Merge cockroachdb#121904

121904: colexec: fix evaluation of the IN expression with INT2 and INT4 types r=yuzefovich a=yuzefovich

This commit fixes a possible incorrect result of evaluation of the IN expression where the left side is INT2 or INT4 type and the right side has integers outside of the range of the left side's type. We do so by upcasting the "filter" row to INT8. Note that we already do the appropriate upcast in the comparison function, so the incorrect results could only be produced due to the "filter row" no longer being sorted in case a value overflows.

Fixes: cockroachdb#102864.

Release note (bug fix): CockroachDB could previously incorrectly evaluate IN expressions that had INT2 or INT4 type on the left side and values outside of the range of the left side on the right side. The bug has been present since at least 21.1 and is now fixed.

Co-authored-by: Yahor Yuzefovich <yahor@cockroachlabs.com>

pkg/sql/colexec/execgen/cmd/execgen/select_in_gen.go

@@ -25,6 +25,7 @@ func genSelectIn(inputFileContents string, wr io.Writer) error {
 		"_CANONICAL_TYPE_FAMILY", "{{.CanonicalTypeFamilyStr}}",
 		"_TYPE_WIDTH", typeWidthReplacement,
 		"_GOTYPESLICE", "{{.GoTypeSliceName}}",
+		"_GOTYPE_UPCAST_INT", `{{if or (eq .VecMethod "Int16") (eq .VecMethod "Int32")}}int64{{else}}{{.GoType}}{{end}}`,
 		"_GOTYPE", "{{.GoType}}",
 		"_TYPE", "{{.VecMethod}}",
 		"TemplateType", "{{.VecMethod}}",

pkg/sql/colexec/select_in_tmpl.go

@@ -56,6 +56,7 @@ var (
 // {{/*
 
 type _GOTYPESLICE interface{}
+type _GOTYPE_UPCAST_INT interface{}
 type _GOTYPE interface{}
 type _TYPE interface{}
 
@@ -146,7 +147,7 @@ func GetInOperator(
 
 type selectInOp_TYPE struct {
 	colexecop.OneInputHelper
-	filterRow []_GOTYPE
+	filterRow []_GOTYPE_UPCAST_INT
 	colIdx    int
 	hasNulls  bool
 	negate    bool
@@ -157,7 +158,7 @@ var _ colexecop.Operator = &selectInOp_TYPE{}
 type projectInOp_TYPE struct {
 	colexecop.OneInputHelper
 	allocator *colmem.Allocator
-	filterRow []_GOTYPE
+	filterRow []_GOTYPE_UPCAST_INT
 	colIdx    int
 	outputIdx int
 	hasNulls  bool
@@ -168,27 +169,32 @@ var _ colexecop.Operator = &projectInOp_TYPE{}
 
 func fillDatumRow_TYPE(
 	evalCtx *eval.Context, t *types.T, datumTuple *tree.DTuple,
-) ([]_GOTYPE, bool) {
+) ([]_GOTYPE_UPCAST_INT, bool) {
 	// Sort the contents of the tuple, if they are not already sorted.
 	datumTuple.Normalize(evalCtx)
 
+	// {{if or (eq .VecMethod "Int16") (eq .VecMethod "Int32")}}
+	// Ensure that we always upcast all integer types.
+	conv := colconv.GetDatumToPhysicalFn(types.Int)
+	//{{else}}
 	conv := colconv.GetDatumToPhysicalFn(t)
-	var result []_GOTYPE
+	// {{end}}
+	var result []_GOTYPE_UPCAST_INT
 	hasNulls := false
 	for _, d := range datumTuple.D {
 		if d == tree.DNull {
 			hasNulls = true
 		} else {
 			convRaw := conv(d)
-			converted := convRaw.(_GOTYPE)
+			converted := convRaw.(_GOTYPE_UPCAST_INT)
 			result = append(result, converted)
 		}
 	}
 	return result, hasNulls
 }
 
 func cmpIn_TYPE(
-	targetElem _GOTYPE, targetCol _GOTYPESLICE, filterRow []_GOTYPE, hasNulls bool,
+	targetElem _GOTYPE, targetCol _GOTYPESLICE, filterRow []_GOTYPE_UPCAST_INT, hasNulls bool,
 ) comparisonResult {
 	// Filter row input was already sorted in fillDatumRow_TYPE, so we can
 	// perform a binary search.

pkg/sql/logictest/testdata/logic_test/select

@@ -938,3 +938,14 @@ ORDER BY 1 DESC
 ----
 5ebfedee-0dcf-41e6-a315-5fa0b51b9883  2  1999-11-30 23:59:58 +0000 +0000
 5ebfedee-0dcf-41e6-a315-5fa0b51b9882  1  1999-11-30 23:59:59 +0000 +0000
+
+# Regression test for incorrect evaluation of the IN filter due to integer
+# overflow when working with INT4 type (#102864).
+statement ok
+CREATE TABLE t102864 (c INT4);
+INSERT INTO t102864 (c) VALUES (0);
+
+query I
+SELECT c FROM t102864 WHERE c IN (0, 862827606027206657::INT8);
+----
+0