Merge #40094

40094: exec: randomize types when testing columnar operators against processors r=yuzefovich a=yuzefovich

This commit adds type randomization when testing columnar operators
against processors. A few edge cases have been uncovered in the vectorize
engine when handling special values, and those are now fixed (we might need
to revisit the fixes though).

Additionally, the generation of random datums has been adjusted to
take into account the requested width for integers and floats. Without
the adjustment, we observe different behavior between the row and the
vectorized engines since the former always upcasts to the maximum
width type. For example, if a special math.MaxInt64 value was generated,
but the requested type was Int4, then the row engine would use the full
int64 value, but the vectorized engine would use only the int32 value.

Release note: None

Co-authored-by: Yahor Yuzefovich <[email protected]>

pkg/sql/distsqlrun/column_exec_setup.go

@@ -673,6 +673,12 @@ func (r *newColOperatorResult) planFilterExpr(
 	if err != nil {
 		return columnTypes, err
 	}
+	if helper.expr == tree.DNull {
+		// The filter expression is tree.DNull meaning that it is always false, so
+		// we put a zero operator.
+		r.op = exec.NewZeroOp(r.op)
+		return columnTypes, nil
+	}
 	var filterColumnTypes []types.T
 	r.op, _, filterColumnTypes, selectionMem, err = planSelectionOperators(flowCtx.NewEvalCtx(), helper.expr, columnTypes, r.op)
 	if err != nil {