Allow the built-in ordered-set aggregates to share transition state.

The built-in OSAs all share the same transition function, so they can
share transition state as long as the final functions cooperate to not
do the sort step more than once.  To avoid running the tuplesort object
in randomAccess mode unnecessarily, add a bit of infrastructure to
nodeAgg.c to let the aggregate functions find out whether the transition
state is actually being shared or not.

This doesn't work for the hypothetical aggregates, since those inject
a hypothetical row that isn't traceable to the shared input state.
So they remain marked aggfinalmodify = 'w'.

Discussion: https://postgr.es/m/CAB4ELO5RZhOamuT9Xsf72ozbenDLLXZKSk07FiSVsuJNZB861A@mail.gmail.com

src/backend/executor/nodeAgg.c

@@ -254,6 +254,11 @@ typedef struct AggStatePerTransData
 	 */
 	Aggref	   *aggref;
 
+	/*
+	 * Is this state value actually being shared by more than one Aggref?
+	 */
+	bool		aggshared;
+
 	/*
 	 * Nominal number of arguments for aggregate function.  For plain aggs,
 	 * this excludes any ORDER BY expressions.  For ordered-set aggs, this
@@ -3360,9 +3365,10 @@ ExecInitAgg(Agg *node, EState *estate, int eflags)
 		{
 			/*
 			 * Existing compatible trans found, so just point the 'peragg' to
-			 * the same per-trans struct.
+			 * the same per-trans struct, and mark the trans state as shared.
 			 */
 			pertrans = &pertransstates[existing_transno];
+			pertrans->aggshared = true;
 			peragg->transno = existing_transno;
 		}
 		else
@@ -3512,6 +3518,7 @@ build_pertrans_for_aggref(AggStatePerTrans pertrans,
 
 	/* Begin filling in the pertrans data */
 	pertrans->aggref = aggref;
+	pertrans->aggshared = false;
 	pertrans->aggCollation = aggref->inputcollid;
 	pertrans->transfn_oid = aggtransfn;
 	pertrans->serialfn_oid = aggserialfn;
@@ -4161,17 +4168,18 @@ AggGetAggref(FunctionCallInfo fcinfo)
 {
 	if (fcinfo->context && IsA(fcinfo->context, AggState))
 	{
+		AggState   *aggstate = (AggState *) fcinfo->context;
 		AggStatePerAgg curperagg;
 		AggStatePerTrans curpertrans;
 
 		/* check curperagg (valid when in a final function) */
-		curperagg = ((AggState *) fcinfo->context)->curperagg;
+		curperagg = aggstate->curperagg;
 
 		if (curperagg)
 			return curperagg->aggref;
 
 		/* check curpertrans (valid when in a transition function) */
-		curpertrans = ((AggState *) fcinfo->context)->curpertrans;
+		curpertrans = aggstate->curpertrans;
 
 		if (curpertrans)
 			return curpertrans->aggref;
@@ -4201,6 +4209,44 @@ AggGetTempMemoryContext(FunctionCallInfo fcinfo)
 	return NULL;
 }
 
+/*
+ * AggStateIsShared - find out whether transition state is shared
+ *
+ * If the function is being called as an aggregate support function,
+ * return TRUE if the aggregate's transition state is shared across
+ * multiple aggregates, FALSE if it is not.
+ *
+ * Returns TRUE if not called as an aggregate support function.
+ * This is intended as a conservative answer, ie "no you'd better not
+ * scribble on your input".  In particular, will return TRUE if the
+ * aggregate is being used as a window function, which is a scenario
+ * in which changing the transition state is a bad idea.  We might
+ * want to refine the behavior for the window case in future.
+ */
+bool
+AggStateIsShared(FunctionCallInfo fcinfo)
+{
+	if (fcinfo->context && IsA(fcinfo->context, AggState))
+	{
+		AggState   *aggstate = (AggState *) fcinfo->context;
+		AggStatePerAgg curperagg;
+		AggStatePerTrans curpertrans;
+
+		/* check curperagg (valid when in a final function) */
+		curperagg = aggstate->curperagg;
+
+		if (curperagg)
+			return aggstate->pertrans[curperagg->transno].aggshared;
+
+		/* check curpertrans (valid when in a transition function) */
+		curpertrans = aggstate->curpertrans;
+
+		if (curpertrans)
+			return curpertrans->aggshared;
+	}
+	return true;
+}
+
 /*
  * AggRegisterCallback - register a cleanup callback for an aggregate
  *

src/include/catalog/catversion.h

@@ -53,6 +53,6 @@
  */
 
 /*							yyyymmddN */
-#define CATALOG_VERSION_NO	201710141
+#define CATALOG_VERSION_NO	201710161
 
 #endif

src/include/catalog/pg_aggregate.h

@@ -318,13 +318,13 @@ DATA(insert ( 3267	n 0 jsonb_agg_transfn	jsonb_agg_finalfn				-	-	-	-				-				-
 DATA(insert ( 3270	n 0 jsonb_object_agg_transfn jsonb_object_agg_finalfn	-	-	-	-				-				-			f f r r 0	2281	0	0		0	_null_ _null_ ));
 
 /* ordered-set and hypothetical-set aggregates */
-DATA(insert ( 3972	o 1 ordered_set_transition			percentile_disc_final					-	-	-	-		-		-		t f w w 0	2281	0	0		0	_null_ _null_ ));
-DATA(insert ( 3974	o 1 ordered_set_transition			percentile_cont_float8_final			-	-	-	-		-		-		f f w w 0	2281	0	0		0	_null_ _null_ ));
-DATA(insert ( 3976	o 1 ordered_set_transition			percentile_cont_interval_final			-	-	-	-		-		-		f f w w 0	2281	0	0		0	_null_ _null_ ));
-DATA(insert ( 3978	o 1 ordered_set_transition			percentile_disc_multi_final				-	-	-	-		-		-		t f w w 0	2281	0	0		0	_null_ _null_ ));
-DATA(insert ( 3980	o 1 ordered_set_transition			percentile_cont_float8_multi_final		-	-	-	-		-		-		f f w w 0	2281	0	0		0	_null_ _null_ ));
-DATA(insert ( 3982	o 1 ordered_set_transition			percentile_cont_interval_multi_final	-	-	-	-		-		-		f f w w 0	2281	0	0		0	_null_ _null_ ));
-DATA(insert ( 3984	o 0 ordered_set_transition			mode_final								-	-	-	-		-		-		t f w w 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3972	o 1 ordered_set_transition			percentile_disc_final					-	-	-	-		-		-		t f s s 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3974	o 1 ordered_set_transition			percentile_cont_float8_final			-	-	-	-		-		-		f f s s 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3976	o 1 ordered_set_transition			percentile_cont_interval_final			-	-	-	-		-		-		f f s s 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3978	o 1 ordered_set_transition			percentile_disc_multi_final				-	-	-	-		-		-		t f s s 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3980	o 1 ordered_set_transition			percentile_cont_float8_multi_final		-	-	-	-		-		-		f f s s 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3982	o 1 ordered_set_transition			percentile_cont_interval_multi_final	-	-	-	-		-		-		f f s s 0	2281	0	0		0	_null_ _null_ ));
+DATA(insert ( 3984	o 0 ordered_set_transition			mode_final								-	-	-	-		-		-		t f s s 0	2281	0	0		0	_null_ _null_ ));
 DATA(insert ( 3986	h 1 ordered_set_transition_multi	rank_final								-	-	-	-		-		-		t f w w 0	2281	0	0		0	_null_ _null_ ));
 DATA(insert ( 3988	h 1 ordered_set_transition_multi	percent_rank_final						-	-	-	-		-		-		t f w w 0	2281	0	0		0	_null_ _null_ ));
 DATA(insert ( 3990	h 1 ordered_set_transition_multi	cume_dist_final							-	-	-	-		-		-		t f w w 0	2281	0	0		0	_null_ _null_ ));

src/include/fmgr.h

@@ -698,6 +698,7 @@ extern int AggCheckCallContext(FunctionCallInfo fcinfo,
 					MemoryContext *aggcontext);
 extern fmAggrefPtr AggGetAggref(FunctionCallInfo fcinfo);
 extern MemoryContext AggGetTempMemoryContext(FunctionCallInfo fcinfo);
+extern bool AggStateIsShared(FunctionCallInfo fcinfo);
 extern void AggRegisterCallback(FunctionCallInfo fcinfo,
 					fmExprContextCallbackFunction func,
 					Datum arg);

src/test/regress/expected/aggregates.out

@@ -1866,7 +1866,7 @@ NOTICE:  avg_transfn called with 3
       2 |      6
 (1 row)
 
--- ideally these would share state, but we have to fix the OSAs first.
+-- exercise cases where OSAs share state
 select
   percentile_cont(0.5) within group (order by a),
   percentile_disc(0.5) within group (order by a)
@@ -1876,6 +1876,16 @@ from (values(1::float8),(3),(5),(7)) t(a);
                4 |               3
 (1 row)
 
+select
+  percentile_cont(0.25) within group (order by a),
+  percentile_disc(0.5) within group (order by a)
+from (values(1::float8),(3),(5),(7)) t(a);
+ percentile_cont | percentile_disc 
+-----------------+-----------------
+             2.5 |               3
+(1 row)
+
+-- these can't share state currently
 select
   rank(4) within group (order by a),
   dense_rank(4) within group (order by a)

src/test/regress/sql/aggregates.sql

@@ -741,12 +741,18 @@ select my_avg(one) filter (where one > 1),my_sum(one) from (values(1),(3)) t(one
 -- this should not share the state due to different input columns.
 select my_avg(one),my_sum(two) from (values(1,2),(3,4)) t(one,two);
 
--- ideally these would share state, but we have to fix the OSAs first.
+-- exercise cases where OSAs share state
 select
   percentile_cont(0.5) within group (order by a),
   percentile_disc(0.5) within group (order by a)
 from (values(1::float8),(3),(5),(7)) t(a);
 
+select
+  percentile_cont(0.25) within group (order by a),
+  percentile_disc(0.5) within group (order by a)
+from (values(1::float8),(3),(5),(7)) t(a);
+
+-- these can't share state currently
 select
   rank(4) within group (order by a),
   dense_rank(4) within group (order by a)