Correctly copy the target host identification in PQcancelCreate.

PQcancelCreate failed to copy struct pg_conn_host's "type" field,
instead leaving it zero (a/k/a CHT_HOST_NAME).  This seemingly
has no great ill effects if it should have been CHT_UNIX_SOCKET
instead, but if it should have been CHT_HOST_ADDRESS then a
null-pointer dereference will occur when the cancelConn is used.

Bug: #18974
Reported-by: Maxim Boguk <maxim.boguk@gmail.com>
Author: Sergei Kornilov <sk@zsrv.org>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/18974-575f02b2168b36b3@postgresql.org
Backpatch-through: 17

doc/src/sgml/pgbuffercache.sgml

@@ -37,12 +37,12 @@
 
  <para>
   This module provides the <function>pg_buffercache_pages()</function>
-  function (wrapped in the <structname>pg_buffercache</structname> view),
+  function (wrapped in the <structname>pg_buffercache</structname> view), the
   <function>pg_buffercache_numa_pages()</function> function (wrapped in the
   <structname>pg_buffercache_numa</structname> view), the
   <function>pg_buffercache_summary()</function> function, the
   <function>pg_buffercache_usage_counts()</function> function, the
-  <function>pg_buffercache_evict()</function>, the
+  <function>pg_buffercache_evict()</function> function, the
   <function>pg_buffercache_evict_relation()</function> function and the
   <function>pg_buffercache_evict_all()</function> function.
  </para>
@@ -55,7 +55,7 @@
  </para>
 
  <para>
-  The <function>pg_buffercache_numa_pages()</function> provides
+  The <function>pg_buffercache_numa_pages()</function> function provides
   <acronym>NUMA</acronym> node mappings for shared buffer entries. This
   information is not part of <function>pg_buffercache_pages()</function>
   itself, as it is much slower to retrieve.

src/backend/access/heap/vacuumlazy.c

@@ -431,7 +431,7 @@ static void find_next_unskippable_block(LVRelState *vacrel, bool *skipsallvis);
 static bool lazy_scan_new_or_empty(LVRelState *vacrel, Buffer buf,
 								   BlockNumber blkno, Page page,
 								   bool sharelock, Buffer vmbuffer);
-static void lazy_scan_prune(LVRelState *vacrel, Buffer buf,
+static int	lazy_scan_prune(LVRelState *vacrel, Buffer buf,
 							BlockNumber blkno, Page page,
 							Buffer vmbuffer, bool all_visible_according_to_vm,
 							bool *has_lpdead_items, bool *vm_page_frozen);
@@ -1245,6 +1245,7 @@ lazy_scan_heap(LVRelState *vacrel)
 		Buffer		buf;
 		Page		page;
 		uint8		blk_info = 0;
+		int			ndeleted = 0;
 		bool		has_lpdead_items;
 		void	   *per_buffer_data = NULL;
 		bool		vm_page_frozen = false;
@@ -1387,10 +1388,10 @@ lazy_scan_heap(LVRelState *vacrel)
 		 * line pointers previously marked LP_DEAD.
 		 */
 		if (got_cleanup_lock)
-			lazy_scan_prune(vacrel, buf, blkno, page,
+			ndeleted = lazy_scan_prune(vacrel, buf, blkno, page,
-							vmbuffer,
+									   vmbuffer,
-							blk_info & VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM,
+									   blk_info & VAC_BLK_ALL_VISIBLE_ACCORDING_TO_VM,
-							&has_lpdead_items, &vm_page_frozen);
+									   &has_lpdead_items, &vm_page_frozen);
 
 		/*
 		 * Count an eagerly scanned page as a failure or a success.
@@ -1481,7 +1482,7 @@ lazy_scan_heap(LVRelState *vacrel)
 			 * table has indexes. There will only be newly-freed space if we
 			 * held the cleanup lock and lazy_scan_prune() was called.
 			 */
-			if (got_cleanup_lock && vacrel->nindexes == 0 && has_lpdead_items &&
+			if (got_cleanup_lock && vacrel->nindexes == 0 && ndeleted > 0 &&
 				blkno - next_fsm_block_to_vacuum >= VACUUM_FSM_EVERY_PAGES)
 			{
 				FreeSpaceMapVacuumRange(vacrel->rel, next_fsm_block_to_vacuum,
@@ -1936,8 +1937,10 @@ cmpOffsetNumbers(const void *a, const void *b)
  * *vm_page_frozen is set to true if the page is newly set all-frozen in the
  * VM. The caller currently only uses this for determining whether an eagerly
  * scanned page was successfully set all-frozen.
+ *
+ * Returns the number of tuples deleted from the page during HOT pruning.
  */
-static void
+static int
 lazy_scan_prune(LVRelState *vacrel,
 				Buffer buf,
 				BlockNumber blkno,
@@ -2208,6 +2211,8 @@ lazy_scan_prune(LVRelState *vacrel,
 			*vm_page_frozen = true;
 		}
 	}
+
+	return presult.ndeleted;
 }
 
 /*

src/backend/access/nbtree/nbtpreprocesskeys.c

@@ -16,6 +16,7 @@
 #include "postgres.h"
 
 #include "access/nbtree.h"
+#include "common/int.h"
 #include "lib/qunique.h"
 #include "utils/array.h"
 #include "utils/lsyscache.h"
@@ -56,6 +57,8 @@ static void _bt_skiparray_strat_decrement(IndexScanDesc scan, ScanKey arraysk,
 										  BTArrayKeyInfo *array);
 static void _bt_skiparray_strat_increment(IndexScanDesc scan, ScanKey arraysk,
 										  BTArrayKeyInfo *array);
+static void _bt_unmark_keys(IndexScanDesc scan, int *keyDataMap);
+static int	_bt_reorder_array_cmp(const void *a, const void *b);
 static ScanKey _bt_preprocess_array_keys(IndexScanDesc scan, int *new_numberOfKeys);
 static void _bt_preprocess_array_keys_final(IndexScanDesc scan, int *keyDataMap);
 static int	_bt_num_array_keys(IndexScanDesc scan, Oid *skip_eq_ops_out,
@@ -96,7 +99,7 @@ static int	_bt_compare_array_elements(const void *a, const void *b, void *arg);
  * incomplete sets of cross-type operators, we may fail to detect redundant
  * or contradictory keys, but we can survive that.)
  *
- * The output keys must be sorted by index attribute.  Presently we expect
+ * Required output keys are sorted by index attribute.  Presently we expect
  * (but verify) that the input keys are already so sorted --- this is done
  * by match_clauses_to_index() in indxpath.c.  Some reordering of the keys
  * within each attribute may be done as a byproduct of the processing here.
@@ -127,29 +130,36 @@ static int	_bt_compare_array_elements(const void *a, const void *b, void *arg);
  * This has the potential to be much more efficient than a full index scan
  * (though it behaves like a full scan when there's many distinct "x" values).
  *
- * If possible, redundant keys are eliminated: we keep only the tightest
+ * Typically, redundant keys are eliminated: we keep only the tightest
  * >/>= bound and the tightest </<= bound, and if there's an = key then
  * that's the only one returned.  (So, we return either a single = key,
  * or one or two boundary-condition keys for each attr.)  However, if we
  * cannot compare two keys for lack of a suitable cross-type operator,
- * we cannot eliminate either.  If there are two such keys of the same
+ * we cannot eliminate either key.
- * operator strategy, the second one is just pushed into the output array
- * without further processing here.  We may also emit both >/>= or both
- * </<= keys if we can't compare them.  The logic about required keys still
- * works if we don't eliminate redundant keys.
  *
- * Note that one reason we need direction-sensitive required-key flags is
+ * When all redundant keys could not be eliminated, we'll output a key array
- * precisely that we may not be able to eliminate redundant keys.  Suppose
+ * that can more or less be treated as if it had no redundant keys.  Suppose
- * we have "x > 4::int AND x > 10::bigint", and we are unable to determine
+ * we have "x > 4::int AND x > 10::bigint AND x < 70", and we are unable to
- * which key is more restrictive for lack of a suitable cross-type operator.
+ * determine which > key is more restrictive for lack of a suitable cross-type
- * _bt_first will arbitrarily pick one of the keys to do the initial
+ * operator.  We'll arbitrarily pick one of the > keys; the other > key won't
- * positioning with.  If it picks x > 4, then the x > 10 condition will fail
+ * be marked required.  Obviously, the scan will be less efficient if we
- * until we reach index entries > 10; but we can't stop the scan just because
+ * choose x > 4 over x > 10 -- but it can still largely proceed as if there
- * x > 10 is failing.  On the other hand, if we are scanning backwards, then
+ * was only a single > condition.  "x > 10" will be placed at the end of the
- * failure of either key is indeed enough to stop the scan.  (In general, when
+ * so->keyData[] output array.  It'll always be evaluated last, after the keys
- * inequality keys are present, the initial-positioning code only promises to
+ * that could be marked required in the usual way (after "x > 4 AND x < 70").
- * position before the first possible match, not exactly at the first match,
+ * This can sometimes result in so->keyData[] keys that aren't even in index
- * for a forward scan; or after the last match for a backward scan.)
+ * attribute order (if the qual involves multiple attributes).  The scan's
+ * required keys will still be in attribute order, though, so it can't matter.
+ *
+ * This scheme ensures that _bt_first always uses the same set of keys at the
+ * start of a forwards scan as those _bt_checkkeys uses to determine when to
+ * end a similar backwards scan (and vice-versa).  _bt_advance_array_keys
+ * depends on this: it expects to be able to reliably predict what the next
+ * _bt_first call will do by testing whether _bt_checkkeys' routines report
+ * that the final tuple on the page is past the end of matches for the scan's
+ * keys with the scan direction flipped.  If it is (if continuescan=false),
+ * then it follows that calling _bt_first will, at a minimum, relocate the
+ * scan to the very next leaf page (in the current scan direction).
  *
  * As a byproduct of this work, we can detect contradictory quals such
  * as "x = 1 AND x > 2".  If we see that, we return so->qual_ok = false,
@@ -188,7 +198,8 @@ _bt_preprocess_keys(IndexScanDesc scan)
 	int			numberOfEqualCols;
 	ScanKey		inkeys;
 	BTScanKeyPreproc xform[BTMaxStrategyNumber];
-	bool		test_result;
+	bool		test_result,
+				redundant_key_kept = false;
 	AttrNumber	attno;
 	ScanKey		arrayKeyData;
 	int		   *keyDataMap = NULL;
@@ -388,7 +399,8 @@ _bt_preprocess_keys(IndexScanDesc scan)
 						xform[j].inkey = NULL;
 						xform[j].inkeyi = -1;
 					}
-					/* else, cannot determine redundancy, keep both keys */
+					else
+						redundant_key_kept = true;
 				}
 				/* track number of attrs for which we have "=" keys */
 				numberOfEqualCols++;
@@ -409,6 +421,8 @@ _bt_preprocess_keys(IndexScanDesc scan)
 					else
 						xform[BTLessStrategyNumber - 1].inkey = NULL;
 				}
+				else
+					redundant_key_kept = true;
 			}
 
 			/* try to keep only one of >, >= */
@@ -426,6 +440,8 @@ _bt_preprocess_keys(IndexScanDesc scan)
 					else
 						xform[BTGreaterStrategyNumber - 1].inkey = NULL;
 				}
+				else
+					redundant_key_kept = true;
 			}
 
 			/*
@@ -466,25 +482,6 @@ _bt_preprocess_keys(IndexScanDesc scan)
 		/* check strategy this key's operator corresponds to */
 		j = inkey->sk_strategy - 1;
 
-		/* if row comparison, push it directly to the output array */
-		if (inkey->sk_flags & SK_ROW_HEADER)
-		{
-			ScanKey		outkey = &so->keyData[new_numberOfKeys++];
-
-			memcpy(outkey, inkey, sizeof(ScanKeyData));
-			if (arrayKeyData)
-				keyDataMap[new_numberOfKeys - 1] = i;
-			if (numberOfEqualCols == attno - 1)
-				_bt_mark_scankey_required(outkey);
-
-			/*
-			 * We don't support RowCompare using equality; such a qual would
-			 * mess up the numberOfEqualCols tracking.
-			 */
-			Assert(j != (BTEqualStrategyNumber - 1));
-			continue;
-		}
-
 		if (inkey->sk_strategy == BTEqualStrategyNumber &&
 			(inkey->sk_flags & SK_SEARCHARRAY))
 		{
@@ -593,9 +590,8 @@ _bt_preprocess_keys(IndexScanDesc scan)
 				 * the new scan key.
 				 *
 				 * Note: We do things this way around so that our arrays are
-				 * always in the same order as their corresponding scan keys,
+				 * always in the same order as their corresponding scan keys.
-				 * even with incomplete opfamilies.  _bt_advance_array_keys
+				 * _bt_preprocess_array_keys_final expects this.
-				 * depends on this.
 				 */
 				ScanKey		outkey = &so->keyData[new_numberOfKeys++];
 
@@ -607,6 +603,7 @@ _bt_preprocess_keys(IndexScanDesc scan)
 				xform[j].inkey = inkey;
 				xform[j].inkeyi = i;
 				xform[j].arrayidx = arrayidx;
+				redundant_key_kept = true;
 			}
 		}
 	}
@@ -622,6 +619,15 @@ _bt_preprocess_keys(IndexScanDesc scan)
 	if (arrayKeyData)
 		_bt_preprocess_array_keys_final(scan, keyDataMap);
 
+	/*
+	 * If there are remaining redundant inequality keys, we must make sure
+	 * that each index attribute has no more than one required >/>= key, and
+	 * no more than one required </<= key.  Attributes that have one or more
+	 * required = keys now must keep only one required key (the first = key).
+	 */
+	if (unlikely(redundant_key_kept) && so->qual_ok)
+		_bt_unmark_keys(scan, keyDataMap);
+
 	/* Could pfree arrayKeyData/keyDataMap now, but not worth the cycles */
 }
 
@@ -746,9 +752,12 @@ _bt_fix_scankey_strategy(ScanKey skey, int16 *indoption)
  *
  * Depending on the operator type, the key may be required for both scan
  * directions or just one.  Also, if the key is a row comparison header,
- * we have to mark its first subsidiary ScanKey as required.  (Subsequent
+ * we have to mark the appropriate subsidiary ScanKeys as required.  In such
- * subsidiary ScanKeys are normally for lower-order columns, and thus
+ * cases, the first subsidiary key is required, but subsequent ones are
- * cannot be required, since they're after the first non-equality scankey.)
+ * required only as long as they correspond to successive index columns and
+ * match the leading column as to sort direction.  Otherwise the row
+ * comparison ordering is different from the index ordering and so we can't
+ * stop the scan on the basis of those lower-order columns.
  *
  * Note: when we set required-key flag bits in a subsidiary scankey, we are
  * scribbling on a data structure belonging to the index AM's caller, not on
@@ -786,12 +795,25 @@ _bt_mark_scankey_required(ScanKey skey)
 	if (skey->sk_flags & SK_ROW_HEADER)
 	{
 		ScanKey		subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
+		AttrNumber	attno = skey->sk_attno;
 
 		/* First subkey should be same column/operator as the header */
-		Assert(subkey->sk_flags & SK_ROW_MEMBER);
+		Assert(subkey->sk_attno == attno);
-		Assert(subkey->sk_attno == skey->sk_attno);
 		Assert(subkey->sk_strategy == skey->sk_strategy);
-		subkey->sk_flags |= addflags;
+
+		for (;;)
+		{
+			Assert(subkey->sk_flags & SK_ROW_MEMBER);
+			if (subkey->sk_attno != attno)
+				break;			/* non-adjacent key, so not required */
+			if (subkey->sk_strategy != skey->sk_strategy)
+				break;			/* wrong direction, so not required */
+			subkey->sk_flags |= addflags;
+			if (subkey->sk_flags & SK_ROW_END)
+				break;
+			subkey++;
+			attno++;
+		}
 	}
 }
 
@@ -847,8 +869,7 @@ _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op,
 				cmp_op;
 	StrategyNumber strat;
 
-	Assert(!((leftarg->sk_flags | rightarg->sk_flags) &
+	Assert(!((leftarg->sk_flags | rightarg->sk_flags) & SK_ROW_MEMBER));
-			 (SK_ROW_HEADER | SK_ROW_MEMBER)));
 
 	/*
 	 * First, deal with cases where one or both args are NULL.  This should
@@ -924,6 +945,16 @@ _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op,
 		return true;
 	}
 
+	/*
+	 * We don't yet know how to determine redundancy when it involves a row
+	 * compare key (barring simple cases involving IS NULL/IS NOT NULL)
+	 */
+	if ((leftarg->sk_flags | rightarg->sk_flags) & SK_ROW_HEADER)
+	{
+		Assert(!((leftarg->sk_flags | rightarg->sk_flags) & SK_BT_SKIP));
+		return false;
+	}
+
 	/*
 	 * If either leftarg or rightarg are equality-type array scankeys, we need
 	 * specialized handling (since by now we know that IS NULL wasn't used)
@@ -1467,6 +1498,283 @@ _bt_skiparray_strat_increment(IndexScanDesc scan, ScanKey arraysk,
 	}
 }
 
+/*
+ *	_bt_unmark_keys() -- make superfluous required keys nonrequired after all
+ *
+ * When _bt_preprocess_keys fails to eliminate one or more redundant keys, it
+ * calls here to make sure that no index attribute has more than one > or >=
+ * key marked required, and no more than one required < or <= key.  Attributes
+ * with = keys will always get one = key as their required key.  All other
+ * keys that were initially marked required get "unmarked" here.  That way,
+ * _bt_first and _bt_checkkeys will reliably agree on which keys to use to
+ * start and/or to end the scan.
+ *
+ * We also relocate keys that become/started out nonrequired to the end of
+ * so->keyData[].  That way, _bt_first and _bt_checkkeys cannot fail to reach
+ * a required key due to some earlier nonrequired key getting in the way.
+ *
+ * Only call here when _bt_compare_scankey_args returned false at least once
+ * (otherwise, calling here will just waste cycles).
+ */
+static void
+_bt_unmark_keys(IndexScanDesc scan, int *keyDataMap)
+{
+	BTScanOpaque so = (BTScanOpaque) scan->opaque;
+	AttrNumber	attno;
+	bool	   *unmarkikey;
+	int			nunmark,
+				nunmarked,
+				nkept,
+				firsti;
+	ScanKey		keepKeys,
+				unmarkKeys;
+	FmgrInfo   *keepOrderProcs = NULL,
+			   *unmarkOrderProcs = NULL;
+	bool		haveReqEquals,
+				haveReqForward,
+				haveReqBackward;
+
+	/*
+	 * Do an initial pass over so->keyData[] that determines which keys to
+	 * keep as required.  We expect so->keyData[] to still be in attribute
+	 * order when we're called (though we don't expect any particular order
+	 * among each attribute's keys).
+	 *
+	 * When both equality and inequality keys remain on a single attribute, we
+	 * *must* make sure that exactly one of the equalities remains required.
+	 * Any requiredness markings that we might leave on later keys/attributes
+	 * are predicated on there being required = keys on all prior columns.
+	 */
+	unmarkikey = palloc0(so->numberOfKeys * sizeof(bool));
+	nunmark = 0;
+
+	/* Set things up for first key's attribute */
+	attno = so->keyData[0].sk_attno;
+	firsti = 0;
+	haveReqEquals = false;
+	haveReqForward = false;
+	haveReqBackward = false;
+	for (int i = 0; i < so->numberOfKeys; i++)
+	{
+		ScanKey		origkey = &so->keyData[i];
+
+		if (origkey->sk_attno != attno)
+		{
+			/* Reset for next attribute */
+			attno = origkey->sk_attno;
+			firsti = i;
+
+			haveReqEquals = false;
+			haveReqForward = false;
+			haveReqBackward = false;
+		}
+
+		/* Equalities get priority over inequalities */
+		if (haveReqEquals)
+		{
+			/*
+			 * We already found the first "=" key for this attribute.  We've
+			 * already decided that all its other keys will be unmarked.
+			 */
+			Assert(!(origkey->sk_flags & SK_SEARCHNULL));
+			unmarkikey[i] = true;
+			nunmark++;
+			continue;
+		}
+		else if ((origkey->sk_flags & SK_BT_REQFWD) &&
+				 (origkey->sk_flags & SK_BT_REQBKWD))
+		{
+			/*
+			 * Found the first "=" key for attno.  All other attno keys will
+			 * be unmarked.
+			 */
+			Assert(origkey->sk_strategy == BTEqualStrategyNumber);
+
+			haveReqEquals = true;
+			for (int j = firsti; j < i; j++)
+			{
+				/* Unmark any prior inequality keys on attno after all */
+				if (!unmarkikey[j])
+				{
+					unmarkikey[j] = true;
+					nunmark++;
+				}
+			}
+			continue;
+		}
+
+		/* Deal with inequalities next */
+		if ((origkey->sk_flags & SK_BT_REQFWD) && !haveReqForward)
+		{
+			haveReqForward = true;
+			continue;
+		}
+		else if ((origkey->sk_flags & SK_BT_REQBKWD) && !haveReqBackward)
+		{
+			haveReqBackward = true;
+			continue;
+		}
+
+		/*
+		 * We have either a redundant inequality key that will be unmarked, or
+		 * we have a key that wasn't marked required in the first place
+		 */
+		unmarkikey[i] = true;
+		nunmark++;
+	}
+
+	/* Should only be called when _bt_compare_scankey_args reported failure */
+	Assert(nunmark > 0);
+
+	/*
+	 * Next, allocate temp arrays: one for required keys that'll remain
+	 * required, the other for all remaining keys
+	 */
+	unmarkKeys = palloc(nunmark * sizeof(ScanKeyData));
+	keepKeys = palloc((so->numberOfKeys - nunmark) * sizeof(ScanKeyData));
+	nunmarked = 0;
+	nkept = 0;
+	if (so->numArrayKeys)
+	{
+		unmarkOrderProcs = palloc(nunmark * sizeof(FmgrInfo));
+		keepOrderProcs = palloc((so->numberOfKeys - nunmark) * sizeof(FmgrInfo));
+	}
+
+	/*
+	 * Next, copy the contents of so->keyData[] into the appropriate temp
+	 * array.
+	 *
+	 * Scans with = array keys need us to maintain invariants around the order
+	 * of so->orderProcs[] and so->arrayKeys[] relative to so->keyData[].  See
+	 * _bt_preprocess_array_keys_final for a full explanation.
+	 */
+	for (int i = 0; i < so->numberOfKeys; i++)
+	{
+		ScanKey		origkey = &so->keyData[i];
+		ScanKey		unmark;
+
+		if (!unmarkikey[i])
+		{
+			/*
+			 * Key gets to keep its original requiredness markings.
+			 *
+			 * Key will stay in its original position, unless we're going to
+			 * unmark an earlier key (in which case this key gets moved back).
+			 */
+			memcpy(keepKeys + nkept, origkey, sizeof(ScanKeyData));
+
+			if (so->numArrayKeys)
+			{
+				keyDataMap[i] = nkept;
+				memcpy(keepOrderProcs + nkept, &so->orderProcs[i],
+					   sizeof(FmgrInfo));
+			}
+
+			nkept++;
+			continue;
+		}
+
+		/*
+		 * Key will be unmarked as needed, and moved to the end of the array,
+		 * next to other keys that will become (or always were) nonrequired
+		 */
+		unmark = unmarkKeys + nunmarked;
+		memcpy(unmark, origkey, sizeof(ScanKeyData));
+
+		if (so->numArrayKeys)
+		{
+			keyDataMap[i] = (so->numberOfKeys - nunmark) + nunmarked;
+			memcpy(&unmarkOrderProcs[nunmarked], &so->orderProcs[i],
+				   sizeof(FmgrInfo));
+		}
+
+		/*
+		 * Preprocessing only generates skip arrays when it knows that they'll
+		 * be the only required = key on the attr.  We'll never unmark them.
+		 */
+		Assert(!(unmark->sk_flags & SK_BT_SKIP));
+
+		/*
+		 * Also shouldn't have to unmark an IS NULL or an IS NOT NULL key.
+		 * They aren't cross-type, so an incomplete opfamily can't matter.
+		 */
+		Assert(!(unmark->sk_flags & SK_ISNULL) ||
+			   !(unmark->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)));
+
+		/* Clear requiredness flags on redundant key (and on any subkeys) */
+		unmark->sk_flags &= ~(SK_BT_REQFWD | SK_BT_REQBKWD);
+		if (unmark->sk_flags & SK_ROW_HEADER)
+		{
+			ScanKey		subkey = (ScanKey) DatumGetPointer(unmark->sk_argument);
+
+			Assert(subkey->sk_strategy == unmark->sk_strategy);
+			for (;;)
+			{
+				Assert(subkey->sk_flags & SK_ROW_MEMBER);
+				subkey->sk_flags &= ~(SK_BT_REQFWD | SK_BT_REQBKWD);
+				if (subkey->sk_flags & SK_ROW_END)
+					break;
+				subkey++;
+			}
+		}
+
+		nunmarked++;
+	}
+
+	/* Copy both temp arrays back into so->keyData[] to reorder */
+	Assert(nkept == so->numberOfKeys - nunmark);
+	Assert(nunmarked == nunmark);
+	memcpy(so->keyData, keepKeys, sizeof(ScanKeyData) * nkept);
+	memcpy(so->keyData + nkept, unmarkKeys, sizeof(ScanKeyData) * nunmarked);
+
+	/* Done with temp arrays */
+	pfree(unmarkikey);
+	pfree(keepKeys);
+	pfree(unmarkKeys);
+
+	/*
+	 * Now copy so->orderProcs[] temp entries needed by scans with = array
+	 * keys back (just like with the so->keyData[] temp arrays)
+	 */
+	if (so->numArrayKeys)
+	{
+		memcpy(so->orderProcs, keepOrderProcs, sizeof(FmgrInfo) * nkept);
+		memcpy(so->orderProcs + nkept, unmarkOrderProcs,
+			   sizeof(FmgrInfo) * nunmarked);
+
+		/* Also fix-up array->scan_key references */
+		for (int arridx = 0; arridx < so->numArrayKeys; arridx++)
+		{
+			BTArrayKeyInfo *array = &so->arrayKeys[arridx];
+
+			array->scan_key = keyDataMap[array->scan_key];
+		}
+
+		/*
+		 * Sort so->arrayKeys[] based on its new BTArrayKeyInfo.scan_key
+		 * offsets, so that its order matches so->keyData[] order as expected
+		 */
+		qsort(so->arrayKeys, so->numArrayKeys, sizeof(BTArrayKeyInfo),
+			  _bt_reorder_array_cmp);
+
+		/* Done with temp arrays */
+		pfree(unmarkOrderProcs);
+		pfree(keepOrderProcs);
+	}
+}
+
+/*
+ * qsort comparator for reordering so->arrayKeys[] BTArrayKeyInfo entries
+ */
+static int
+_bt_reorder_array_cmp(const void *a, const void *b)
+{
+	BTArrayKeyInfo *arraya = (BTArrayKeyInfo *) a;
+	BTArrayKeyInfo *arrayb = (BTArrayKeyInfo *) b;
+
+	return pg_cmp_s32(arraya->scan_key, arrayb->scan_key);
+}
+
 /*
  *	_bt_preprocess_array_keys() -- Preprocess SK_SEARCHARRAY scan keys
  *

src/backend/access/nbtree/nbtsearch.c

@@ -960,46 +960,51 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 
 	/*----------
 	 * Examine the scan keys to discover where we need to start the scan.
+	 * The selected scan keys (at most one per index column) are remembered by
+	 * storing their addresses into the local startKeys[] array.  The final
+	 * startKeys[] entry's strategy is set in strat_total. (Actually, there
+	 * are a couple of cases where we force a less/more restrictive strategy.)
 	 *
-	 * We want to identify the keys that can be used as starting boundaries;
+	 * We must use the key that was marked required (in the direction opposite
-	 * these are =, >, or >= keys for a forward scan or =, <, <= keys for
+	 * our own scan's) during preprocessing.  Each index attribute can only
-	 * a backwards scan.  We can use keys for multiple attributes so long as
+	 * have one such required key.  In general, the keys that we use to find
-	 * the prior attributes had only =, >= (resp. =, <=) keys.  Once we accept
+	 * an initial position when scanning forwards are the same keys that end
-	 * a > or < boundary or find an attribute with no boundary (which can be
+	 * the scan on the leaf level when scanning backwards (and vice-versa).
-	 * thought of as the same as "> -infinity"), we can't use keys for any
-	 * attributes to its right, because it would break our simplistic notion
-	 * of what initial positioning strategy to use.
 	 *
 	 * When the scan keys include cross-type operators, _bt_preprocess_keys
-	 * may not be able to eliminate redundant keys; in such cases we will
+	 * may not be able to eliminate redundant keys; in such cases it will
-	 * arbitrarily pick a usable one for each attribute.  This is correct
+	 * arbitrarily pick a usable key for each attribute (and scan direction),
-	 * but possibly not optimal behavior.  (For example, with keys like
+	 * ensuring that there is no more than one key required in each direction.
-	 * "x >= 4 AND x >= 5" we would elect to scan starting at x=4 when
+	 * We stop considering further keys once we reach the first nonrequired
-	 * x=5 would be more efficient.)  Since the situation only arises given
+	 * key (which must come after all required keys), so this can't affect us.
-	 * a poorly-worded query plus an incomplete opfamily, live with it.
 	 *
-	 * When both equality and inequality keys appear for a single attribute
+	 * The required keys that we use as starting boundaries have to be =, >,
-	 * (again, only possible when cross-type operators appear), we *must*
+	 * or >= keys for a forward scan or =, <, <= keys for a backwards scan.
-	 * select one of the equality keys for the starting point, because
+	 * We can use keys for multiple attributes so long as the prior attributes
-	 * _bt_checkkeys() will stop the scan as soon as an equality qual fails.
+	 * had only =, >= (resp. =, <=) keys.  These rules are very similar to the
-	 * For example, if we have keys like "x >= 4 AND x = 10" and we elect to
+	 * rules that preprocessing used to determine which keys to mark required.
-	 * start at x=4, we will fail and stop before reaching x=10.  If multiple
+	 * We cannot always use every required key as a positioning key, though.
-	 * equality quals survive preprocessing, however, it doesn't matter which
+	 * Skip arrays necessitate independently applying our own rules here.
-	 * one we use --- by definition, they are either redundant or
+	 * Skip arrays are always generally considered = array keys, but we'll
-	 * contradictory.
+	 * nevertheless treat them as inequalities at certain points of the scan.
+	 * When that happens, it _might_ have implications for the number of
+	 * required keys that we can safely use for initial positioning purposes.
 	 *
-	 * In practice we rarely see any "attribute boundary key gaps" here.
+	 * For example, a forward scan with a skip array on its leading attribute
-	 * Preprocessing can usually backfill skip array keys for any attributes
+	 * (with no low_compare/high_compare) will have at least two required scan
-	 * that were omitted from the original scan->keyData[] input keys.  All
+	 * keys, but we won't use any of them as boundary keys during the scan's
-	 * array keys are always considered = keys, but we'll sometimes need to
+	 * initial call here.  Our positioning key during the first call here can
-	 * treat the current key value as if we were using an inequality strategy.
+	 * be thought of as representing "> -infinity".  Similarly, if such a skip
-	 * This happens with range skip arrays, which store inequality keys in the
+	 * array's low_compare is "a > 'foo'", then we position using "a > 'foo'"
-	 * array's low_compare/high_compare fields (used to find the first/last
+	 * during the scan's initial call here; a lower-order key such as "b = 42"
-	 * set of matches, when = key will lack a usable sk_argument value).
+	 * can't be used until the "a" array advances beyond MINVAL/low_compare.
-	 * These are always preferred over any redundant "standard" inequality
+	 *
-	 * keys on the same column (per the usual rule about preferring = keys).
+	 * On the other hand, if such a skip array's low_compare was "a >= 'foo'",
-	 * Note also that any column with an = skip array key can never have an
+	 * then we _can_ use "a >= 'foo' AND b = 42" during the initial call here.
-	 * additional, contradictory = key.
+	 * A subsequent call here might have us use "a = 'fop' AND b = 42".  Note
+	 * that we treat = and >= as equivalent when scanning forwards (just as we
+	 * treat = and <= as equivalent when scanning backwards).  We effectively
+	 * do the same thing (though with a distinct "a" element/value) each time.
 	 *
 	 * All keys (with the exception of SK_SEARCHNULL keys and SK_BT_SKIP
 	 * array keys whose array is "null_elem=true") imply a NOT NULL qualifier.
@@ -1011,21 +1016,20 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	 * traversing a lot of null entries at the start of the scan.
 	 *
 	 * In this loop, row-comparison keys are treated the same as keys on their
-	 * first (leftmost) columns.  We'll add on lower-order columns of the row
+	 * first (leftmost) columns.  We'll add all lower-order columns of the row
-	 * comparison below, if possible.
+	 * comparison that were marked required during preprocessing below.
 	 *
-	 * The selected scan keys (at most one per index column) are remembered by
+	 * _bt_advance_array_keys needs to know exactly how we'll reposition the
-	 * storing their addresses into the local startKeys[] array.
+	 * scan (should it opt to schedule another primitive index scan).  It is
-	 *
+	 * critical that primscans only be scheduled when they'll definitely make
-	 * _bt_checkkeys/_bt_advance_array_keys decide whether and when to start
+	 * some useful progress.  _bt_advance_array_keys does this by calling
-	 * the next primitive index scan (for scans with array keys) based in part
+	 * _bt_checkkeys routines that report whether a tuple is past the end of
-	 * on an understanding of how it'll enable us to reposition the scan.
+	 * matches for the scan's keys (given the scan's current array elements).
-	 * They're directly aware of how we'll sometimes cons up an explicit
+	 * If the page's final tuple is "after the end of matches" for a scan that
-	 * SK_SEARCHNOTNULL key.  They'll even end primitive scans by applying a
+	 * uses the *opposite* scan direction, then it must follow that it's also
-	 * symmetric "deduce NOT NULL" rule of their own.  This allows top-level
+	 * "before the start of matches" for the actual current scan direction.
-	 * scans to skip large groups of NULLs through repeated deductions about
+	 * It is therefore essential that all of our initial positioning rules are
-	 * key strictness (for a required inequality key) and whether NULLs in the
+	 * symmetric with _bt_checkkeys's corresponding continuescan=false rule.
-	 * key's index column are stored last or first (relative to non-NULLs).
 	 * If you update anything here, _bt_checkkeys/_bt_advance_array_keys might
 	 * need to be kept in sync.
 	 *----------
@@ -1034,18 +1038,17 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	if (so->numberOfKeys > 0)
 	{
 		AttrNumber	curattr;
-		ScanKey		chosen;
+		ScanKey		bkey;
 		ScanKey		impliesNN;
 		ScanKey		cur;
 
 		/*
-		 * chosen is the so-far-chosen key for the current attribute, if any.
+		 * bkey will be set to the key that preprocessing left behind as the
-		 * We don't cast the decision in stone until we reach keys for the
+		 * boundary key for this attribute, in this scan direction (if any)
-		 * next attribute.
 		 */
 		cur = so->keyData;
 		curattr = 1;
-		chosen = NULL;
+		bkey = NULL;
 		/* Also remember any scankey that implies a NOT NULL constraint */
 		impliesNN = NULL;
 
@@ -1058,23 +1061,29 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 		{
 			if (i >= so->numberOfKeys || cur->sk_attno != curattr)
 			{
+				/* Done looking for the curattr boundary key */
+				Assert(bkey == NULL ||
+					   (bkey->sk_attno == curattr &&
+						(bkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))));
+				Assert(impliesNN == NULL ||
+					   (impliesNN->sk_attno == curattr &&
+						(impliesNN->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))));
+
 				/*
-				 * Done looking at keys for curattr.
-				 *
 				 * If this is a scan key for a skip array whose current
 				 * element is MINVAL, choose low_compare (when scanning
 				 * backwards it'll be MAXVAL, and we'll choose high_compare).
 				 *
-				 * Note: if the array's low_compare key makes 'chosen' NULL,
+				 * Note: if the array's low_compare key makes 'bkey' NULL,
 				 * then we behave as if the array's first element is -inf,
 				 * except when !array->null_elem implies a usable NOT NULL
 				 * constraint.
 				 */
-				if (chosen != NULL &&
+				if (bkey != NULL &&
-					(chosen->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL)))
+					(bkey->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL)))
 				{
-					int			ikey = chosen - so->keyData;
+					int			ikey = bkey - so->keyData;
-					ScanKey		skipequalitykey = chosen;
+					ScanKey		skipequalitykey = bkey;
 					BTArrayKeyInfo *array = NULL;
 
 					for (int arridx = 0; arridx < so->numArrayKeys; arridx++)
@@ -1087,35 +1096,35 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 					if (ScanDirectionIsForward(dir))
 					{
 						Assert(!(skipequalitykey->sk_flags & SK_BT_MAXVAL));
-						chosen = array->low_compare;
+						bkey = array->low_compare;
 					}
 					else
 					{
 						Assert(!(skipequalitykey->sk_flags & SK_BT_MINVAL));
-						chosen = array->high_compare;
+						bkey = array->high_compare;
 					}
 
-					Assert(chosen == NULL ||
+					Assert(bkey == NULL ||
-						   chosen->sk_attno == skipequalitykey->sk_attno);
+						   bkey->sk_attno == skipequalitykey->sk_attno);
 
 					if (!array->null_elem)
 						impliesNN = skipequalitykey;
 					else
-						Assert(chosen == NULL && impliesNN == NULL);
+						Assert(bkey == NULL && impliesNN == NULL);
 				}
 
 				/*
 				 * If we didn't find a usable boundary key, see if we can
 				 * deduce a NOT NULL key
 				 */
-				if (chosen == NULL && impliesNN != NULL &&
+				if (bkey == NULL && impliesNN != NULL &&
 					((impliesNN->sk_flags & SK_BT_NULLS_FIRST) ?
 					 ScanDirectionIsForward(dir) :
 					 ScanDirectionIsBackward(dir)))
 				{
 					/* Yes, so build the key in notnullkeys[keysz] */
-					chosen = &notnullkeys[keysz];
+					bkey = &notnullkeys[keysz];
-					ScanKeyEntryInitialize(chosen,
+					ScanKeyEntryInitialize(bkey,
 										   (SK_SEARCHNOTNULL | SK_ISNULL |
 											(impliesNN->sk_flags &
 											 (SK_BT_DESC | SK_BT_NULLS_FIRST))),
@@ -1130,12 +1139,12 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 				}
 
 				/*
-				 * If we still didn't find a usable boundary key, quit; else
+				 * If preprocessing didn't leave a usable boundary key, quit;
-				 * save the boundary key pointer in startKeys.
+				 * else save the boundary key pointer in startKeys[]
 				 */
-				if (chosen == NULL)
+				if (bkey == NULL)
 					break;
-				startKeys[keysz++] = chosen;
+				startKeys[keysz++] = bkey;
 
 				/*
 				 * We can only consider adding more boundary keys when the one
@@ -1143,7 +1152,7 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 				 * (during backwards scans we can only do so when the key that
 				 * we just added to startKeys[] uses the = or <= strategy)
 				 */
-				strat_total = chosen->sk_strategy;
+				strat_total = bkey->sk_strategy;
 				if (strat_total == BTGreaterStrategyNumber ||
 					strat_total == BTLessStrategyNumber)
 					break;
@@ -1154,19 +1163,19 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 				 * make strat_total > or < (and stop adding boundary keys).
 				 * This can only happen with opclasses that lack skip support.
 				 */
-				if (chosen->sk_flags & (SK_BT_NEXT | SK_BT_PRIOR))
+				if (bkey->sk_flags & (SK_BT_NEXT | SK_BT_PRIOR))
 				{
-					Assert(chosen->sk_flags & SK_BT_SKIP);
+					Assert(bkey->sk_flags & SK_BT_SKIP);
 					Assert(strat_total == BTEqualStrategyNumber);
 
 					if (ScanDirectionIsForward(dir))
 					{
-						Assert(!(chosen->sk_flags & SK_BT_PRIOR));
+						Assert(!(bkey->sk_flags & SK_BT_PRIOR));
 						strat_total = BTGreaterStrategyNumber;
 					}
 					else
 					{
-						Assert(!(chosen->sk_flags & SK_BT_NEXT));
+						Assert(!(bkey->sk_flags & SK_BT_NEXT));
 						strat_total = BTLessStrategyNumber;
 					}
 
@@ -1180,24 +1189,30 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 
 				/*
 				 * Done if that was the last scan key output by preprocessing.
-				 * Also done if there is a gap index attribute that lacks a
+				 * Also done if we've now examined all keys marked required.
-				 * usable key (only possible when preprocessing was unable to
-				 * generate a skip array key to "fill in the gap").
 				 */
 				if (i >= so->numberOfKeys ||
-					cur->sk_attno != curattr + 1)
+					!(cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
 					break;
 
 				/*
 				 * Reset for next attr.
 				 */
+				Assert(cur->sk_attno == curattr + 1);
 				curattr = cur->sk_attno;
-				chosen = NULL;
+				bkey = NULL;
 				impliesNN = NULL;
 			}
 
 			/*
-			 * Can we use this key as a starting boundary for this attr?
+			 * If we've located the starting boundary key for curattr, we have
+			 * no interest in curattr's other required key
+			 */
+			if (bkey != NULL)
+				continue;
+
+			/*
+			 * Is this key the starting boundary key for curattr?
 			 *
 			 * If not, does it imply a NOT NULL constraint?  (Because
 			 * SK_SEARCHNULL keys are always assigned BTEqualStrategyNumber,
@@ -1207,27 +1222,20 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 			{
 				case BTLessStrategyNumber:
 				case BTLessEqualStrategyNumber:
-					if (chosen == NULL)
+					if (ScanDirectionIsBackward(dir))
-					{
+						bkey = cur;
-						if (ScanDirectionIsBackward(dir))
+					else if (impliesNN == NULL)
-							chosen = cur;
+						impliesNN = cur;
-						else
-							impliesNN = cur;
-					}
 					break;
 				case BTEqualStrategyNumber:
-					/* override any non-equality choice */
+					bkey = cur;
-					chosen = cur;
 					break;
 				case BTGreaterEqualStrategyNumber:
 				case BTGreaterStrategyNumber:
-					if (chosen == NULL)
+					if (ScanDirectionIsForward(dir))
-					{
+						bkey = cur;
-						if (ScanDirectionIsForward(dir))
+					else if (impliesNN == NULL)
-							chosen = cur;
+						impliesNN = cur;
-						else
-							impliesNN = cur;
-					}
 					break;
 			}
 		}
@@ -1253,16 +1261,18 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	Assert(keysz <= INDEX_MAX_KEYS);
 	for (int i = 0; i < keysz; i++)
 	{
-		ScanKey		cur = startKeys[i];
+		ScanKey		bkey = startKeys[i];
 
-		Assert(cur->sk_attno == i + 1);
+		Assert(bkey->sk_attno == i + 1);
 
-		if (cur->sk_flags & SK_ROW_HEADER)
+		if (bkey->sk_flags & SK_ROW_HEADER)
 		{
 			/*
 			 * Row comparison header: look to the first row member instead
 			 */
-			ScanKey		subkey = (ScanKey) DatumGetPointer(cur->sk_argument);
+			ScanKey		subkey = (ScanKey) DatumGetPointer(bkey->sk_argument);
+			bool		loosen_strat = false,
+						tighten_strat = false;
 
 			/*
 			 * Cannot be a NULL in the first row member: _bt_preprocess_keys
@@ -1270,9 +1280,18 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 			 * ever getting this far
 			 */
 			Assert(subkey->sk_flags & SK_ROW_MEMBER);
-			Assert(subkey->sk_attno == cur->sk_attno);
+			Assert(subkey->sk_attno == bkey->sk_attno);
 			Assert(!(subkey->sk_flags & SK_ISNULL));
 
+			/*
+			 * This is either a > or >= key (during backwards scans it is
+			 * either < or <=) that was marked required during preprocessing.
+			 * Later so->keyData[] keys can't have been marked required, so
+			 * our row compare header key must be the final startKeys[] entry.
+			 */
+			Assert(subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD));
+			Assert(i == keysz - 1);
+
 			/*
 			 * The member scankeys are already in insertion format (ie, they
 			 * have sk_func = 3-way-comparison function)
@@ -1280,112 +1299,141 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 			memcpy(inskey.scankeys + i, subkey, sizeof(ScanKeyData));
 
 			/*
-			 * If the row comparison is the last positioning key we accepted,
+			 * Now look to later row compare members.
-			 * try to add additional keys from the lower-order row members.
+			 *
-			 * (If we accepted independent conditions on additional index
+			 * If there's an "index attribute gap" between two row compare
-			 * columns, we use those instead --- doesn't seem worth trying to
+			 * members, the second member won't have been marked required, and
-			 * determine which is more restrictive.)  Note that this is OK
+			 * so can't be used as a starting boundary key here.  The part of
-			 * even if the row comparison is of ">" or "<" type, because the
+			 * the row comparison that we do still use has to be treated as a
-			 * condition applied to all but the last row member is effectively
+			 * ">=" or "<=" condition.  For example, a qual "(a, c) > (1, 42)"
-			 * ">=" or "<=", and so the extra keys don't break the positioning
+			 * with an omitted intervening index attribute "b" will use an
-			 * scheme.  But, by the same token, if we aren't able to use all
+			 * insertion scan key "a >= 1".  Even the first "a = 1" tuple on
-			 * the row members, then the part of the row comparison that we
+			 * the leaf level might satisfy the row compare qual.
-			 * did use has to be treated as just a ">=" or "<=" condition, and
+			 *
-			 * so we'd better adjust strat_total accordingly.
+			 * We're able to use a _more_ restrictive strategy when we reach a
+			 * NULL row compare member, since they're always unsatisfiable.
+			 * For example, a qual "(a, b, c) >= (1, NULL, 77)" will use an
+			 * insertion scan key "a > 1".  All tuples where "a = 1" cannot
+			 * possibly satisfy the row compare qual, so this is safe.
 			 */
-			if (i == keysz - 1)
+			Assert(!(subkey->sk_flags & SK_ROW_END));
+			for (;;)
 			{
-				bool		used_all_subkeys = false;
+				subkey++;
+				Assert(subkey->sk_flags & SK_ROW_MEMBER);
 
-				Assert(!(subkey->sk_flags & SK_ROW_END));
+				if (subkey->sk_flags & SK_ISNULL)
-				for (;;)
 				{
-					subkey++;
+					/*
-					Assert(subkey->sk_flags & SK_ROW_MEMBER);
+					 * NULL member key, can only use earlier keys.
-					if (subkey->sk_attno != keysz + 1)
+					 *
-						break;	/* out-of-sequence, can't use it */
+					 * We deliberately avoid checking if this key is marked
-					if (subkey->sk_strategy != cur->sk_strategy)
+					 * required.  All earlier keys are required, and this key
-						break;	/* wrong direction, can't use it */
+					 * is unsatisfiable either way, so we can't miss anything.
-					if (subkey->sk_flags & SK_ISNULL)
+					 */
-						break;	/* can't use null keys */
+					tighten_strat = true;
-					Assert(keysz < INDEX_MAX_KEYS);
+					break;
-					memcpy(inskey.scankeys + keysz, subkey,
-						   sizeof(ScanKeyData));
-					keysz++;
-					if (subkey->sk_flags & SK_ROW_END)
-					{
-						used_all_subkeys = true;
-						break;
-					}
 				}
-				if (!used_all_subkeys)
+
+				if (!(subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
 				{
-					switch (strat_total)
+					/* nonrequired member key, can only use earlier keys */
-					{
+					loosen_strat = true;
-						case BTLessStrategyNumber:
+					break;
-							strat_total = BTLessEqualStrategyNumber;
-							break;
-						case BTGreaterStrategyNumber:
-							strat_total = BTGreaterEqualStrategyNumber;
-							break;
-					}
 				}
-				break;			/* done with outer loop */
+
+				Assert(subkey->sk_attno == keysz + 1);
+				Assert(subkey->sk_strategy == bkey->sk_strategy);
+				Assert(keysz < INDEX_MAX_KEYS);
+
+				memcpy(inskey.scankeys + keysz, subkey,
+					   sizeof(ScanKeyData));
+				keysz++;
+				if (subkey->sk_flags & SK_ROW_END)
+					break;
 			}
+			Assert(!(loosen_strat && tighten_strat));
+			if (loosen_strat)
+			{
+				/* Use less restrictive strategy (and fewer member keys) */
+				switch (strat_total)
+				{
+					case BTLessStrategyNumber:
+						strat_total = BTLessEqualStrategyNumber;
+						break;
+					case BTGreaterStrategyNumber:
+						strat_total = BTGreaterEqualStrategyNumber;
+						break;
+				}
+			}
+			if (tighten_strat)
+			{
+				/* Use more restrictive strategy (and fewer member keys) */
+				switch (strat_total)
+				{
+					case BTLessEqualStrategyNumber:
+						strat_total = BTLessStrategyNumber;
+						break;
+					case BTGreaterEqualStrategyNumber:
+						strat_total = BTGreaterStrategyNumber;
+						break;
+				}
+			}
+
+			/* done adding to inskey (row comparison keys always come last) */
+			break;
+		}
+
+		/*
+		 * Ordinary comparison key/search-style key.
+		 *
+		 * Transform the search-style scan key to an insertion scan key by
+		 * replacing the sk_func with the appropriate btree 3-way-comparison
+		 * function.
+		 *
+		 * If scankey operator is not a cross-type comparison, we can use the
+		 * cached comparison function; otherwise gotta look it up in the
+		 * catalogs.  (That can't lead to infinite recursion, since no
+		 * indexscan initiated by syscache lookup will use cross-data-type
+		 * operators.)
+		 *
+		 * We support the convention that sk_subtype == InvalidOid means the
+		 * opclass input type; this hack simplifies life for ScanKeyInit().
+		 */
+		if (bkey->sk_subtype == rel->rd_opcintype[i] ||
+			bkey->sk_subtype == InvalidOid)
+		{
+			FmgrInfo   *procinfo;
+
+			procinfo = index_getprocinfo(rel, bkey->sk_attno, BTORDER_PROC);
+			ScanKeyEntryInitializeWithInfo(inskey.scankeys + i,
+										   bkey->sk_flags,
+										   bkey->sk_attno,
+										   InvalidStrategy,
+										   bkey->sk_subtype,
+										   bkey->sk_collation,
+										   procinfo,
+										   bkey->sk_argument);
 		}
 		else
 		{
-			/*
+			RegProcedure cmp_proc;
-			 * Ordinary comparison key.  Transform the search-style scan key
-			 * to an insertion scan key by replacing the sk_func with the
-			 * appropriate btree comparison function.
-			 *
-			 * If scankey operator is not a cross-type comparison, we can use
-			 * the cached comparison function; otherwise gotta look it up in
-			 * the catalogs.  (That can't lead to infinite recursion, since no
-			 * indexscan initiated by syscache lookup will use cross-data-type
-			 * operators.)
-			 *
-			 * We support the convention that sk_subtype == InvalidOid means
-			 * the opclass input type; this is a hack to simplify life for
-			 * ScanKeyInit().
-			 */
-			if (cur->sk_subtype == rel->rd_opcintype[i] ||
-				cur->sk_subtype == InvalidOid)
-			{
-				FmgrInfo   *procinfo;
 
-				procinfo = index_getprocinfo(rel, cur->sk_attno, BTORDER_PROC);
+			cmp_proc = get_opfamily_proc(rel->rd_opfamily[i],
-				ScanKeyEntryInitializeWithInfo(inskey.scankeys + i,
+										 rel->rd_opcintype[i],
-											   cur->sk_flags,
+										 bkey->sk_subtype, BTORDER_PROC);
-											   cur->sk_attno,
+			if (!RegProcedureIsValid(cmp_proc))
-											   InvalidStrategy,
+				elog(ERROR, "missing support function %d(%u,%u) for attribute %d of index \"%s\"",
-											   cur->sk_subtype,
+					 BTORDER_PROC, rel->rd_opcintype[i], bkey->sk_subtype,
-											   cur->sk_collation,
+					 bkey->sk_attno, RelationGetRelationName(rel));
-											   procinfo,
+			ScanKeyEntryInitialize(inskey.scankeys + i,
-											   cur->sk_argument);
+								   bkey->sk_flags,
-			}
+								   bkey->sk_attno,
-			else
+								   InvalidStrategy,
-			{
+								   bkey->sk_subtype,
-				RegProcedure cmp_proc;
+								   bkey->sk_collation,
-
+								   cmp_proc,
-				cmp_proc = get_opfamily_proc(rel->rd_opfamily[i],
+								   bkey->sk_argument);
-											 rel->rd_opcintype[i],
-											 cur->sk_subtype,
-											 BTORDER_PROC);
-				if (!RegProcedureIsValid(cmp_proc))
-					elog(ERROR, "missing support function %d(%u,%u) for attribute %d of index \"%s\"",
-						 BTORDER_PROC, rel->rd_opcintype[i], cur->sk_subtype,
-						 cur->sk_attno, RelationGetRelationName(rel));
-				ScanKeyEntryInitialize(inskey.scankeys + i,
-									   cur->sk_flags,
-									   cur->sk_attno,
-									   InvalidStrategy,
-									   cur->sk_subtype,
-									   cur->sk_collation,
-									   cmp_proc,
-									   cur->sk_argument);
-			}
 		}
 	}
 
@@ -1474,6 +1522,8 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 
 	if (!BufferIsValid(so->currPos.buf))
 	{
+		Assert(!so->needPrimScan);
+
 		/*
 		 * We only get here if the index is completely empty. Lock relation
 		 * because nothing finer to lock exists.  Without a buffer lock, it's
@@ -1492,7 +1542,6 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 
 		if (!BufferIsValid(so->currPos.buf))
 		{
-			Assert(!so->needPrimScan);
 			_bt_parallel_done(scan);
 			return false;
 		}

src/backend/access/nbtree/nbtutils.c

@@ -44,7 +44,6 @@ static bool _bt_array_decrement(Relation rel, ScanKey skey, BTArrayKeyInfo *arra
 static bool _bt_array_increment(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
 static bool _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
 											 bool *skip_array_set);
-static void _bt_rewind_nonrequired_arrays(IndexScanDesc scan, ScanDirection dir);
 static bool _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
 										 IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
 										 bool readpagetup, int sktrig, bool *scanBehind);
@@ -52,7 +51,6 @@ static bool _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
 								   IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
 								   int sktrig, bool sktrig_required);
 #ifdef USE_ASSERT_CHECKING
-static bool _bt_verify_arrays_bt_first(IndexScanDesc scan, ScanDirection dir);
 static bool _bt_verify_keys_with_arraykeys(IndexScanDesc scan);
 #endif
 static bool _bt_oppodir_checkkeys(IndexScanDesc scan, ScanDirection dir,
@@ -1034,73 +1032,6 @@ _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
 	return false;
 }
 
-/*
- * _bt_rewind_nonrequired_arrays() -- Rewind SAOP arrays not marked required
- *
- * Called when _bt_advance_array_keys decides to start a new primitive index
- * scan on the basis of the current scan position being before the position
- * that _bt_first is capable of repositioning the scan to by applying an
- * inequality operator required in the opposite-to-scan direction only.
- *
- * Although equality strategy scan keys (for both arrays and non-arrays alike)
- * are either marked required in both directions or in neither direction,
- * there is a sense in which non-required arrays behave like required arrays.
- * With a qual such as "WHERE a IN (100, 200) AND b >= 3 AND c IN (5, 6, 7)",
- * the scan key on "c" is non-required, but nevertheless enables positioning
- * the scan at the first tuple >= "(100, 3, 5)" on the leaf level during the
- * first descent of the tree by _bt_first.  Later on, there could also be a
- * second descent, that places the scan right before tuples >= "(200, 3, 5)".
- * _bt_first must never be allowed to build an insertion scan key whose "c"
- * entry is set to a value other than 5, the "c" array's first element/value.
- * (Actually, it's the first in the current scan direction.  This example uses
- * a forward scan.)
- *
- * Calling here resets the array scan key elements for the scan's non-required
- * arrays.  This is strictly necessary for correctness in a subset of cases
- * involving "required in opposite direction"-triggered primitive index scans.
- * Not all callers are at risk of _bt_first using a non-required array like
- * this, but advancement always resets the arrays when another primitive scan
- * is scheduled, just to keep things simple.  Array advancement even makes
- * sure to reset non-required arrays during scans that have no inequalities.
- * (Advancement still won't call here when there are no inequalities, though
- * that's just because it's all handled indirectly instead.)
- *
- * Note: _bt_verify_arrays_bt_first is called by an assertion to enforce that
- * everybody got this right.
- *
- * Note: In practice almost all SAOP arrays are marked required during
- * preprocessing (if necessary by generating skip arrays).  It is hardly ever
- * truly necessary to call here, but consistently doing so is simpler.
- */
-static void
-_bt_rewind_nonrequired_arrays(IndexScanDesc scan, ScanDirection dir)
-{
-	Relation	rel = scan->indexRelation;
-	BTScanOpaque so = (BTScanOpaque) scan->opaque;
-	int			arrayidx = 0;
-
-	for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
-	{
-		ScanKey		cur = so->keyData + ikey;
-		BTArrayKeyInfo *array = NULL;
-
-		if (!(cur->sk_flags & SK_SEARCHARRAY) ||
-			cur->sk_strategy != BTEqualStrategyNumber)
-			continue;
-
-		array = &so->arrayKeys[arrayidx++];
-		Assert(array->scan_key == ikey);
-
-		if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
-			continue;
-
-		Assert(array->num_elems != -1); /* No non-required skip arrays */
-
-		_bt_array_set_low_or_high(rel, cur, array,
-								  ScanDirectionIsForward(dir));
-	}
-}
-
 /*
  * _bt_tuple_before_array_skeys() -- too early to advance required arrays?
  *
@@ -1380,8 +1311,6 @@ _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir)
 	 */
 	if (so->needPrimScan)
 	{
-		Assert(_bt_verify_arrays_bt_first(scan, dir));
-
 		/*
 		 * Flag was set -- must call _bt_first again, which will reset the
 		 * scan's needPrimScan flag
@@ -2007,14 +1936,7 @@ _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
 	 */
 	else if (has_required_opposite_direction_only && pstate->finaltup &&
 			 unlikely(!_bt_oppodir_checkkeys(scan, dir, pstate->finaltup)))
-	{
-		/*
-		 * Make sure that any SAOP arrays that were not marked required by
-		 * preprocessing are reset to their first element for this direction
-		 */
-		_bt_rewind_nonrequired_arrays(scan, dir);
 		goto new_prim_scan;
-	}
 
 continue_scan:
 
@@ -2045,8 +1967,6 @@ continue_scan:
 		 */
 		so->oppositeDirCheck = has_required_opposite_direction_only;
 
-		_bt_rewind_nonrequired_arrays(scan, dir);
-
 		/*
 		 * skip by setting "look ahead" mechanism's offnum for forwards scans
 		 * (backwards scans check scanBehind flag directly instead)
@@ -2142,48 +2062,6 @@ end_toplevel_scan:
 }
 
 #ifdef USE_ASSERT_CHECKING
-/*
- * Verify that the scan's qual state matches what we expect at the point that
- * _bt_start_prim_scan is about to start a just-scheduled new primitive scan.
- *
- * We enforce a rule against non-required array scan keys: they must start out
- * with whatever element is the first for the scan's current scan direction.
- * See _bt_rewind_nonrequired_arrays comments for an explanation.
- */
-static bool
-_bt_verify_arrays_bt_first(IndexScanDesc scan, ScanDirection dir)
-{
-	BTScanOpaque so = (BTScanOpaque) scan->opaque;
-	int			arrayidx = 0;
-
-	for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
-	{
-		ScanKey		cur = so->keyData + ikey;
-		BTArrayKeyInfo *array = NULL;
-		int			first_elem_dir;
-
-		if (!(cur->sk_flags & SK_SEARCHARRAY) ||
-			cur->sk_strategy != BTEqualStrategyNumber)
-			continue;
-
-		array = &so->arrayKeys[arrayidx++];
-
-		if (((cur->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
-			((cur->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
-			continue;
-
-		if (ScanDirectionIsForward(dir))
-			first_elem_dir = 0;
-		else
-			first_elem_dir = array->num_elems - 1;
-
-		if (array->cur_elem != first_elem_dir)
-			return false;
-	}
-
-	return _bt_verify_keys_with_arraykeys(scan);
-}
-
 /*
  * Verify that the scan's "so->keyData[]" scan keys are in agreement with
  * its array key state
@@ -2194,6 +2072,7 @@ _bt_verify_keys_with_arraykeys(IndexScanDesc scan)
 	BTScanOpaque so = (BTScanOpaque) scan->opaque;
 	int			last_sk_attno = InvalidAttrNumber,
 				arrayidx = 0;
+	bool		nonrequiredseen = false;
 
 	if (!so->qual_ok)
 		return false;
@@ -2217,8 +2096,16 @@ _bt_verify_keys_with_arraykeys(IndexScanDesc scan)
 		if (array->num_elems != -1 &&
 			cur->sk_argument != array->elem_values[array->cur_elem])
 			return false;
-		if (last_sk_attno > cur->sk_attno)
+		if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
-			return false;
+		{
+			if (last_sk_attno > cur->sk_attno)
+				return false;
+			if (nonrequiredseen)
+				return false;
+		}
+		else
+			nonrequiredseen = true;
+
 		last_sk_attno = cur->sk_attno;
 	}
 
@@ -2551,37 +2438,12 @@ _bt_set_startikey(IndexScanDesc scan, BTReadPageState *pstate)
 		if (!(key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
 		{
 			/* Scan key isn't marked required (corner case) */
-			Assert(!(key->sk_flags & SK_ROW_HEADER));
 			break;				/* unsafe */
 		}
 		if (key->sk_flags & SK_ROW_HEADER)
 		{
-			/*
+			/* RowCompare inequalities currently aren't supported */
-			 * RowCompare inequality.
+			break;				/* "unsafe" */
-			 *
-			 * Only the first subkey from a RowCompare can ever be marked
-			 * required (that happens when the row header is marked required).
-			 * There is no simple, general way for us to transitively deduce
-			 * whether or not every tuple on the page satisfies a RowCompare
-			 * key based only on firsttup and lasttup -- so we just give up.
-			 */
-			if (!start_past_saop_eq && !so->skipScan)
-				break;			/* unsafe to go further */
-
-			/*
-			 * We have to be even more careful with RowCompares that come
-			 * after an array: we assume it's unsafe to even bypass the array.
-			 * Calling _bt_start_array_keys to recover the scan's arrays
-			 * following use of forcenonrequired mode isn't compatible with
-			 * _bt_check_rowcompare's continuescan=false behavior with NULL
-			 * row compare members.  _bt_advance_array_keys must not make a
-			 * decision on the basis of a key not being satisfied in the
-			 * opposite-to-scan direction until the scan reaches a leaf page
-			 * where the same key begins to be satisfied in scan direction.
-			 * The _bt_first !used_all_subkeys behavior makes this limitation
-			 * hard to work around some other way.
-			 */
-			return;				/* completely unsafe to set pstate.startikey */
 		}
 		if (key->sk_strategy != BTEqualStrategyNumber)
 		{
@@ -3078,76 +2940,7 @@ _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
 
 		Assert(subkey->sk_flags & SK_ROW_MEMBER);
 
-		if (subkey->sk_attno > tupnatts)
+		/* When a NULL row member is compared, the row never matches */
-		{
-			/*
-			 * This attribute is truncated (must be high key).  The value for
-			 * this attribute in the first non-pivot tuple on the page to the
-			 * right could be any possible value.  Assume that truncated
-			 * attribute passes the qual.
-			 */
-			Assert(BTreeTupleIsPivot(tuple));
-			cmpresult = 0;
-			if (subkey->sk_flags & SK_ROW_END)
-				break;
-			subkey++;
-			continue;
-		}
-
-		datum = index_getattr(tuple,
-							  subkey->sk_attno,
-							  tupdesc,
-							  &isNull);
-
-		if (isNull)
-		{
-			if (forcenonrequired)
-			{
-				/* treating scan's keys as non-required */
-			}
-			else if (subkey->sk_flags & SK_BT_NULLS_FIRST)
-			{
-				/*
-				 * Since NULLs are sorted before non-NULLs, we know we have
-				 * reached the lower limit of the range of values for this
-				 * index attr.  On a backward scan, we can stop if this qual
-				 * is one of the "must match" subset.  We can stop regardless
-				 * of whether the qual is > or <, so long as it's required,
-				 * because it's not possible for any future tuples to pass. On
-				 * a forward scan, however, we must keep going, because we may
-				 * have initially positioned to the start of the index.
-				 * (_bt_advance_array_keys also relies on this behavior during
-				 * forward scans.)
-				 */
-				if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
-					ScanDirectionIsBackward(dir))
-					*continuescan = false;
-			}
-			else
-			{
-				/*
-				 * Since NULLs are sorted after non-NULLs, we know we have
-				 * reached the upper limit of the range of values for this
-				 * index attr.  On a forward scan, we can stop if this qual is
-				 * one of the "must match" subset.  We can stop regardless of
-				 * whether the qual is > or <, so long as it's required,
-				 * because it's not possible for any future tuples to pass. On
-				 * a backward scan, however, we must keep going, because we
-				 * may have initially positioned to the end of the index.
-				 * (_bt_advance_array_keys also relies on this behavior during
-				 * backward scans.)
-				 */
-				if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
-					ScanDirectionIsForward(dir))
-					*continuescan = false;
-			}
-
-			/*
-			 * In any case, this indextuple doesn't match the qual.
-			 */
-			return false;
-		}
-
 		if (subkey->sk_flags & SK_ISNULL)
 		{
 			/*
@@ -3172,6 +2965,114 @@ _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
 			return false;
 		}
 
+		if (subkey->sk_attno > tupnatts)
+		{
+			/*
+			 * This attribute is truncated (must be high key).  The value for
+			 * this attribute in the first non-pivot tuple on the page to the
+			 * right could be any possible value.  Assume that truncated
+			 * attribute passes the qual.
+			 */
+			Assert(BTreeTupleIsPivot(tuple));
+			return true;
+		}
+
+		datum = index_getattr(tuple,
+							  subkey->sk_attno,
+							  tupdesc,
+							  &isNull);
+
+		if (isNull)
+		{
+			int			reqflags;
+
+			if (forcenonrequired)
+			{
+				/* treating scan's keys as non-required */
+			}
+			else if (subkey->sk_flags & SK_BT_NULLS_FIRST)
+			{
+				/*
+				 * Since NULLs are sorted before non-NULLs, we know we have
+				 * reached the lower limit of the range of values for this
+				 * index attr.  On a backward scan, we can stop if this qual
+				 * is one of the "must match" subset.  However, on a forwards
+				 * scan, we must keep going, because we may have initially
+				 * positioned to the start of the index.
+				 *
+				 * All required NULLS FIRST > row members can use NULL tuple
+				 * values to end backwards scans, just like with other values.
+				 * A qual "WHERE (a, b, c) > (9, 42, 'foo')" can terminate a
+				 * backwards scan upon reaching the index's rightmost "a = 9"
+				 * tuple whose "b" column contains a NULL (if not sooner).
+				 * Since "b" is NULLS FIRST, we can treat its NULLs as "<" 42.
+				 */
+				reqflags = SK_BT_REQBKWD;
+
+				/*
+				 * When a most significant required NULLS FIRST < row compare
+				 * member sees NULL tuple values during a backwards scan, it
+				 * signals the end of matches for the whole row compare/scan.
+				 * A qual "WHERE (a, b, c) < (9, 42, 'foo')" will terminate a
+				 * backwards scan upon reaching the rightmost tuple whose "a"
+				 * column has a NULL.  The "a" NULL value is "<" 9, and yet
+				 * our < row compare will still end the scan.  (This isn't
+				 * safe with later/lower-order row members.  Notice that it
+				 * can only happen with an "a" NULL some time after the scan
+				 * completely stops needing to use its "b" and "c" members.)
+				 */
+				if (subkey == (ScanKey) DatumGetPointer(skey->sk_argument))
+					reqflags |= SK_BT_REQFWD;	/* safe, first row member */
+
+				if ((subkey->sk_flags & reqflags) &&
+					ScanDirectionIsBackward(dir))
+					*continuescan = false;
+			}
+			else
+			{
+				/*
+				 * Since NULLs are sorted after non-NULLs, we know we have
+				 * reached the upper limit of the range of values for this
+				 * index attr.  On a forward scan, we can stop if this qual is
+				 * one of the "must match" subset.  However, on a backward
+				 * scan, we must keep going, because we may have initially
+				 * positioned to the end of the index.
+				 *
+				 * All required NULLS LAST < row members can use NULL tuple
+				 * values to end forwards scans, just like with other values.
+				 * A qual "WHERE (a, b, c) < (9, 42, 'foo')" can terminate a
+				 * forwards scan upon reaching the index's leftmost "a = 9"
+				 * tuple whose "b" column contains a NULL (if not sooner).
+				 * Since "b" is NULLS LAST, we can treat its NULLs as ">" 42.
+				 */
+				reqflags = SK_BT_REQFWD;
+
+				/*
+				 * When a most significant required NULLS LAST > row compare
+				 * member sees NULL tuple values during a forwards scan, it
+				 * signals the end of matches for the whole row compare/scan.
+				 * A qual "WHERE (a, b, c) > (9, 42, 'foo')" will terminate a
+				 * forwards scan upon reaching the leftmost tuple whose "a"
+				 * column has a NULL.  The "a" NULL value is ">" 9, and yet
+				 * our > row compare will end the scan.  (This isn't safe with
+				 * later/lower-order row members.  Notice that it can only
+				 * happen with an "a" NULL some time after the scan completely
+				 * stops needing to use its "b" and "c" members.)
+				 */
+				if (subkey == (ScanKey) DatumGetPointer(skey->sk_argument))
+					reqflags |= SK_BT_REQBKWD;	/* safe, first row member */
+
+				if ((subkey->sk_flags & reqflags) &&
+					ScanDirectionIsForward(dir))
+					*continuescan = false;
+			}
+
+			/*
+			 * In any case, this indextuple doesn't match the qual.
+			 */
+			return false;
+		}
+
 		/* Perform the test --- three-way comparison not bool operator */
 		cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
 													subkey->sk_collation,

src/backend/parser/gram.y

@@ -2668,6 +2668,12 @@ alter_table_cmd:
 						c->alterDeferrability = true;
 					if ($4 & CAS_NO_INHERIT)
 						c->alterInheritability = true;
+					/* handle unsupported case with specific error message */
+					if ($4 & CAS_NOT_VALID)
+						ereport(ERROR,
+								errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+								errmsg("constraints cannot be altered to be NOT VALID"),
+								parser_errposition(@4));
 					processCASbits($4, @4, "FOREIGN KEY",
 									&c->deferrable,
 									&c->initdeferred,

src/bin/pg_verifybackup/t/008_untar.pl

@@ -16,6 +16,22 @@ my $primary = PostgreSQL::Test::Cluster->new('primary');
 $primary->init(allows_streaming => 1);
 $primary->start;
 
+# Create file with some random data and an arbitrary size, useful to check
+# the solidity of the compression and decompression logic.  The size of the
+# file is chosen to be around 640kB.  This has proven to be large enough to
+# detect some issues related to LZ4, and low enough to not impact the runtime
+# of the test significantly.
+my $junk_data = $primary->safe_psql(
+	'postgres', qq(
+		SELECT string_agg(encode(sha256(i::bytea), 'hex'), '')
+		FROM generate_series(1, 10240) s(i);));
+my $data_dir = $primary->data_dir;
+my $junk_file = "$data_dir/junk";
+open my $jf, '>', $junk_file
+  or die "Could not create junk file: $!";
+print $jf $junk_data;
+close $jf;
+
 # Create a tablespace directory.
 my $source_ts_path = PostgreSQL::Test::Utils::tempdir_short();
 
@@ -52,6 +68,12 @@ my @test_configuration = (
 		'backup_archive' => [ 'base.tar.lz4', "$tsoid.tar.lz4" ],
 		'enabled' => check_pg_config("#define USE_LZ4 1")
 	},
+	{
+		'compression_method' => 'lz4',
+		'backup_flags' => [ '--compress', 'server-lz4:5' ],
+		'backup_archive' => [ 'base.tar.lz4', "$tsoid.tar.lz4" ],
+		'enabled' => check_pg_config("#define USE_LZ4 1")
+	},
 	{
 		'compression_method' => 'zstd',
 		'backup_flags' => [ '--compress', 'server-zstd' ],

src/bin/pg_verifybackup/t/010_client_untar.pl

@@ -15,6 +15,22 @@ my $primary = PostgreSQL::Test::Cluster->new('primary');
 $primary->init(allows_streaming => 1);
 $primary->start;
 
+# Create file with some random data and an arbitrary size, useful to check
+# the solidity of the compression and decompression logic.  The size of the
+# file is chosen to be around 640kB.  This has proven to be large enough to
+# detect some issues related to LZ4, and low enough to not impact the runtime
+# of the test significantly.
+my $junk_data = $primary->safe_psql(
+	'postgres', qq(
+		SELECT string_agg(encode(sha256(i::bytea), 'hex'), '')
+		FROM generate_series(1, 10240) s(i);));
+my $data_dir = $primary->data_dir;
+my $junk_file = "$data_dir/junk";
+open my $jf, '>', $junk_file
+  or die "Could not create junk file: $!";
+print $jf $junk_data;
+close $jf;
+
 my $backup_path = $primary->backup_dir . '/client-backup';
 my $extract_path = $primary->backup_dir . '/extracted-backup';
 
@@ -37,6 +53,12 @@ my @test_configuration = (
 		'backup_archive' => 'base.tar.lz4',
 		'enabled' => check_pg_config("#define USE_LZ4 1")
 	},
+	{
+		'compression_method' => 'lz4',
+		'backup_flags' => [ '--compress', 'client-lz4:1' ],
+		'backup_archive' => 'base.tar.lz4',
+		'enabled' => check_pg_config("#define USE_LZ4 1")
+	},
 	{
 		'compression_method' => 'zstd',
 		'backup_flags' => [ '--compress', 'client-zstd:5' ],

src/fe_utils/astreamer_lz4.c

@@ -322,9 +322,9 @@ astreamer_lz4_decompressor_content(astreamer *streamer,
 
 	mystreamer = (astreamer_lz4_frame *) streamer;
 	next_in = (uint8 *) data;
-	next_out = (uint8 *) mystreamer->base.bbs_buffer.data;
+	next_out = (uint8 *) mystreamer->base.bbs_buffer.data + mystreamer->bytes_written;
 	avail_in = len;
-	avail_out = mystreamer->base.bbs_buffer.maxlen;
+	avail_out = mystreamer->base.bbs_buffer.maxlen - mystreamer->bytes_written;
 
 	while (avail_in > 0)
 	{

src/include/port/pg_crc32c.h

@@ -72,7 +72,7 @@ pg_comp_crc32c_dispatch(pg_crc32c crc, const void *data, size_t len)
 {
 	if (__builtin_constant_p(len) && len < 32)
 	{
-		const unsigned char *p = data;
+		const unsigned char *p = (const unsigned char *) data;
 
 		/*
 		 * For small constant inputs, inline the computation to avoid a

src/interfaces/libpq/fe-cancel.c

@@ -137,6 +137,7 @@ PQcancelCreate(PGconn *conn)
 		goto oom_error;
 
 	originalHost = conn->connhost[conn->whichhost];
+	cancelConn->connhost[0].type = originalHost.type;
 	if (originalHost.host)
 	{
 		cancelConn->connhost[0].host = strdup(originalHost.host);

src/test/regress/expected/btree_index.out

@@ -195,54 +195,123 @@ ORDER BY proname DESC, proargtypes DESC, pronamespace DESC LIMIT 1;
 (1 row)
 
 --
--- Add coverage for RowCompare quals whose rhs row has a NULL that ends scan
+-- Forwards scan RowCompare qual whose row arg has a NULL that affects our
+-- initial positioning strategy
 --
 explain (costs off)
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) < ('abs', NULL)
+   WHERE (proname, proargtypes) >= ('abs', NULL) AND proname <= 'abs'
 ORDER BY proname, proargtypes, pronamespace;
-                                                 QUERY PLAN                                                  
+                                                  QUERY PLAN                                                   
--------------------------------------------------------------------------------------------------------------
+---------------------------------------------------------------------------------------------------------------
  Index Only Scan using pg_proc_proname_args_nsp_index on pg_proc
-   Index Cond: ((ROW(proname, proargtypes) < ROW('abs'::name, NULL::oidvector)) AND (proname = 'abs'::name))
+   Index Cond: ((ROW(proname, proargtypes) >= ROW('abs'::name, NULL::oidvector)) AND (proname <= 'abs'::name))
 (2 rows)
 
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) < ('abs', NULL)
+   WHERE (proname, proargtypes) >= ('abs', NULL) AND proname <= 'abs'
 ORDER BY proname, proargtypes, pronamespace;
  proname | proargtypes | pronamespace 
 ---------+-------------+--------------
 (0 rows)
 
 --
--- Add coverage for backwards scan RowCompare quals whose rhs row has a NULL
+-- Forwards scan RowCompare quals whose row arg has a NULL that ends scan
--- that ends scan
 --
 explain (costs off)
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) > ('abs', NULL)
+   WHERE proname >= 'abs' AND (proname, proargtypes) < ('abs', NULL)
-ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
+ORDER BY proname, proargtypes, pronamespace;
-                                                 QUERY PLAN                                                  
+                                                  QUERY PLAN                                                  
--------------------------------------------------------------------------------------------------------------
+--------------------------------------------------------------------------------------------------------------
- Index Only Scan Backward using pg_proc_proname_args_nsp_index on pg_proc
+ Index Only Scan using pg_proc_proname_args_nsp_index on pg_proc
-   Index Cond: ((ROW(proname, proargtypes) > ROW('abs'::name, NULL::oidvector)) AND (proname = 'abs'::name))
+   Index Cond: ((proname >= 'abs'::name) AND (ROW(proname, proargtypes) < ROW('abs'::name, NULL::oidvector)))
 (2 rows)
 
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) > ('abs', NULL)
+   WHERE proname >= 'abs' AND (proname, proargtypes) < ('abs', NULL)
+ORDER BY proname, proargtypes, pronamespace;
+ proname | proargtypes | pronamespace 
+---------+-------------+--------------
+(0 rows)
+
+--
+-- Backwards scan RowCompare qual whose row arg has a NULL that affects our
+-- initial positioning strategy
+--
+explain (costs off)
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname >= 'abs' AND (proname, proargtypes) <= ('abs', NULL)
+ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
+                                                  QUERY PLAN                                                   
+---------------------------------------------------------------------------------------------------------------
+ Index Only Scan Backward using pg_proc_proname_args_nsp_index on pg_proc
+   Index Cond: ((proname >= 'abs'::name) AND (ROW(proname, proargtypes) <= ROW('abs'::name, NULL::oidvector)))
+(2 rows)
+
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname >= 'abs' AND (proname, proargtypes) <= ('abs', NULL)
 ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
  proname | proargtypes | pronamespace 
 ---------+-------------+--------------
 (0 rows)
 
 --
--- Add coverage for recheck of > key following array advancement on previous
+-- Backwards scan RowCompare qual whose row arg has a NULL that ends scan
--- (left sibling) page that used a high key whose attribute value corresponding
+--
--- to the > key was -inf (due to being truncated when the high key was created).
+explain (costs off)
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE (proname, proargtypes) > ('abs', NULL) AND proname <= 'abs'
+ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
+                                                  QUERY PLAN                                                  
+--------------------------------------------------------------------------------------------------------------
+ Index Only Scan Backward using pg_proc_proname_args_nsp_index on pg_proc
+   Index Cond: ((ROW(proname, proargtypes) > ROW('abs'::name, NULL::oidvector)) AND (proname <= 'abs'::name))
+(2 rows)
+
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE (proname, proargtypes) > ('abs', NULL) AND proname <= 'abs'
+ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
+ proname | proargtypes | pronamespace 
+---------+-------------+--------------
+(0 rows)
+
+-- Makes B-Tree preprocessing deal with unmarking redundant keys that were
+-- initially marked required (test case relies on current row compare
+-- preprocessing limitations)
+explain (costs off)
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname = 'zzzzzz' AND (proname, proargtypes) > ('abs', NULL)
+   AND pronamespace IN (1, 2, 3) AND proargtypes IN ('26 23', '5077')
+ORDER BY proname, proargtypes, pronamespace;
+                                                                                                     QUERY PLAN                                                                                                     
+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
+ Index Only Scan using pg_proc_proname_args_nsp_index on pg_proc
+   Index Cond: ((ROW(proname, proargtypes) > ROW('abs'::name, NULL::oidvector)) AND (proname = 'zzzzzz'::name) AND (proargtypes = ANY ('{"26 23",5077}'::oidvector[])) AND (pronamespace = ANY ('{1,2,3}'::oid[])))
+(2 rows)
+
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname = 'zzzzzz' AND (proname, proargtypes) > ('abs', NULL)
+   AND pronamespace IN (1, 2, 3) AND proargtypes IN ('26 23', '5077')
+ORDER BY proname, proargtypes, pronamespace;
+ proname | proargtypes | pronamespace 
+---------+-------------+--------------
+(0 rows)
+
+--
+-- Performs a recheck of > key following array advancement on previous (left
+-- sibling) page that used a high key whose attribute value corresponding to
+-- the > key was -inf (due to being truncated when the high key was created).
 --
 -- XXX This relies on the assumption that tenk1_thous_tenthous has a truncated
 -- high key "(183, -inf)" on the first page that we'll scan.  The test will only

src/test/regress/expected/constraints.out

@@ -748,6 +748,11 @@ ALTER TABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key ENFORCED;
 ERROR:  cannot alter enforceability of constraint "unique_tbl_i_key" of relation "unique_tbl"
 ALTER TABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key NOT ENFORCED;
 ERROR:  cannot alter enforceability of constraint "unique_tbl_i_key" of relation "unique_tbl"
+-- can't make an existing constraint NOT VALID
+ALTER TABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key NOT VALID;
+ERROR:  constraints cannot be altered to be NOT VALID
+LINE 1: ...ABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key NOT VALID;
+                                                             ^
 DROP TABLE unique_tbl;
 --
 -- EXCLUDE constraints

src/test/regress/expected/foreign_key.out

@@ -1359,7 +1359,7 @@ LINE 1: ...e ALTER CONSTRAINT fktable_fk_fkey NOT DEFERRABLE INITIALLY ...
 ALTER TABLE fktable ALTER CONSTRAINT fktable_fk_fkey NO INHERIT;
 ERROR:  constraint "fktable_fk_fkey" of relation "fktable" is not a not-null constraint
 ALTER TABLE fktable ALTER CONSTRAINT fktable_fk_fkey NOT VALID;
-ERROR:  FOREIGN KEY constraints cannot be marked NOT VALID
+ERROR:  constraints cannot be altered to be NOT VALID
 LINE 1: ...ER TABLE fktable ALTER CONSTRAINT fktable_fk_fkey NOT VALID;
                                                              ^
 ALTER TABLE fktable ALTER CONSTRAINT fktable_fk_fkey ENFORCED NOT ENFORCED;

src/test/regress/sql/btree_index.sql

@@ -143,38 +143,83 @@ SELECT proname, proargtypes, pronamespace
 ORDER BY proname DESC, proargtypes DESC, pronamespace DESC LIMIT 1;
 
 --
--- Add coverage for RowCompare quals whose rhs row has a NULL that ends scan
+-- Forwards scan RowCompare qual whose row arg has a NULL that affects our
+-- initial positioning strategy
 --
 explain (costs off)
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) < ('abs', NULL)
+   WHERE (proname, proargtypes) >= ('abs', NULL) AND proname <= 'abs'
 ORDER BY proname, proargtypes, pronamespace;
 
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) < ('abs', NULL)
+   WHERE (proname, proargtypes) >= ('abs', NULL) AND proname <= 'abs'
 ORDER BY proname, proargtypes, pronamespace;
 
 --
--- Add coverage for backwards scan RowCompare quals whose rhs row has a NULL
+-- Forwards scan RowCompare quals whose row arg has a NULL that ends scan
--- that ends scan
 --
 explain (costs off)
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) > ('abs', NULL)
+   WHERE proname >= 'abs' AND (proname, proargtypes) < ('abs', NULL)
+ORDER BY proname, proargtypes, pronamespace;
+
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname >= 'abs' AND (proname, proargtypes) < ('abs', NULL)
+ORDER BY proname, proargtypes, pronamespace;
+
+--
+-- Backwards scan RowCompare qual whose row arg has a NULL that affects our
+-- initial positioning strategy
+--
+explain (costs off)
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname >= 'abs' AND (proname, proargtypes) <= ('abs', NULL)
 ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
 
 SELECT proname, proargtypes, pronamespace
    FROM pg_proc
-   WHERE proname = 'abs' AND (proname, proargtypes) > ('abs', NULL)
+   WHERE proname >= 'abs' AND (proname, proargtypes) <= ('abs', NULL)
 ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
 
 --
--- Add coverage for recheck of > key following array advancement on previous
+-- Backwards scan RowCompare qual whose row arg has a NULL that ends scan
--- (left sibling) page that used a high key whose attribute value corresponding
+--
--- to the > key was -inf (due to being truncated when the high key was created).
+explain (costs off)
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE (proname, proargtypes) > ('abs', NULL) AND proname <= 'abs'
+ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
+
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE (proname, proargtypes) > ('abs', NULL) AND proname <= 'abs'
+ORDER BY proname DESC, proargtypes DESC, pronamespace DESC;
+
+-- Makes B-Tree preprocessing deal with unmarking redundant keys that were
+-- initially marked required (test case relies on current row compare
+-- preprocessing limitations)
+explain (costs off)
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname = 'zzzzzz' AND (proname, proargtypes) > ('abs', NULL)
+   AND pronamespace IN (1, 2, 3) AND proargtypes IN ('26 23', '5077')
+ORDER BY proname, proargtypes, pronamespace;
+
+SELECT proname, proargtypes, pronamespace
+   FROM pg_proc
+   WHERE proname = 'zzzzzz' AND (proname, proargtypes) > ('abs', NULL)
+   AND pronamespace IN (1, 2, 3) AND proargtypes IN ('26 23', '5077')
+ORDER BY proname, proargtypes, pronamespace;
+
+--
+-- Performs a recheck of > key following array advancement on previous (left
+-- sibling) page that used a high key whose attribute value corresponding to
+-- the > key was -inf (due to being truncated when the high key was created).
 --
 -- XXX This relies on the assumption that tenk1_thous_tenthous has a truncated
 -- high key "(183, -inf)" on the first page that we'll scan.  The test will only

src/test/regress/sql/constraints.sql

@@ -537,6 +537,9 @@ CREATE TABLE UNIQUE_NOTEN_TBL(i int UNIQUE NOT ENFORCED);
 ALTER TABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key ENFORCED;
 ALTER TABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key NOT ENFORCED;
 
+-- can't make an existing constraint NOT VALID
+ALTER TABLE unique_tbl ALTER CONSTRAINT unique_tbl_i_key NOT VALID;
+
 DROP TABLE unique_tbl;
 
 --