Document nbtree row comparison design.

Add comments explaining when and where it is safe for nbtree to treat
row compare keys as if they were simple scalar inequality keys on the
row's most significant column.  This is particularly important within
_bt_advance_array_keys, which deals with required inequality keys in a
general and uniform way, without any special handling for row compares.

Also spell out the implications of _bt_check_rowcompare's approach of
_conditionally_ evaluating lower-order row compare subkeys, particularly
when one of its lower-order subkeys might see NULL index tuple values
(these may or may not affect whether the qual as a whole is satisfied).
The behavior in this area isn't particularly intuitive, so these issues
seem worth going into.

In passing, add a few more defensive/documenting row comparison related
assertions to _bt_first and _bt_check_rowcompare.

Follow-up to commits bd3f59f and ec98602.

Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Victor Yegorov <vyegorov@gmail.com>
Reviewed-By: Chao Li <li.evan.chao@gmail.com>
Discussion: https://postgr.es/m/CAH2-Wznwkak_K7pcAdv9uH8ZfNo8QO7+tHXOaCUddMeTfaCCFw@mail.gmail.com
Backpatch-through: 18

Author: petergeoghegan

src/backend/access/nbtree/nbtsearch.c

@@ -1290,6 +1290,8 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 			 * our row compare header key must be the final startKeys[] entry.
 			 */
 			Assert(subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD));
+			Assert(subkey->sk_strategy == bkey->sk_strategy);
+			Assert(subkey->sk_strategy == strat_total);
 			Assert(i == keysz - 1);
 
 			/*
@@ -1346,9 +1348,9 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 				Assert(subkey->sk_strategy == bkey->sk_strategy);
 				Assert(keysz < INDEX_MAX_KEYS);
 
-				memcpy(inskey.scankeys + keysz, subkey,
-					   sizeof(ScanKeyData));
+				memcpy(inskey.scankeys + keysz, subkey, sizeof(ScanKeyData));
 				keysz++;
+
 				if (subkey->sk_flags & SK_ROW_END)
 					break;
 			}
@@ -1380,7 +1382,7 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 				}
 			}
 
-			/* done adding to inskey (row comparison keys always come last) */
+			/* Done (row compare header key is always last startKeys[] key) */
 			break;
 		}
 

src/backend/access/nbtree/nbtutils.c

@@ -59,7 +59,7 @@ static bool _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
 							  IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
 							  bool advancenonrequired, bool forcenonrequired,
 							  bool *continuescan, int *ikey);
-static bool _bt_check_rowcompare(ScanKey skey,
+static bool _bt_check_rowcompare(ScanKey header,
 								 IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
 								 ScanDirection dir, bool forcenonrequired, bool *continuescan);
 static void _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
@@ -2911,19 +2911,64 @@ _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
  * it's not possible for any future tuples in the current scan direction
  * to pass the qual.
  *
- * This is a subroutine for _bt_checkkeys/_bt_check_compare.
+ * This is a subroutine for _bt_checkkeys/_bt_check_compare.  Caller passes us
+ * a row compare header key taken from so->keyData[].
+ *
+ * Row value comparisons can be described in terms of logical expansions that
+ * use only scalar operators.  Consider the following example row comparison:
+ *
+ * "(a, b, c) > (7, 'bar', 62)"
+ *
+ * This can be evaluated as:
+ *
+ * "(a = 7 AND b = 'bar' AND c > 62) OR (a = 7 AND b > 'bar') OR (a > 7)".
+ *
+ * Notice that this condition is satisfied by _all_ rows that satisfy "a > 7",
+ * and by a subset of all rows that satisfy "a >= 7" (possibly all such rows).
+ * It _can't_ be satisfied by other rows (where "a < 7" or where "a IS NULL").
+ * A row comparison header key can therefore often be treated as if it was a
+ * simple scalar inequality on the row compare's most significant column.
+ * (For example, _bt_advance_array_keys and most preprocessing routines treat
+ * row compares like any other same-strategy inequality on the same column.)
+ *
+ * Things get more complicated for our row compare given a row where "a = 7".
+ * Note that a row compare isn't necessarily satisfied by _every_ tuple that
+ * appears between the first and last satisfied tuple returned by the scan,
+ * due to the way that its lower-order subkeys are only conditionally applied.
+ * A forwards scan that uses our example qual might initially return a tuple
+ * "(a, b, c) = (7, 'zebra', 54)".  But it won't subsequently return a tuple
+ * "(a, b, c) = (7, NULL, 1)" located to the right of the first matching tuple
+ * (assume that "b" was declared NULLS LAST here).  The scan will only return
+ * additional matches upon reaching tuples where "a > 7".  If you rereview our
+ * example row comparison's logical expansion, you'll understand why this is.
+ * (Here we assume that all subkeys could be marked required, guaranteeing
+ * that row comparison order matches index order.  This is the common case.)
+ *
+ * Note that a row comparison header key behaves _exactly_ the same as a
+ * similar scalar inequality key on the row's most significant column once the
+ * scan reaches the point where it no longer needs to evaluate lower-order
+ * subkeys (or before the point where it starts needing to evaluate them).
+ * For example, once a forwards scan that uses our example qual reaches the
+ * first tuple "a > 7", we'll behave in just the same way as our caller would
+ * behave with a similar scalar inequality "a > 7" for the remainder of the
+ * scan (assuming that the scan never changes direction/never goes backwards).
+ * We'll even set continuescan=false according to exactly the same rules as
+ * the ones our caller applies with simple scalar inequalities, including the
+ * rules it applies when NULL tuple values don't satisfy an inequality qual.
  */
 static bool
-_bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
+_bt_check_rowcompare(ScanKey header, IndexTuple tuple, int tupnatts,
 					 TupleDesc tupdesc, ScanDirection dir,
 					 bool forcenonrequired, bool *continuescan)
 {
-	ScanKey		subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
+	ScanKey		subkey = (ScanKey) DatumGetPointer(header->sk_argument);
 	int32		cmpresult = 0;
 	bool		result;
 
 	/* First subkey should be same as the header says */
-	Assert(subkey->sk_attno == skey->sk_attno);
+	Assert(header->sk_flags & SK_ROW_HEADER);
+	Assert(subkey->sk_attno == header->sk_attno);
+	Assert(subkey->sk_strategy == header->sk_strategy);
 
 	/* Loop over columns of the row condition */
 	for (;;)
@@ -2943,7 +2988,7 @@ _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
 			 * columns are required for the scan direction, we can stop the
 			 * scan, because there can't be another tuple that will succeed.
 			 */
-			Assert(subkey != (ScanKey) DatumGetPointer(skey->sk_argument));
+			Assert(subkey != (ScanKey) DatumGetPointer(header->sk_argument));
 			subkey--;
 			if (forcenonrequired)
 			{
@@ -3014,7 +3059,7 @@ _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
 				 * 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))
+				if (subkey == (ScanKey) DatumGetPointer(header->sk_argument))
 					reqflags |= SK_BT_REQFWD;	/* safe, first row member */
 
 				if ((subkey->sk_flags & reqflags) &&
@@ -3052,7 +3097,7 @@ _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
 				 * 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))
+				if (subkey == (ScanKey) DatumGetPointer(header->sk_argument))
 					reqflags |= SK_BT_REQBKWD;	/* safe, first row member */
 
 				if ((subkey->sk_flags & reqflags) &&