opt: fix panic with empty column sets in set operations

Fix panic "column statistics cannot be determined for empty column set"
when performing set operations (UNION, INTERSECT, EXCEPT) on tables
with no columns (e.g., created with "SELECT ALL").

Relatedly, relax the "invalid union" assertion in
CanConvertUnionToDistinctUnionAll that incorrectly assumed that we will
never have an operation with no columns.

Fixes #130593

Release note: None

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

Author: yuzefovich

pkg/sql/execinfrapb/processors_sql.proto

@@ -407,7 +407,8 @@ message DistinctSpec {
   // check for distinct rows. If A,B,C are in distinct_columns and there is a
   // 4th column D which is not included in distinct_columns, its values are not
   // considered, so rows A1,B1,C1,D1 and A1,B1,C1,D2 are considered equal and
-  // only one of them (the first) is output.
+  // only one of them (the first) is output. distinct_columns can be empty
+  // (which means that all rows are considered equal).
   repeated uint32 distinct_columns = 2;
   // If true, then NULL values are treated as not equal to one another. Each NULL
   // value will cause a new row group to be created. For example:

pkg/sql/logictest/testdata/logic_test/union

@@ -750,3 +750,52 @@ query II
 SELECT k, v FROM v127043_2 INNER JOIN t127043_3 ON k = k3 WHERE k = 1 LIMIT 1;
 ----
 1  1
+
+# Regression tests for #130593 (set operations with empty column sets).
+statement ok
+CREATE TABLE t130591_1 AS SELECT ALL;
+CREATE TABLE t130591_2 AS SELECT ALL
+
+statement count 0
+TABLE t130591_1 EXCEPT DISTINCT TABLE t130591_1;
+
+statement count 1
+TABLE t130591_1 INTERSECT DISTINCT TABLE t130591_1;
+
+statement count 1
+TABLE t130591_1 UNION DISTINCT TABLE t130591_1;
+
+statement count 0
+TABLE t130591_1 EXCEPT DISTINCT TABLE t130591_2;
+
+statement count 1
+TABLE t130591_1 INTERSECT DISTINCT TABLE t130591_2;
+
+statement count 1
+TABLE t130591_1 UNION DISTINCT TABLE t130591_2;
+
+statement ok
+INSERT INTO t130591_1 DEFAULT VALUES;
+
+query I
+SELECT count(*) FROM t130591_1;
+----
+2
+
+statement count 0
+TABLE t130591_1 EXCEPT DISTINCT TABLE t130591_1;
+
+statement count 1
+TABLE t130591_1 INTERSECT DISTINCT TABLE t130591_1;
+
+statement count 1
+TABLE t130591_1 UNION DISTINCT TABLE t130591_1;
+
+statement count 0
+TABLE t130591_1 EXCEPT TABLE t130591_1;
+
+statement count 1
+TABLE t130591_1 INTERSECT TABLE t130591_1;
+
+statement count 1
+TABLE t130591_1 UNION TABLE t130591_1;

pkg/sql/opt/memo/statistics_builder.go

@@ -2193,8 +2193,16 @@ func (sb *statisticsBuilder) buildSetNode(setNode RelExpr, relProps *props.Relat
 		// count will equal the distinct count of the set of output columns.
 		setPrivate := setNode.Private().(*SetPrivate)
 		outputCols := setPrivate.OutCols.ToSet()
-		colStat := sb.colStatSetNodeImpl(outputCols, setNode, relProps)
-		s.RowCount = colStat.DistinctCount
+		if outputCols.Empty() {
+			// When there are no output columns (e.g., SELECT ALL), the distinct
+			// count is either 0 or 1, so we use the already computed row count.
+			if s.RowCount > 0 {
+				s.RowCount = 1
+			}
+		} else {
+			colStat := sb.colStatSetNodeImpl(outputCols, setNode, relProps)
+			s.RowCount = colStat.DistinctCount
+		}
 	}
 
 	sb.finalizeFromCardinality(relProps)

pkg/sql/opt/norm/set_funcs.go

@@ -103,9 +103,13 @@ func (c *CustomFuncs) PruneSetPrivate(needed opt.ColSet, set *memo.SetPrivate) *
 func (c *CustomFuncs) CanConvertUnionToDistinctUnionAll(
 	leftCols, rightCols opt.ColList,
 ) (keyCols opt.ColSet, ok bool) {
-	if len(leftCols) != len(rightCols) || len(leftCols) == 0 {
+	if len(leftCols) != len(rightCols) {
 		panic(errors.AssertionFailedf("invalid union"))
 	}
+	if len(leftCols) == 0 {
+		// We must have some columns so that we have a non-empty key.
+		return opt.ColSet{}, false
+	}
 	md := c.mem.Metadata()
 	leftTableID, rightTableID := md.ColumnMeta(leftCols[0]).Table, md.ColumnMeta(rightCols[0]).Table
 	if leftTableID == opt.TableID(0) || rightTableID == opt.TableID(0) {

pkg/sql/rowexec/distinct.go

@@ -75,10 +75,6 @@ func newDistinct(
 	input execinfra.RowSource,
 	post *execinfrapb.PostProcessSpec,
 ) (execinfra.RowSourcedProcessor, error) {
-	if len(spec.DistinctColumns) == 0 {
-		return nil, errors.AssertionFailedf("0 distinct columns specified for distinct processor")
-	}
-
 	nonOrderedCols := make([]uint32, 0, len(spec.DistinctColumns)-len(spec.OrderedColumns))
 	for _, col := range spec.DistinctColumns {
 		ordered := false