word;
 
 #ifdef IGNORECASE
-		word = str_tolower(buf, bytelen, DEFAULT_COLLATION_OID);
+		word = str_tolower(buf, bytelen, collation);
 		bytelen = strlen(word);
 #else
 		word = buf;
@@ -1177,7 +1177,7 @@ show_trgm(PG_FUNCTION_ARGS)
 	trgm	   *ptr;
 	int			i;
 
-	trg = generate_trgm(VARDATA_ANY(in), VARSIZE_ANY_EXHDR(in));
+	trg = generate_trgm(VARDATA_ANY(in), VARSIZE_ANY_EXHDR(in), PG_GET_COLLATION());
 	d = palloc_array(Datum, 1 + ARRNELEM(trg));
 
 	for (i = 0, ptr = GETARR(trg); i < ARRNELEM(trg); i++, ptr++)
@@ -1344,8 +1344,8 @@ similarity(PG_FUNCTION_ARGS)
 			   *trg2;
 	float4		res;
 
-	trg1 = generate_trgm(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1));
-	trg2 = generate_trgm(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2));
+	trg1 = generate_trgm(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1), PG_GET_COLLATION());
+	trg2 = generate_trgm(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2), PG_GET_COLLATION());
 
 	res = cnt_sml(trg1, trg2, false);
 
@@ -1366,7 +1366,7 @@ word_similarity(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
 							   VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
-							   0);
+							   0, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1382,7 +1382,7 @@ strict_word_similarity(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
 							   VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
-							   WORD_SIMILARITY_STRICT);
+							   WORD_SIMILARITY_STRICT, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1392,9 +1392,10 @@ strict_word_similarity(PG_FUNCTION_ARGS)
 Datum
 similarity_dist(PG_FUNCTION_ARGS)
 {
-	float4		res = DatumGetFloat4(DirectFunctionCall2(similarity,
-														 PG_GETARG_DATUM(0),
-														 PG_GETARG_DATUM(1)));
+	float4		res = DatumGetFloat4(DirectFunctionCall2Coll(similarity,
+															 PG_GET_COLLATION(),
+															 PG_GETARG_DATUM(0),
+															 PG_GETARG_DATUM(1)));
 
 	PG_RETURN_FLOAT4(1.0 - res);
 }
@@ -1402,9 +1403,10 @@ similarity_dist(PG_FUNCTION_ARGS)
 Datum
 similarity_op(PG_FUNCTION_ARGS)
 {
-	float4		res = DatumGetFloat4(DirectFunctionCall2(similarity,
-														 PG_GETARG_DATUM(0),
-														 PG_GETARG_DATUM(1)));
+	float4		res = DatumGetFloat4(DirectFunctionCall2Coll(similarity,
+															 PG_GET_COLLATION(),
+															 PG_GETARG_DATUM(0),
+															 PG_GETARG_DATUM(1)));
 
 	PG_RETURN_BOOL(res >= similarity_threshold);
 }
@@ -1418,7 +1420,7 @@ word_similarity_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
 							   VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
-							   WORD_SIMILARITY_CHECK_ONLY);
+							   WORD_SIMILARITY_CHECK_ONLY, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1434,7 +1436,7 @@ word_similarity_commutator_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
 							   VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
-							   WORD_SIMILARITY_CHECK_ONLY);
+							   WORD_SIMILARITY_CHECK_ONLY, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1450,7 +1452,7 @@ word_similarity_dist_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
 							   VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
-							   0);
+							   0, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1466,7 +1468,7 @@ word_similarity_dist_commutator_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
 							   VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
-							   0);
+							   0, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1482,7 +1484,8 @@ strict_word_similarity_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
 							   VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
-							   WORD_SIMILARITY_CHECK_ONLY | WORD_SIMILARITY_STRICT);
+							   WORD_SIMILARITY_CHECK_ONLY | WORD_SIMILARITY_STRICT,
+							   PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1498,7 +1501,8 @@ strict_word_similarity_commutator_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
 							   VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
-							   WORD_SIMILARITY_CHECK_ONLY | WORD_SIMILARITY_STRICT);
+							   WORD_SIMILARITY_CHECK_ONLY | WORD_SIMILARITY_STRICT,
+							   PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1514,7 +1518,7 @@ strict_word_similarity_dist_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
 							   VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
-							   WORD_SIMILARITY_STRICT);
+							   WORD_SIMILARITY_STRICT, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
@@ -1530,7 +1534,7 @@ strict_word_similarity_dist_commutator_op(PG_FUNCTION_ARGS)
 
 	res = calc_word_similarity(VARDATA_ANY(in2), VARSIZE_ANY_EXHDR(in2),
 							   VARDATA_ANY(in1), VARSIZE_ANY_EXHDR(in1),
-							   WORD_SIMILARITY_STRICT);
+							   WORD_SIMILARITY_STRICT, PG_GET_COLLATION());
 
 	PG_FREE_IF_COPY(in1, 0);
 	PG_FREE_IF_COPY(in2, 1);
diff --git a/contrib/pg_trgm/trgm_regexp.c b/contrib/pg_trgm/trgm_regexp.c
index b4eaeec6090..b4086a75e17 100644
--- a/contrib/pg_trgm/trgm_regexp.c
+++ b/contrib/pg_trgm/trgm_regexp.c
@@ -479,11 +479,11 @@ typedef struct
 
 /* prototypes for private functions */
 static TRGM *createTrgmNFAInternal(regex_t *regex, TrgmPackedGraph **graph,
-								   MemoryContext rcontext);
+								   MemoryContext rcontext, Oid collation);
 static void RE_compile(regex_t *regex, text *text_re,
 					   int cflags, Oid collation);
-static void getColorInfo(regex_t *regex, TrgmNFA *trgmNFA);
-static int	convertPgWchar(pg_wchar c, trgm_mb_char *result);
+static void getColorInfo(regex_t *regex, TrgmNFA *trgmNFA, Oid collation);
+static int	convertPgWchar(pg_wchar c, trgm_mb_char *result, Oid collation);
 static void transformGraph(TrgmNFA *trgmNFA);
 static void processState(TrgmNFA *trgmNFA, TrgmState *state);
 static void addKey(TrgmNFA *trgmNFA, TrgmState *state, TrgmStateKey *key);
@@ -551,7 +551,7 @@ createTrgmNFA(text *text_re, Oid collation,
 			   REG_ADVANCED | REG_NOSUB, collation);
 #endif
 
-	trg = createTrgmNFAInternal(&regex, graph, rcontext);
+	trg = createTrgmNFAInternal(&regex, graph, rcontext, collation);
 
 	/* Clean up all the cruft we created (including regex) */
 	MemoryContextSwitchTo(oldcontext);
@@ -565,7 +565,7 @@ createTrgmNFA(text *text_re, Oid collation,
  */
 static TRGM *
 createTrgmNFAInternal(regex_t *regex, TrgmPackedGraph **graph,
-					  MemoryContext rcontext)
+					  MemoryContext rcontext, Oid collation)
 {
 	TRGM	   *trg;
 	TrgmNFA		trgmNFA;
@@ -573,7 +573,7 @@ createTrgmNFAInternal(regex_t *regex, TrgmPackedGraph **graph,
 	trgmNFA.regex = regex;
 
 	/* Collect color information from the regex */
-	getColorInfo(regex, &trgmNFA);
+	getColorInfo(regex, &trgmNFA, collation);
 
 #ifdef TRGM_REGEXP_DEBUG
 	printSourceNFA(regex, trgmNFA.colorInfo, trgmNFA.ncolors);
@@ -762,7 +762,7 @@ RE_compile(regex_t *regex, text *text_re, int cflags, Oid collation)
  * Fill TrgmColorInfo structure for each color using regex export functions.
  */
 static void
-getColorInfo(regex_t *regex, TrgmNFA *trgmNFA)
+getColorInfo(regex_t *regex, TrgmNFA *trgmNFA, Oid collation)
 {
 	int			colorsCount = pg_reg_getnumcolors(regex);
 	int			i;
@@ -807,7 +807,7 @@ getColorInfo(regex_t *regex, TrgmNFA *trgmNFA)
 		for (j = 0; j < charsCount; j++)
 		{
 			trgm_mb_char c;
-			int			clen = convertPgWchar(chars[j], &c);
+			int			clen = convertPgWchar(chars[j], &c, collation);
 
 			if (!clen)
 				continue;		/* ok to ignore it altogether */
@@ -827,7 +827,7 @@ getColorInfo(regex_t *regex, TrgmNFA *trgmNFA)
  * byte length.
  */
 static int
-convertPgWchar(pg_wchar c, trgm_mb_char *result)
+convertPgWchar(pg_wchar c, trgm_mb_char *result, Oid collation)
 {
 	/* "s" has enough space for a multibyte character and a trailing NUL */
 	char		s[MAX_MULTIBYTE_CHAR_LEN + 1];
@@ -860,7 +860,7 @@ convertPgWchar(pg_wchar c, trgm_mb_char *result)
 	 */
 #ifdef IGNORECASE
 	{
-		char	   *lowerCased = str_tolower(s, clen, DEFAULT_COLLATION_OID);
+		char	   *lowerCased = str_tolower(s, clen, collation);
 
 		if (strcmp(lowerCased, s) != 0)
 		{
-- 
2.51.0