Avoid index rebuild for no-rewrite ALTER TABLE .. ALTER TYPE.

Noah Misch.  Review and minor cosmetic changes by me.

doc/src/sgml/xindex.sgml

@@ -834,8 +834,10 @@ ALTER OPERATOR FAMILY integer_ops USING btree ADD
   <para>
    In a B-tree operator family, all the operators in the family must sort
    compatibly, meaning that the transitive laws hold across all the data types
-   supported by the family: <quote>if A = B and B = C, then A =
+   supported by the family: <quote>if A = B and B = C, then A = C</>,
-   C</>, and <quote>if A &lt; B and B &lt; C, then A &lt; C</>.  For each
+   and <quote>if A &lt; B and B &lt; C, then A &lt; C</>.  Moreover, implicit
+   or binary coercion casts between types represented in the operator family
+   must not change the associated sort ordering.  For each
    operator in the family there must be a support function having the same
    two input data types as the operator.  It is recommended that a family be
    complete, i.e., for each combination of data types, all operators are
@@ -851,6 +853,9 @@ ALTER OPERATOR FAMILY integer_ops USING btree ADD
    by the family's equality operators, even when the values are of different
    types.  This is usually difficult to accomplish when the types have
    different physical representations, but it can be done in some cases.
+   Furthermore, casting a value from one data type represented in the operator
+   family to another data type also represented in the operator family via
+   an implicit or binary coercion cast must not change the computed hash value.
    Notice that there is only one support function per data type, not one
    per equality operator.  It is recommended that a family be complete, i.e.,
    provide an equality operator for each combination of data types.

src/backend/bootstrap/bootparse.y

@@ -217,6 +217,7 @@ Boot_CreateStmt:
 												   PG_CATALOG_NAMESPACE,
 												   shared_relation ? GLOBALTABLESPACE_OID : 0,
 												   $3,
+												   InvalidOid,
 												   tupdesc,
 												   RELKIND_RELATION,
 												   RELPERSISTENCE_PERMANENT,
@@ -284,6 +285,7 @@ Boot_DeclareIndexStmt:
 					DefineIndex(makeRangeVar(NULL, $6, -1),
 								$3,
 								$4,
+								InvalidOid,
 								$8,
 								NULL,
 								$10,
@@ -302,6 +304,7 @@ Boot_DeclareUniqueIndexStmt:
 					DefineIndex(makeRangeVar(NULL, $7, -1),
 								$4,
 								$5,
+								InvalidOid,
 								$9,
 								NULL,
 								$11,

src/backend/catalog/heap.c

@@ -229,7 +229,8 @@ SystemAttributeByName(const char *attname, bool relhasoids)
  *		heap_create		- Create an uncataloged heap relation
  *
  *		Note API change: the caller must now always provide the OID
- *		to use for the relation.
+ *		to use for the relation.  The relfilenode may (and, normally,
+ *		should) be left unspecified.
  *
  *		rel->rd_rel is initialized by RelationBuildLocalRelation,
  *		and is mostly zeroes at return.
@@ -240,6 +241,7 @@ heap_create(const char *relname,
 			Oid relnamespace,
 			Oid reltablespace,
 			Oid relid,
+			Oid relfilenode,
 			TupleDesc tupDesc,
 			char relkind,
 			char relpersistence,
@@ -296,6 +298,16 @@ heap_create(const char *relname,
 			break;
 	}
 
+	/*
+	 * Unless otherwise requested, the physical ID (relfilenode) is initially
+	 * the same as the logical ID (OID).  When the caller did specify a
+	 * relfilenode, it already exists; do not attempt to create it.
+	 */
+	if (OidIsValid(relfilenode))
+		create_storage = false;
+	else
+		relfilenode = relid;
+
 	/*
 	 * Never allow a pg_class entry to explicitly specify the database's
 	 * default tablespace in reltablespace; force it to zero instead. This
@@ -315,6 +327,7 @@ heap_create(const char *relname,
 									 relnamespace,
 									 tupDesc,
 									 relid,
+									 relfilenode,
 									 reltablespace,
 									 shared_relation,
 									 mapped_relation,
@@ -1103,6 +1116,7 @@ heap_create_with_catalog(const char *relname,
 							   relnamespace,
 							   reltablespace,
 							   relid,
+							   InvalidOid,
 							   tupdesc,
 							   relkind,
 							   relpersistence,

src/backend/catalog/index.c

@@ -649,6 +649,8 @@ UpdateIndexRelation(Oid indexoid,
  * indexRelationId: normally, pass InvalidOid to let this routine
  *		generate an OID for the index.	During bootstrap this may be
  *		nonzero to specify a preselected OID.
+ * relFileNode: normally, pass InvalidOid to get new storage.  May be
+ *		nonzero to attach an existing valid build.
  * indexInfo: same info executor uses to insert into the index
  * indexColNames: column names to use for index (List of char *)
  * accessMethodObjectId: OID of index AM to use
@@ -674,6 +676,7 @@ Oid
 index_create(Relation heapRelation,
 			 const char *indexRelationName,
 			 Oid indexRelationId,
+			 Oid relFileNode,
 			 IndexInfo *indexInfo,
 			 List *indexColNames,
 			 Oid accessMethodObjectId,
@@ -813,6 +816,7 @@ index_create(Relation heapRelation,
 								namespaceId,
 								tableSpaceId,
 								indexRelationId,
+								relFileNode,
 								indexTupDesc,
 								RELKIND_INDEX,
 								relpersistence,

src/backend/catalog/storage.c

@@ -206,10 +206,13 @@ RelationDropStorage(Relation rel)
  * The relation mapper fixes this by telling us to not delete such relations
  * after all as part of its commit.
  *
+ * We also use this to reuse an old build of an index during ALTER TABLE, this
+ * time removing the delete-at-commit entry.
+ *
  * No-op if the relation is not among those scheduled for deletion.
  */
 void
-RelationPreserveStorage(RelFileNode rnode)
+RelationPreserveStorage(RelFileNode rnode, bool atCommit)
 {
 	PendingRelDelete *pending;
 	PendingRelDelete *prev;
@@ -219,11 +222,9 @@ RelationPreserveStorage(RelFileNode rnode)
 	for (pending = pendingDeletes; pending != NULL; pending = next)
 	{
 		next = pending->next;
-		if (RelFileNodeEquals(rnode, pending->relnode))
+		if (RelFileNodeEquals(rnode, pending->relnode)
+			&& pending->atCommit == atCommit)
 		{
-			/* we should only find delete-on-abort entries, else trouble */
-			if (pending->atCommit)
-				elog(ERROR, "cannot preserve a delete-on-commit relation");
 			/* unlink and delete list entry */
 			if (prev)
 				prev->next = next;

src/backend/catalog/toasting.c

@@ -274,7 +274,7 @@ create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid, Datum reloptio
 	coloptions[0] = 0;
 	coloptions[1] = 0;
 
-	index_create(toast_rel, toast_idxname, toastIndexOid,
+	index_create(toast_rel, toast_idxname, toastIndexOid, InvalidOid,
 				 indexInfo,
 				 list_make2("chunk_id", "chunk_seq"),
 				 BTREE_AM_OID,

src/backend/commands/indexcmds.c

@@ -72,6 +72,198 @@ static Oid GetIndexOpClass(List *opclass, Oid attrType,
 static char *ChooseIndexNameAddition(List *colnames);
 
 
+/*
+ * CheckIndexCompatible
+ *		Determine whether an existing index definition is compatible with a
+ *		prospective index definition, such that the existing index storage
+ *		could become the storage of the new index, avoiding a rebuild.
+ *
+ * 'heapRelation': the relation the index would apply to.
+ * 'accessMethodName': name of the AM to use.
+ * 'attributeList': a list of IndexElem specifying columns and expressions
+ *		to index on.
+ * 'exclusionOpNames': list of names of exclusion-constraint operators,
+ *		or NIL if not an exclusion constraint.
+ *
+ * This is tailored to the needs of ALTER TABLE ALTER TYPE, which recreates
+ * any indexes that depended on a changing column from their pg_get_indexdef
+ * or pg_get_constraintdef definitions.  We omit some of the sanity checks of
+ * DefineIndex.  We assume that the old and new indexes have the same number
+ * of columns and that if one has an expression column or predicate, both do.
+ * Errors arising from the attribute list still apply.
+ *
+ * Most column type changes that can skip a table rewrite will not invalidate
+ * indexes.  For btree and hash indexes, we assume continued validity when
+ * each column of an index would have the same operator family before and
+ * after the change.  Since we do not document a contract for GIN or GiST
+ * operator families, we require an exact operator class match for them and
+ * for any other access methods.
+ *
+ * DefineIndex always verifies that each exclusion operator shares an operator
+ * family with its corresponding index operator class.  For access methods
+ * having no operator family contract, confirm that the old and new indexes
+ * use the exact same exclusion operator.  For btree and hash, there's nothing
+ * more to check.
+ *
+ * We do not yet implement a test to verify compatibility of expression
+ * columns or predicates, so assume any such index is incompatible.
+ */
+bool
+CheckIndexCompatible(Oid oldId,
+					 RangeVar *heapRelation,
+					 char *accessMethodName,
+					 List *attributeList,
+					 List *exclusionOpNames)
+{
+	bool		isconstraint;
+	Oid		   *collationObjectId;
+	Oid		   *classObjectId;
+	Oid			accessMethodId;
+	Oid			relationId;
+	HeapTuple	tuple;
+	Form_pg_am	accessMethodForm;
+	bool		amcanorder;
+	RegProcedure amoptions;
+	int16	   *coloptions;
+	IndexInfo  *indexInfo;
+	int			numberOfAttributes;
+	int			old_natts;
+	bool		isnull;
+	bool		family_am;
+	bool		ret = true;
+	oidvector  *old_indclass;
+	oidvector  *old_indcollation;
+	int			i;
+	Datum		d;
+
+	/* Caller should already have the relation locked in some way. */
+	relationId = RangeVarGetRelid(heapRelation, NoLock, false, false);
+	/*
+	 * We can pretend isconstraint = false unconditionally.  It only serves to
+	 * decide the text of an error message that should never happen for us.
+	 */
+	isconstraint = false;
+
+	numberOfAttributes = list_length(attributeList);
+	Assert(numberOfAttributes > 0);
+	Assert(numberOfAttributes <= INDEX_MAX_KEYS);
+
+	/* look up the access method */
+	tuple = SearchSysCache1(AMNAME, PointerGetDatum(accessMethodName));
+	if (!HeapTupleIsValid(tuple))
+		ereport(ERROR,
+				(errcode(ERRCODE_UNDEFINED_OBJECT),
+				 errmsg("access method \"%s\" does not exist",
+						accessMethodName)));
+	accessMethodId = HeapTupleGetOid(tuple);
+	accessMethodForm = (Form_pg_am) GETSTRUCT(tuple);
+	amcanorder = accessMethodForm->amcanorder;
+	amoptions = accessMethodForm->amoptions;
+	ReleaseSysCache(tuple);
+
+	/*
+	 * Compute the operator classes, collations, and exclusion operators
+	 * for the new index, so we can test whether it's compatible with the
+	 * existing one.  Note that ComputeIndexAttrs might fail here, but that's
+	 * OK: DefineIndex would have called this function with the same arguments
+	 * later on, and it would have failed then anyway.
+	 */
+	indexInfo = makeNode(IndexInfo);
+	indexInfo->ii_Expressions = NIL;
+	indexInfo->ii_ExpressionsState = NIL;
+	indexInfo->ii_PredicateState = NIL;
+	indexInfo->ii_ExclusionOps = NULL;
+	indexInfo->ii_ExclusionProcs = NULL;
+	indexInfo->ii_ExclusionStrats = NULL;
+	collationObjectId = (Oid *) palloc(numberOfAttributes * sizeof(Oid));
+	classObjectId = (Oid *) palloc(numberOfAttributes * sizeof(Oid));
+	coloptions = (int16 *) palloc(numberOfAttributes * sizeof(int16));
+	ComputeIndexAttrs(indexInfo, collationObjectId, classObjectId,
+					  coloptions, attributeList,
+					  exclusionOpNames, relationId,
+					  accessMethodName, accessMethodId,
+					  amcanorder, isconstraint);
+
+
+	/* Get the soon-obsolete pg_index tuple. */
+	tuple = SearchSysCache1(INDEXRELID, ObjectIdGetDatum(oldId));
+	if (!HeapTupleIsValid(tuple))
+		elog(ERROR, "cache lookup failed for index %u", oldId);
+
+	/* We don't assess expressions or predicates; assume incompatibility. */
+	if (!(heap_attisnull(tuple, Anum_pg_index_indpred) &&
+		  heap_attisnull(tuple, Anum_pg_index_indexprs)))
+	{
+		ReleaseSysCache(tuple);
+		return false;
+	}
+
+	/*
+	 * If the old and new operator class of any index column differ in
+	 * operator family or collation, regard the old index as incompatible.
+	 * For access methods other than btree and hash, a family match has no
+	 * defined meaning; require an exact operator class match.
+	 */
+	old_natts = ((Form_pg_index) GETSTRUCT(tuple))->indnatts;
+	Assert(old_natts == numberOfAttributes);
+
+	d = SysCacheGetAttr(INDEXRELID, tuple, Anum_pg_index_indcollation, &isnull);
+	Assert(!isnull);
+	old_indcollation = (oidvector *) DatumGetPointer(d);
+
+	d = SysCacheGetAttr(INDEXRELID, tuple, Anum_pg_index_indclass, &isnull);
+	Assert(!isnull);
+	old_indclass = (oidvector *) DatumGetPointer(d);
+
+	family_am = accessMethodId == BTREE_AM_OID || accessMethodId == HASH_AM_OID;
+
+	for (i = 0; i < old_natts; i++)
+	{
+		Oid			old_class = old_indclass->values[i];
+		Oid			new_class = classObjectId[i];
+
+		if (!(old_indcollation->values[i] == collationObjectId[i]
+			  && (old_class == new_class
+				  || (family_am && (get_opclass_family(old_class)
+									== get_opclass_family(new_class))))))
+		{
+			ret = false;
+			break;
+		}
+	}
+
+	ReleaseSysCache(tuple);
+
+	/*
+	 * For btree and hash, exclusion operators need only fall in the same
+	 * operator family; ComputeIndexAttrs already verified that much.  If we
+	 * get this far, we know that the index operator family has not changed,
+	 * and we're done.  For other access methods, require exact matches for
+	 * all exclusion operators.
+	 */
+	if (ret && !family_am && indexInfo->ii_ExclusionOps != NULL)
+	{
+		Relation	irel;
+		Oid		   *old_operators, *old_procs;
+		uint16	   *old_strats;
+
+		/* Caller probably already holds a stronger lock. */
+		irel = index_open(oldId, AccessShareLock);
+		RelationGetExclusionInfo(irel, &old_operators, &old_procs, &old_strats);
+
+		for (i = 0; i < old_natts; i++)
+			if (old_operators[i] != indexInfo->ii_ExclusionOps[i])
+			{
+				ret = false;
+				break;
+			}
+
+		index_close(irel, NoLock);
+	}
+
+	return ret;
+}
+
 /*
  * DefineIndex
  *		Creates a new index.
@@ -81,6 +273,8 @@ static char *ChooseIndexNameAddition(List *colnames);
  *		that a nonconflicting default name should be picked.
  * 'indexRelationId': normally InvalidOid, but during bootstrap can be
  *		nonzero to specify a preselected OID for the index.
+ * 'relFileNode': normally InvalidOid, but can be nonzero to specify existing
+ *		storage constituting a valid build of this index.
  * 'accessMethodName': name of the AM to use.
  * 'tableSpaceName': name of the tablespace to create the index in.
  *		NULL specifies using the appropriate default.
@@ -103,11 +297,14 @@ static char *ChooseIndexNameAddition(List *colnames);
  *		it will be filled later.
  * 'quiet': suppress the NOTICE chatter ordinarily provided for constraints.
  * 'concurrent': avoid blocking writers to the table while building.
+ *
+ * Returns the OID of the created index.
  */
-void
+Oid
 DefineIndex(RangeVar *heapRelation,
 			char *indexRelationName,
 			Oid indexRelationId,
+			Oid relFileNode,
 			char *accessMethodName,
 			char *tableSpaceName,
 			List *attributeList,
@@ -402,12 +599,18 @@ DefineIndex(RangeVar *heapRelation,
 				  indexRelationName, RelationGetRelationName(rel))));
 	}
 
+	/*
+	 * A valid relFileNode implies that we already have a built form of the
+	 * index.  The caller should also decline any index build.
+	 */
+	Assert(!OidIsValid(relFileNode) || (skip_build && !concurrent));
+
 	/*
 	 * Make the catalog entries for the index, including constraints. Then, if
 	 * not skip_build || concurrent, actually build the index.
 	 */
 	indexRelationId =
-		index_create(rel, indexRelationName, indexRelationId,
+		index_create(rel, indexRelationName, indexRelationId, relFileNode,
 					 indexInfo, indexColNames,
 					 accessMethodId, tablespaceId,
 					 collationObjectId, classObjectId,
@@ -421,7 +624,7 @@ DefineIndex(RangeVar *heapRelation,
 	{
 		/* Close the heap and we're done, in the non-concurrent case */
 		heap_close(rel, NoLock);
-		return;
+		return indexRelationId;
 	}
 
 	/* save lockrelid and locktag for below, then close rel */
@@ -709,6 +912,8 @@ DefineIndex(RangeVar *heapRelation,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
+
+	return indexRelationId;
 }
 
 

src/backend/commands/tablecmds.c

@@ -347,7 +347,9 @@ static bool ATColumnChangeRequiresRewrite(Node *expr, AttrNumber varattno);
 static void ATExecAlterColumnType(AlteredTableInfo *tab, Relation rel,
 					  AlterTableCmd *cmd, LOCKMODE lockmode);
 static void ATPostAlterTypeCleanup(List **wqueue, AlteredTableInfo *tab, LOCKMODE lockmode);
-static void ATPostAlterTypeParse(char *cmd, List **wqueue, LOCKMODE lockmode);
+static void ATPostAlterTypeParse(Oid oldId, char *cmd,
+					 List **wqueue, LOCKMODE lockmode, bool rewrite);
+static void TryReuseIndex(Oid oldId, IndexStmt *stmt);
 static void change_owner_recurse_to_sequences(Oid relationOid,
 								  Oid newOwnerId, LOCKMODE lockmode);
 static void ATExecClusterOn(Relation rel, const char *indexName, LOCKMODE lockmode);
@@ -5232,21 +5234,23 @@ ATExecAddIndex(AlteredTableInfo *tab, Relation rel,
 	bool		check_rights;
 	bool		skip_build;
 	bool		quiet;
+	Oid			new_index;
 
 	Assert(IsA(stmt, IndexStmt));
 
 	/* suppress schema rights check when rebuilding existing index */
 	check_rights = !is_rebuild;
-	/* skip index build if phase 3 will have to rewrite table anyway */
+	/* skip index build if phase 3 will do it or we're reusing an old one */
-	skip_build = tab->rewrite;
+	skip_build = tab->rewrite || OidIsValid(stmt->oldNode);
 	/* suppress notices when rebuilding existing index */
 	quiet = is_rebuild;
 
 	/* The IndexStmt has already been through transformIndexStmt */
 
-	DefineIndex(stmt->relation, /* relation */
+	new_index = DefineIndex(stmt->relation,		/* relation */
 							stmt->idxname,		/* index name */
 							InvalidOid, /* no predefined OID */
+							stmt->oldNode,
 							stmt->accessMethod, /* am name */
 							stmt->tableSpace,
 							stmt->indexParams,	/* parameters */
@@ -5263,6 +5267,19 @@ ATExecAddIndex(AlteredTableInfo *tab, Relation rel,
 							skip_build,
 							quiet,
 							false);
+
+	/*
+	 * If TryReuseIndex() stashed a relfilenode for us, we used it for the new
+	 * index instead of building from scratch.  The DROP of the old edition of
+	 * this index will have scheduled the storage for deletion at commit, so
+	 * cancel that pending deletion.
+	 */
+	if (OidIsValid(stmt->oldNode))
+	{
+		Relation	irel = index_open(new_index, NoLock);
+		RelationPreserveStorage(irel->rd_node, true);
+		index_close(irel, NoLock);
+	}
 }
 
 /*
@@ -7389,7 +7406,8 @@ static void
 ATPostAlterTypeCleanup(List **wqueue, AlteredTableInfo *tab, LOCKMODE lockmode)
 {
 	ObjectAddress obj;
-	ListCell   *l;
+	ListCell   *def_item;
+	ListCell   *oid_item;
 
 	/*
 	 * Re-parse the index and constraint definitions, and attach them to the
@@ -7399,10 +7417,14 @@ ATPostAlterTypeCleanup(List **wqueue, AlteredTableInfo *tab, LOCKMODE lockmode)
 	 * that before dropping.  It's safe because the parser won't actually look
 	 * at the catalogs to detect the existing entry.
 	 */
-	foreach(l, tab->changedIndexDefs)
+	forboth(oid_item, tab->changedConstraintOids,
-		ATPostAlterTypeParse((char *) lfirst(l), wqueue, lockmode);
+			def_item, tab->changedConstraintDefs)
-	foreach(l, tab->changedConstraintDefs)
+		ATPostAlterTypeParse(lfirst_oid(oid_item), (char *) lfirst(def_item),
-		ATPostAlterTypeParse((char *) lfirst(l), wqueue, lockmode);
+							 wqueue, lockmode, tab->rewrite);
+	forboth(oid_item, tab->changedIndexOids,
+			def_item, tab->changedIndexDefs)
+		ATPostAlterTypeParse(lfirst_oid(oid_item), (char *) lfirst(def_item),
+							 wqueue, lockmode, tab->rewrite);
 
 	/*
 	 * Now we can drop the existing constraints and indexes --- constraints
@@ -7412,18 +7434,18 @@ ATPostAlterTypeCleanup(List **wqueue, AlteredTableInfo *tab, LOCKMODE lockmode)
 	 * should be okay to use DROP_RESTRICT here, since nothing else should be
 	 * depending on these objects.
 	 */
-	foreach(l, tab->changedConstraintOids)
+	foreach(oid_item, tab->changedConstraintOids)
 	{
 		obj.classId = ConstraintRelationId;
-		obj.objectId = lfirst_oid(l);
+		obj.objectId = lfirst_oid(oid_item);
 		obj.objectSubId = 0;
 		performDeletion(&obj, DROP_RESTRICT);
 	}
 
-	foreach(l, tab->changedIndexOids)
+	foreach(oid_item, tab->changedIndexOids)
 	{
 		obj.classId = RelationRelationId;
-		obj.objectId = lfirst_oid(l);
+		obj.objectId = lfirst_oid(oid_item);
 		obj.objectSubId = 0;
 		performDeletion(&obj, DROP_RESTRICT);
 	}
@@ -7435,7 +7457,8 @@ ATPostAlterTypeCleanup(List **wqueue, AlteredTableInfo *tab, LOCKMODE lockmode)
 }
 
 static void
-ATPostAlterTypeParse(char *cmd, List **wqueue, LOCKMODE lockmode)
+ATPostAlterTypeParse(Oid oldId, char *cmd,
+					 List **wqueue, LOCKMODE lockmode, bool rewrite)
 {
 	List	   *raw_parsetree_list;
 	List	   *querytree_list;
@@ -7482,6 +7505,9 @@ ATPostAlterTypeParse(char *cmd, List **wqueue, LOCKMODE lockmode)
 					IndexStmt  *stmt = (IndexStmt *) stm;
 					AlterTableCmd *newcmd;
 
+					if (!rewrite)
+						TryReuseIndex(oldId, stmt);
+
 					rel = relation_openrv(stmt->relation, lockmode);
 					tab = ATGetQueueEntry(wqueue, rel);
 					newcmd = makeNode(AlterTableCmd);
@@ -7506,6 +7532,10 @@ ATPostAlterTypeParse(char *cmd, List **wqueue, LOCKMODE lockmode)
 						switch (cmd->subtype)
 						{
 							case AT_AddIndex:
+								Assert(IsA(cmd->def, IndexStmt));
+								if (!rewrite)
+									TryReuseIndex(get_constraint_index(oldId),
+												  (IndexStmt *) cmd->def);
 								cmd->subtype = AT_ReAddIndex;
 								tab->subcmds[AT_PASS_OLD_INDEX] =
 									lappend(tab->subcmds[AT_PASS_OLD_INDEX], cmd);
@@ -7529,6 +7559,26 @@ ATPostAlterTypeParse(char *cmd, List **wqueue, LOCKMODE lockmode)
 	}
 }
 
+/*
+ * Subroutine for ATPostAlterTypeParse().  Calls out to CheckIndexCompatible()
+ * for the real analysis, then mutates the IndexStmt based on that verdict.
+*/
+static void
+TryReuseIndex(Oid oldId, IndexStmt *stmt)
+{
+
+	if (CheckIndexCompatible(oldId,
+							 stmt->relation,
+							 stmt->accessMethod,
+							 stmt->indexParams,
+							 stmt->excludeOpNames))
+	{
+		Relation irel = index_open(oldId, NoLock);
+		stmt->oldNode = irel->rd_node.relNode;
+		index_close(irel, NoLock);
+	}
+}
+
 
 /*
  * ALTER TABLE OWNER

src/backend/nodes/copyfuncs.c

@@ -2807,6 +2807,7 @@ _copyIndexStmt(IndexStmt *from)
 	COPY_NODE_FIELD(whereClause);
 	COPY_NODE_FIELD(excludeOpNames);
 	COPY_SCALAR_FIELD(indexOid);
+	COPY_SCALAR_FIELD(oldNode);
 	COPY_SCALAR_FIELD(unique);
 	COPY_SCALAR_FIELD(primary);
 	COPY_SCALAR_FIELD(isconstraint);

src/backend/nodes/equalfuncs.c

@@ -1245,6 +1245,7 @@ _equalIndexStmt(IndexStmt *a, IndexStmt *b)
 	COMPARE_NODE_FIELD(whereClause);
 	COMPARE_NODE_FIELD(excludeOpNames);
 	COMPARE_SCALAR_FIELD(indexOid);
+	COMPARE_SCALAR_FIELD(oldNode);
 	COMPARE_SCALAR_FIELD(unique);
 	COMPARE_SCALAR_FIELD(primary);
 	COMPARE_SCALAR_FIELD(isconstraint);

src/backend/nodes/outfuncs.c

@@ -1976,6 +1976,7 @@ _outIndexStmt(StringInfo str, IndexStmt *node)
 	WRITE_NODE_FIELD(whereClause);
 	WRITE_NODE_FIELD(excludeOpNames);
 	WRITE_OID_FIELD(indexOid);
+	WRITE_OID_FIELD(oldNode);
 	WRITE_BOOL_FIELD(unique);
 	WRITE_BOOL_FIELD(primary);
 	WRITE_BOOL_FIELD(isconstraint);

src/backend/tcop/utility.c

@@ -953,6 +953,7 @@ standard_ProcessUtility(Node *parsetree,
 				DefineIndex(stmt->relation,		/* relation */
 							stmt->idxname,		/* index name */
 							InvalidOid, /* no predefined OID */
+							InvalidOid, /* no previous storage */
 							stmt->accessMethod, /* am name */
 							stmt->tableSpace,
 							stmt->indexParams,	/* parameters */

src/backend/utils/cache/relcache.c

@@ -2395,6 +2395,7 @@ RelationBuildLocalRelation(const char *relname,
 						   Oid relnamespace,
 						   TupleDesc tupDesc,
 						   Oid relid,
+						   Oid relfilenode,
 						   Oid reltablespace,
 						   bool shared_relation,
 						   bool mapped_relation,
@@ -2529,10 +2530,8 @@ RelationBuildLocalRelation(const char *relname,
 
 	/*
 	 * Insert relation physical and logical identifiers (OIDs) into the right
-	 * places.	Note that the physical ID (relfilenode) is initially the same
+	 * places.  For a mapped relation, we set relfilenode to zero and rely on
-	 * as the logical ID (OID); except that for a mapped relation, we set
+	 * RelationInitPhysicalAddr to consult the map.
-	 * relfilenode to zero and rely on RelationInitPhysicalAddr to consult the
-	 * map.
 	 */
 	rel->rd_rel->relisshared = shared_relation;
 
@@ -2547,10 +2546,10 @@ RelationBuildLocalRelation(const char *relname,
 	{
 		rel->rd_rel->relfilenode = InvalidOid;
 		/* Add it to the active mapping information */
-		RelationMapUpdateMap(relid, relid, shared_relation, true);
+		RelationMapUpdateMap(relid, relfilenode, shared_relation, true);
 	}
 	else
-		rel->rd_rel->relfilenode = relid;
+		rel->rd_rel->relfilenode = relfilenode;
 
 	RelationInitLockInfo(rel);	/* see lmgr.c */
 

src/backend/utils/cache/relmapper.c

@@ -792,7 +792,7 @@ write_relmap_file(bool shared, RelMapFile *newmap,
 			rnode.spcNode = tsid;
 			rnode.dbNode = dbid;
 			rnode.relNode = newmap->mappings[i].mapfilenode;
-			RelationPreserveStorage(rnode);
+			RelationPreserveStorage(rnode, false);
 		}
 	}
 

src/include/catalog/heap.h

@@ -39,6 +39,7 @@ extern Relation heap_create(const char *relname,
 			Oid relnamespace,
 			Oid reltablespace,
 			Oid relid,
+			Oid relfilenode,
 			TupleDesc tupDesc,
 			char relkind,
 			char relpersistence,

src/include/catalog/index.h

@@ -35,6 +35,7 @@ extern void index_check_primary_key(Relation heapRel,
 extern Oid index_create(Relation heapRelation,
 			 const char *indexRelationName,
 			 Oid indexRelationId,
+			 Oid relFileNode,
 			 IndexInfo *indexInfo,
 			 List *indexColNames,
 			 Oid accessMethodObjectId,

src/include/catalog/storage.h

@@ -22,7 +22,7 @@
 
 extern void RelationCreateStorage(RelFileNode rnode, char relpersistence);
 extern void RelationDropStorage(Relation rel);
-extern void RelationPreserveStorage(RelFileNode rnode);
+extern void RelationPreserveStorage(RelFileNode rnode, bool atCommit);
 extern void RelationTruncate(Relation rel, BlockNumber nblocks);
 
 /*

src/include/commands/defrem.h

@@ -18,9 +18,10 @@
 
 
 /* commands/indexcmds.c */
-extern void DefineIndex(RangeVar *heapRelation,
+extern Oid DefineIndex(RangeVar *heapRelation,
 			char *indexRelationName,
 			Oid indexRelationId,
+			Oid relFileNode,
 			char *accessMethodName,
 			char *tableSpaceName,
 			List *attributeList,
@@ -49,6 +50,11 @@ extern char *ChooseIndexName(const char *tabname, Oid namespaceId,
 				List *colnames, List *exclusionOpNames,
 				bool primary, bool isconstraint);
 extern List *ChooseIndexColumnNames(List *indexElems);
+extern bool CheckIndexCompatible(Oid oldId,
+					 RangeVar *heapRelation,
+					 char *accessMethodName,
+					 List *attributeList,
+					 List *exclusionOpNames);
 extern Oid	GetDefaultOpClass(Oid type_id, Oid am_id);
 
 /* commands/functioncmds.c */

src/include/nodes/parsenodes.h

@@ -2062,6 +2062,7 @@ typedef struct IndexStmt
 	Node	   *whereClause;	/* qualification (partial-index predicate) */
 	List	   *excludeOpNames; /* exclusion operator names, or NIL if none */
 	Oid			indexOid;		/* OID of an existing index, if any */
+	Oid			oldNode;		/* relfilenode of my former self */
 	bool		unique;			/* is index unique? */
 	bool		primary;		/* is index on primary key? */
 	bool		isconstraint;	/* is it from a CONSTRAINT clause? */

src/include/utils/relcache.h

@@ -67,6 +67,7 @@ extern Relation RelationBuildLocalRelation(const char *relname,
 						   Oid relnamespace,
 						   TupleDesc tupDesc,
 						   Oid relid,
+						   Oid relfilenode,
 						   Oid reltablespace,
 						   bool shared_relation,
 						   bool mapped_relation,