opt: fix error due to selectivity is NaN

This commit fixes the error "selectivity is NaN" by avoiding division
by zero.

Fixes cockroachdb#84478

There is no release note since this does not manifest as an error in
any existing releases (we have made our checking of valid selectivity
more strict on master).

Release note: None

pkg/sql/opt/memo/statistics_builder.go

@@ -976,7 +976,7 @@ func (sb *statisticsBuilder) colStatSelect(
 	// filter conditions were pushed down into the input after s.Selectivity
 	// was calculated. For example, an index scan or index join created during
 	// exploration could absorb some of the filter conditions.
-	selectivity := props.MakeSelectivity(s.RowCount / inputStats.RowCount)
+	selectivity := props.MakeSelectivityFromFraction(s.RowCount, inputStats.RowCount)
 	colStat.ApplySelectivity(selectivity, inputStats.RowCount)
 	if colSet.Intersects(relProps.NotNullCols) {
 		colStat.NullCount = 0

pkg/sql/opt/memo/testdata/stats/select

@@ -3232,3 +3232,71 @@ select
  │                     <--- 0 ------- 100000000000
  └── filters
       └── x:1 = 10 [type=bool, outer=(1), constraints=(/1: [/10 - /10]; tight), fd=()-->(1)]
+
+# Regression test for #84478. Avoid error due to selectivity is NaN.
+exec-ddl
+CREATE TABLE t84478 (
+  col0 FLOAT4,
+  col1 REGPROC NULL,
+  col2 BOX2D NOT NULL,
+  col3 TIMESTAMP NULL,
+  col4 FLOAT8 AS (col0 + NULL) VIRTUAL,
+  UNIQUE (col0) STORING (col1, col2, col3)
+)
+----
+
+norm
+SELECT
+  NULL, 1
+FROM
+  t84478 AS t1
+  JOIN t84478 AS t2
+    JOIN t84478 AS t3 ON
+        (t2.col4) = (t3.col0)
+        AND (t2.col3) = (t3.col3)
+        AND (t2.col2) = (t3.col2) ON (t1.tableoid) = (t2.col1)
+  RIGHT JOIN t84478 AS t4 ON
+      isnan(t1.crdb_internal_mvcc_timestamp::DECIMAL)::BOOL
+GROUP BY
+  t1.col2
+----
+project
+ ├── columns: "?column?":33(unknown) "?column?":34(int!null)
+ ├── immutable
+ ├── stats: [rows=1]
+ ├── fd: ()-->(33,34)
+ ├── distinct-on
+ │    ├── columns: col2:3(box2d)
+ │    ├── grouping columns: col2:3(box2d)
+ │    ├── immutable
+ │    ├── stats: [rows=1, distinct(3)=1, null(3)=1, avgsize(3)=0]
+ │    ├── key: (3)
+ │    └── left-join (cross)
+ │         ├── columns: col2:3(box2d) crdb_internal_mvcc_timestamp:7(decimal)
+ │         ├── multiplicity: left-rows(exactly-one), right-rows(zero-or-more)
+ │         ├── immutable
+ │         ├── stats: [rows=1000, distinct(3)=1, null(3)=1000, avgsize(3)=0]
+ │         ├── scan t84478
+ │         │    ├── computed column expressions
+ │         │    │    └── col4:29
+ │         │    │         └── CAST(NULL AS FLOAT8) [type=float]
+ │         │    └── stats: [rows=1000]
+ │         ├── select
+ │         │    ├── columns: col2:3(box2d!null) crdb_internal_mvcc_timestamp:7(decimal!null)
+ │         │    ├── cardinality: [0 - 0]
+ │         │    ├── immutable
+ │         │    ├── stats: [rows=0, distinct(3)=0, null(3)=0, avgsize(3)=0]
+ │         │    ├── key: ()
+ │         │    ├── fd: ()-->(3,7)
+ │         │    ├── values
+ │         │    │    ├── columns: col2:3(box2d!null) crdb_internal_mvcc_timestamp:7(decimal!null)
+ │         │    │    ├── cardinality: [0 - 0]
+ │         │    │    ├── stats: [rows=0, distinct(3)=0, null(3)=0, avgsize(3)=0]
+ │         │    │    ├── key: ()
+ │         │    │    └── fd: ()-->(3,7)
+ │         │    └── filters
+ │         │         └── isnan(crdb_internal_mvcc_timestamp:7) [type=bool, outer=(7), immutable, constraints=(/7: (/NULL - ])]
+ │         └── filters (true)
+ └── projections
+      ├── NULL [as="?column?":33, type=unknown]
+      └── 1 [as="?column?":34, type=int]

pkg/sql/opt/props/histogram.go

@@ -434,6 +434,9 @@ func (h *Histogram) addBucket(bucket *cat.HistogramBucket, desc bool) {
 // ApplySelectivity reduces the size of each histogram bucket according to
 // the given selectivity, and returns a new histogram with the results.
 func (h *Histogram) ApplySelectivity(selectivity Selectivity) *Histogram {
+	if selectivity == ZeroSelectivity {
+		return nil
+	}
 	res := h.copy()
 	for i := range res.buckets {
 		b := &res.buckets[i]