Only C, POSIX and ICU collations now allowed

- ALTER COLLATION ... REFRESH VERSION forbidden when there are OrioleDB indexes using collation - Added collation check for ALTER COLUMN TYPE
JoshInnis · Sep 9, 2023 · fcd00b5 · fcd00b5
1 parent 9402e42
commit fcd00b5
Show file tree

Hide file tree

Showing 16 changed files with 424 additions and 80 deletions.
diff --git a/.pgtags b/.pgtags
@@ -1,4 +1,4 @@
-13: patches13_14
-14: patches14_14
-15: patches15_15
-16: patches16_15
+13: patches13_15
+14: patches14_15
+15: patches15_17
+16: patches16_17
diff --git a/Makefile b/Makefile
@@ -180,14 +180,18 @@ ifeq ($(shell expr $(MAJORVERSION) \>= 15), 1)
   ISOLATIONCHECKS += isol_merge
 endif
 
+PG_REGRESS_ARGS=--no-locale --encoding=UTF8
+
 regresscheck: | install
 	$(pg_regress_check) \
 		--temp-config orioledb_regression.conf \
+		$(PG_REGRESS_ARGS) \
 		$(REGRESSCHECKS)
 
 isolationcheck: | install
 	$(pg_isolation_regress_check) \
 		--temp-config orioledb_isolation.conf \
+		$(PG_REGRESS_ARGS) \
 		$(ISOLATIONCHECKS)
 
 $(TESTGRESCHECKS_PART_1) $(TESTGRESCHECKS_PART_2): | install

diff --git a/README.md b/README.md
@@ -83,7 +83,7 @@ See [our dockerhub](https://hub.docker.com/r/orioledb/orioledb) for details.
 
 Before building and installing OrioleDB, one should ensure to have the following:
 
- * [PostgreSQL with extensibility patches](https://github.com/orioledb/postgres): [13 (tag: patches13_14)](https://github.com/orioledb/postgres/tree/patches13_14), [14 (tag: patches14_14)](https://github.com/orioledb/postgres/tree/patches14_14), [15 (tag: patches15_15)](https://github.com/orioledb/postgres/tree/patches15_15), or [16 (tag: patches16_15)](https://github.com/orioledb/postgres/tree/patches16_15);
+ * [PostgreSQL with extensibility patches](https://github.com/orioledb/postgres): [13 (tag: patches13_15)](https://github.com/orioledb/postgres/tree/patches13_15), [14 (tag: patches14_15)](https://github.com/orioledb/postgres/tree/patches14_15), [15 (tag: patches15_17)](https://github.com/orioledb/postgres/tree/patches15_17), or [16 (tag: patches16_17)](https://github.com/orioledb/postgres/tree/patches16_17);
  * Development package of libzstd;
  * python 3.5+ with testgres package.
 

diff --git a/doc/usage.md b/doc/usage.md
@@ -5,6 +5,12 @@ OrioleDB is implemented using the table access method interface.  Therefore, to
 
 OrioleDB uses index-organized tables.  So, the selection of the primary key is a very critical decision affecting performance.  If you specify no primary key, then a hidden surrogate primary key will be created over the virtual `ctid` column.
 
+Collations
+----------
+OrioleDB tables support only ICU, C, and POSIX collations. So, make sure the cluster or database is set up with default collations that fall under those options, otherwise you have to write COLLATE for every "text" field of the table.
+
+ALTER COLLATION REFRESH VERSION is also disabled for collations that used for fields and indexes of orioledb tables.
+
 Example
 -------
 

diff --git a/include/catalog/o_sys_cache.h b/include/catalog/o_sys_cache.h
@@ -590,6 +590,7 @@ extern void o_amproc_cache_tup_print(BTreeDescr *desc, StringInfo buf,
 /* o_collation_cache.c */
 O_SYS_CACHE_DECLS(collation_cache, OCollation, 1);
 extern HeapTuple o_collation_cache_search_htup(TupleDesc tupdesc, Oid colloid);
+extern void orioledb_save_collation(Oid colloid);
 
 /* o_database_cache.c */
 typedef struct ODatabase

diff --git a/include/catalog/o_tables.h b/include/catalog/o_tables.h
@@ -230,7 +230,6 @@ extern void o_table_fill_constr(OTable *o_table, Relation rel, int fieldnum,
 extern void o_tupdesc_load_constr(TupleDesc tupdesc, OTable *o_table,
 								  OIndexDescr *descr);
 extern char *o_get_type_name(Oid typid, int32 typmod);
-extern char *o_get_collation_name(Oid colid);
 
 static inline int
 o_table_fieldnum(OTable *table, const char *name)

diff --git a/src/catalog/ddl.c b/src/catalog/ddl.c
@@ -118,6 +118,7 @@ static void orioledb_ExecutorRun_hook(QueryDesc *queryDesc,
 									  bool execute_once);
 static void o_alter_column_type(AlterTableCmd *cmd, const char *queryString,
 								Relation rel);
+static void o_find_collation_dependencies(Oid colloid);
 
 void
 orioledb_setup_ddl_hooks(void)
@@ -1009,6 +1010,13 @@ orioledb_utility_command(PlannedStmt *pstmt,
 			}
 		}
 	}
+	else if (IsA(pstmt->utilityStmt, AlterCollationStmt))
+	{
+		AlterCollationStmt *astmt = (AlterCollationStmt *) pstmt->utilityStmt;
+		Oid			collOid = get_collation_oid(astmt->collname, false);
+
+		o_find_collation_dependencies(collOid);
+	}
 
 	if (call_next)
 	{
@@ -1055,6 +1063,93 @@ o_alter_column_type(AlterTableCmd *cmd, const char *queryString, Relation rel)
 	}
 }
 
+static void
+o_find_collation_dependencies(Oid colloid)
+{
+	Relation	depRel;
+	ScanKeyData key[2];
+	SysScanDesc depScan;
+	HeapTuple	depTup;
+	HeapTuple	collationtup;
+	Form_pg_collation collform;
+
+	/* since this function recurses, it could be driven to stack overflow */
+	check_stack_depth();
+
+	/*
+	 * We scan pg_depend to find those things that depend on the given type.
+	 * (We assume we can ignore refobjsubid for a type.)
+	 */
+	depRel = table_open(DependRelationId, AccessShareLock);
+
+	ScanKeyInit(&key[0],
+				Anum_pg_depend_refclassid,
+				BTEqualStrategyNumber, F_OIDEQ,
+				ObjectIdGetDatum(CollationRelationId));
+	ScanKeyInit(&key[1],
+				Anum_pg_depend_refobjid,
+				BTEqualStrategyNumber, F_OIDEQ,
+				ObjectIdGetDatum(colloid));
+
+	depScan = systable_beginscan(depRel, DependReferenceIndexId, true,
+								 NULL, 2, key);
+
+	collationtup = SearchSysCache1(COLLOID, colloid);
+	if (!HeapTupleIsValid(collationtup))
+		elog(ERROR, "cache lookup failed for collation (%u)", colloid);
+	collform = (Form_pg_collation) GETSTRUCT(collationtup);
+
+	while (HeapTupleIsValid(depTup = systable_getnext(depScan)))
+	{
+		Form_pg_depend pg_depend = (Form_pg_depend) GETSTRUCT(depTup);
+		Relation	rel;
+
+		/* Else, ignore dependees that aren't user columns of relations */
+		/* (we assume system columns are never of interesting types) */
+		if (pg_depend->classid != RelationRelationId)
+			continue;
+
+		rel = relation_open(pg_depend->objid, AccessShareLock);
+
+		if ((rel->rd_rel->relkind == RELKIND_RELATION ||
+			 rel->rd_rel->relkind == RELKIND_MATVIEW) &&
+			is_orioledb_rel(rel))
+		{
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot refresh collation \"%s\" because "
+							"orioledb table \"%s\" uses it",
+							collform->collname.data,
+							RelationGetRelationName(rel))));
+		}
+		else if (rel->rd_rel->relkind == RELKIND_INDEX)
+		{
+			Relation	tbl;
+
+			tbl = relation_open(rel->rd_index->indrelid, AccessShareLock);
+
+			if ((tbl->rd_rel->relkind == RELKIND_RELATION ||
+				 tbl->rd_rel->relkind == RELKIND_MATVIEW) &&
+				is_orioledb_rel(tbl))
+			{
+				ereport(ERROR,
+						(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+						 errmsg("cannot refresh collation \"%s\" because "
+								"orioledb index \"%s\" uses it",
+								collform->collname.data,
+								RelationGetRelationName(rel))));
+			}
+			relation_close(tbl, AccessShareLock);
+		}
+
+		relation_close(rel, AccessShareLock);
+	}
+	ReleaseSysCache(collationtup);
+	systable_endscan(depScan);
+
+	relation_close(depRel, AccessShareLock);
+}
+
 static void
 o_find_composite_type_dependencies(Oid typeOid, Relation origRelation)
 {
@@ -1776,6 +1871,7 @@ orioledb_object_access_hook(ObjectAccessType access, Oid classId, Oid objectId,
 
 					if (!rewrite)
 					{
+						orioledb_save_collation(field->collation);
 						fill_current_oxid_csn(&oxid, &csn);
 						o_tables_meta_lock();
 						o_tables_update(o_table, oxid, csn);

diff --git a/src/catalog/o_collation_cache.c b/src/catalog/o_collation_cache.c
@@ -85,6 +85,7 @@ o_collation_cache_fill_entry(Pointer *entry_ptr, OSysCacheKey *key,
 	Oid			colloid;
 	Datum		datum;
 	bool		isNull;
+	bool		valid;
 
 	colloid = DatumGetObjectId(key->keys[0]);
 
@@ -93,6 +94,17 @@ o_collation_cache_fill_entry(Pointer *entry_ptr, OSysCacheKey *key,
 		elog(ERROR, "cache lookup failed for collation (%u)", colloid);
 	collform = (Form_pg_collation) GETSTRUCT(collationtup);
 
+	valid = collform->collprovider == COLLPROVIDER_ICU ||
+		lc_collate_is_c(colloid);
+
+#if PG_VERSION_NUM >= 150000
+	valid = valid || (colloid == DEFAULT_COLLATION_OID &&
+					  default_locale.provider == COLLPROVIDER_ICU);
+#endif
+	if (!valid)
+		elog(ERROR,
+			 "Only C, POSIX and ICU collations supported for orioledb tables");
+
 	prev_context = MemoryContextSwitchTo(collation_cache->mcxt);
 	if (o_collation != NULL)	/* Existed o_collation updated */
 	{
@@ -282,3 +294,19 @@ o_collation_cache_search_htup(TupleDesc tupdesc, Oid colloid)
 	}
 	return result;
 }
+
+void
+orioledb_save_collation(Oid colloid)
+{
+	if (OidIsValid(colloid))
+	{
+		XLogRecPtr	cur_lsn;
+		Oid			datoid;
+		OClassArg	arg = {.sys_table = true};
+
+		o_sys_cache_set_datoid_lsn(&cur_lsn, &datoid);
+		o_class_cache_add_if_needed(datoid, CollationRelationId, cur_lsn,
+									(Pointer) &arg);
+		o_collation_cache_add_if_needed(datoid, colloid, cur_lsn, NULL);
+	}
+}
diff --git a/src/catalog/o_indices.c b/src/catalog/o_indices.c
@@ -34,6 +34,7 @@
 #include "miscadmin.h"
 #include "optimizer/optimizer.h"
 #include "utils/builtins.h"
+#include "utils/lsyscache.h"
 
 PG_FUNCTION_INFO_V1(orioledb_index_oids);
 PG_FUNCTION_INFO_V1(orioledb_index_description);
@@ -1114,7 +1115,7 @@ describe_index(TupleDesc tupdesc, ORelOids oids, OIndexType type)
 	{
 		OTableField *field = &index->leafFields[i];
 		char	   *typename = o_get_type_name(field->typid, field->typmod);
-		char	   *colname = o_get_collation_name(field->collation);
+		char	   *colname = get_collation_name(field->collation);
 
 		if (max_column_str < strlen(NameStr(field->name)))
 			max_column_str = strlen(NameStr(field->name));
@@ -1144,7 +1145,7 @@ describe_index(TupleDesc tupdesc, ORelOids oids, OIndexType type)
 	{
 		OTableField *field = &index->leafFields[i];
 		char	   *typename = o_get_type_name(field->typid, field->typmod);
-		char	   *colname = o_get_collation_name(field->collation);
+		char	   *colname = get_collation_name(field->collation);
 
 		appendStringInfo(&buf, format.data,
 						 NameStr(field->name),

diff --git a/src/catalog/o_tables.c b/src/catalog/o_tables.c
@@ -417,18 +417,8 @@ o_table_fill_index(OTable *o_table, OIndexNumber ix_num, Relation index_rel)
 			exprField->align = typeTup->typalign;
 			exprField->typmod = exprTypmod(indexkey);
 			exprField->collation = exprCollation(indexkey);
-			if (OidIsValid(exprField->collation))
-			{
-				XLogRecPtr	cur_lsn;
-				Oid			datoid;
-				OClassArg	arg = {.sys_table = true};
-
-				o_sys_cache_set_datoid_lsn(&cur_lsn, &datoid);
-				o_class_cache_add_if_needed(datoid, CollationRelationId,
-											cur_lsn, (Pointer) &arg);
-				o_collation_cache_add_if_needed(datoid, exprField->collation,
-												cur_lsn, NULL);
-			}
+
+			orioledb_save_collation(exprField->collation);
 
 			ReleaseSysCache(tuple);
 
@@ -511,18 +501,7 @@ o_table_fill_index(OTable *o_table, OIndexNumber ix_num, Relation index_rel)
 				ix_field->nullsOrdering = SORTBY_NULLS_FIRST;
 			}
 		}
-		if (OidIsValid(ix_field->collation))
-		{
-			XLogRecPtr	cur_lsn;
-			Oid			datoid;
-			OClassArg	arg = {.sys_table = true};
-
-			o_sys_cache_set_datoid_lsn(&cur_lsn, &datoid);
-			o_class_cache_add_if_needed(datoid, CollationRelationId, cur_lsn,
-										(Pointer) &arg);
-			o_collation_cache_add_if_needed(datoid, ix_field->collation,
-											cur_lsn, NULL);
-		}
+		orioledb_save_collation(ix_field->collation);
 	}
 }
 
@@ -619,6 +598,27 @@ o_table_fill_constr(OTable *o_table, Relation rel, int fieldnum,
 	MemoryContextSwitchTo(oldcxt);
 }
 
+void
+orioledb_attr_to_field(OTableField *field, Form_pg_attribute attr)
+{
+	strlcpy(NameStr(field->name), NameStr(attr->attname), NAMEDATALEN);
+	field->typid = attr->atttypid;
+	field->collation = attr->attcollation;
+	field->typmod = attr->atttypmod;
+	field->typlen = attr->attlen;
+	field->ndims = attr->attndims;
+	field->byval = attr->attbyval;
+	field->align = attr->attalign;
+	field->storage = attr->attstorage;
+#if PG_VERSION_NUM >= 140000
+	field->compression = attr->attcompression;
+#endif
+	field->droped = attr->attisdropped;
+	field->notnull = attr->attnotnull;
+	field->hasmissing = attr->atthasmissing;
+	field->hasdef = attr->atthasdef;
+}
+
 OTable *
 o_table_tableam_create(ORelOids oids, TupleDesc tupdesc)
 {
@@ -641,18 +641,7 @@ o_table_tableam_create(ORelOids oids, TupleDesc tupdesc)
 		OTableField *field = &o_table->fields[i];
 
 		orioledb_attr_to_field(field, attr);
-		if (OidIsValid(field->collation))
-		{
-			XLogRecPtr	cur_lsn;
-			Oid			datoid;
-			OClassArg	arg = {.sys_table = true};
-
-			o_sys_cache_set_datoid_lsn(&cur_lsn, &datoid);
-			o_class_cache_add_if_needed(datoid, CollationRelationId, cur_lsn,
-										(Pointer) &arg);
-			o_collation_cache_add_if_needed(datoid, field->collation, cur_lsn,
-											NULL);
-		}
+		orioledb_save_collation(field->collation);
 	}
 	o_table->nindices = 0;
 	o_table_resize_constr(o_table);
@@ -1249,15 +1238,6 @@ o_get_type_name(Oid typid, int32 typmod)
 								FORMAT_TYPE_ALLOW_INVALID);
 }
 
-char *
-o_get_collation_name(Oid colid)
-{
-	if (colid < FirstGenbkiObjectId)
-		return get_collation_name(colid);
-	else
-		return psprintf("%u", colid);
-}
-
 static text *
 describe_table(ORelOids oids)
 {
@@ -1284,7 +1264,7 @@ describe_table(ORelOids oids)
 	{
 		OTableField *field = &table->fields[i];
 		char	   *typename = o_get_type_name(field->typid, field->typmod);
-		char	   *colname = o_get_collation_name(field->collation);
+		char	   *colname = get_collation_name(field->collation);
 
 		if (max_column_str < strlen(NameStr(field->name)))
 			max_column_str = strlen(NameStr(field->name));
@@ -1323,7 +1303,7 @@ describe_table(ORelOids oids)
 	{
 		OTableField *field = &table->fields[i];
 		char	   *typename = o_get_type_name(field->typid, field->typmod);
-		char	   *colname = o_get_collation_name(field->collation);
+		char	   *colname = get_collation_name(field->collation);
 
 		appendStringInfo(&buf, format.data,
 						 NameStr(field->name),