Merge pull request #9140 from planetscale/hash-join

gen4: Add hash join primitive and planning

go/hack/runtime.go

@@ -51,3 +51,6 @@ func roundupsize(size uintptr) uintptr
 func RuntimeAllocSize(size int64) int64 {
 	return int64(roundupsize(uintptr(size)))
 }
+
+//go:linkname ParseFloatPrefix strconv.parseFloatPrefix
+func ParseFloatPrefix(s string, bitSize int) (float64, int, error)

go/mysql/collations/8bit.go

@@ -82,7 +82,7 @@ func (c *Collation_8bit_bin) WeightString(dst, src []byte, numCodepoints int) []
 	return weightStringPadingSimple(' ', dst, numCodepoints-copyCodepoints, padToMax)
 }
 
-func (c *Collation_8bit_bin) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_8bit_bin) Hash(src []byte, numCodepoints int) HashCode {
 	hash := 0x8b8b0000 | uintptr(c.id)
 	if numCodepoints == 0 {
 		return memhash(src, hash)
@@ -164,7 +164,7 @@ func (c *Collation_8bit_simple_ci) WeightString(dst, src []byte, numCodepoints i
 	return weightStringPadingSimple(' ', dst, numCodepoints-copyCodepoints, padToMax)
 }
 
-func (c *Collation_8bit_simple_ci) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_8bit_simple_ci) Hash(src []byte, numCodepoints int) HashCode {
 	sortOrder := c.sort
 
 	var tocopy = len(src)
@@ -251,7 +251,7 @@ func (c *Collation_binary) WeightString(dst, src []byte, numCodepoints int) []by
 	return dst
 }
 
-func (c *Collation_binary) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_binary) Hash(src []byte, numCodepoints int) HashCode {
 	if numCodepoints > 0 {
 		src = src[:numCodepoints]
 	}

go/mysql/collations/collation.go

@@ -119,7 +119,7 @@ type Collation interface {
 	// the hash will interpret the source string as if it were stored in a `CHAR(n)` column. If the value of
 	// numCodepoints is 0, this is equivalent to setting `numCodepoints = RuneCount(src)`.
 	// For collations with NO PAD, the numCodepoint argument is ignored.
-	Hash(src []byte, numCodepoints int) uintptr
+	Hash(src []byte, numCodepoints int) HashCode
 
 	// Charset returns the Charset with which this collation is encoded
 	Charset() charset.Charset
@@ -128,6 +128,8 @@ type Collation interface {
 	IsBinary() bool
 }
 
+type HashCode = uintptr
+
 const PadToMax = math.MaxInt32
 
 func minInt(i1, i2 int) int {

go/mysql/collations/multibyte.go

@@ -147,7 +147,7 @@ func (c *Collation_multibyte) WeightString(dst, src []byte, numCodepoints int) [
 	return dst
 }
 
-func (c *Collation_multibyte) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_multibyte) Hash(src []byte, numCodepoints int) HashCode {
 	cs := c.charset
 	sortOrder := c.sort
 

go/mysql/collations/remote/collation.go

@@ -169,7 +169,7 @@ func (c *Collation) WeightString(dst, src []byte, numCodepoints int) []byte {
 	return dst
 }
 
-func (c *Collation) Hash(_ []byte, _ int) uintptr {
+func (c *Collation) Hash(_ []byte, _ int) collations.HashCode {
 	panic("unsupported: Hash for remote collations")
 }
 

go/mysql/collations/uca.go

@@ -161,7 +161,7 @@ performPadding:
 	return dst
 }
 
-func (c *Collation_utf8mb4_uca_0900) Hash(src []byte, _ int) uintptr {
+func (c *Collation_utf8mb4_uca_0900) Hash(src []byte, _ int) HashCode {
 	var hash = uintptr(c.id)
 
 	it := c.uca.Iterator(src)
@@ -234,7 +234,7 @@ func (c *Collation_utf8mb4_0900_bin) WeightString(dst, src []byte, numCodepoints
 	return dst
 }
 
-func (c *Collation_utf8mb4_0900_bin) Hash(src []byte, _ int) uintptr {
+func (c *Collation_utf8mb4_0900_bin) Hash(src []byte, _ int) HashCode {
 	return memhash(src, 0xb900b900)
 }
 
@@ -340,7 +340,7 @@ func (c *Collation_uca_legacy) WeightString(dst, src []byte, numCodepoints int)
 	return dst
 }
 
-func (c *Collation_uca_legacy) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_uca_legacy) Hash(src []byte, numCodepoints int) HashCode {
 	it := c.uca.Iterator(src)
 	defer it.Done()
 

go/mysql/collations/unicode.go

@@ -124,7 +124,7 @@ func (c *Collation_unicode_general_ci) WeightString(dst, src []byte, numCodepoin
 	return dst
 }
 
-func (c *Collation_unicode_general_ci) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_unicode_general_ci) Hash(src []byte, numCodepoints int) HashCode {
 	unicaseInfo := c.unicase
 	cs := c.charset
 
@@ -278,7 +278,7 @@ func (c *Collation_unicode_bin) weightStringUnicode(dst, src []byte, numCodepoin
 	return dst
 }
 
-func (c *Collation_unicode_bin) Hash(src []byte, numCodepoints int) uintptr {
+func (c *Collation_unicode_bin) Hash(src []byte, numCodepoints int) HashCode {
 	if c.charset.SupportsSupplementaryChars() {
 		return c.hashUnicode(src, numCodepoints)
 	}

go/sqltypes/type.go

@@ -91,6 +91,11 @@ func IsDate(t querypb.Type) bool {
 	return t == Datetime || t == Date || t == Timestamp || t == Time
 }
 
+// IsNull returns true if the type is NULL type
+func IsNull(t querypb.Type) bool {
+	return t == Null
+}
+
 // Vitess data types. These are idiomatically
 // named synonyms for the querypb.Type values.
 // Although these constants are interchangeable,

go/sqltypes/value.go

@@ -227,6 +227,13 @@ func (v Value) Raw() []byte {
 	return v.val
 }
 
+// RawStr returns the internal representation of the value as a string instead
+// of a byte slice. This is equivalent to calling `string(v.Raw())` but does
+// not allocate.
+func (v Value) RawStr() string {
+	return hack.String(v.val)
+}
+
 // ToBytes returns the value as MySQL would return it as []byte.
 // In contrast, Raw returns the internal representation of the Value, which may not
 // match MySQL's representation for newer types.

go/test/endtoend/vtgate/gen4/column_name_test.go

@@ -20,6 +20,8 @@ import (
 	"context"
 	"testing"
 
+	"vitess.io/vitess/go/test/endtoend/vtgate/utils"
+
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 
@@ -35,12 +37,10 @@ func TestColumnNames(t *testing.T) {
 	require.NoError(t, err)
 	defer conn.Close()
 
-	_, err = exec(t, conn, "create table uks.t2(id bigint,phone bigint,msg varchar(100),primary key(id)) Engine=InnoDB")
-	require.NoError(t, err)
-	defer exec(t, conn, "drop table uks.t2")
+	utils.Exec(t, conn, "create table uks.t2(id bigint,phone bigint,msg varchar(100),primary key(id)) Engine=InnoDB")
+	defer utils.Exec(t, conn, "drop table uks.t2")
 
-	qr, err := exec(t, conn, "SELECT t1.id as t1id, t2.id as t2id, t2.phone as t2phn FROM ks.t1 cross join uks.t2 where t1.id = t2.id ORDER BY t2.phone")
-	require.NoError(t, err)
+	qr := utils.Exec(t, conn, "SELECT t1.id as t1id, t2.id as t2id, t2.phone as t2phn FROM ks.t1 cross join uks.t2 where t1.id = t2.id ORDER BY t2.phone")
 
 	assert.Equal(t, 3, len(qr.Fields))
 	assert.Equal(t, "t1id", qr.Fields[0].Name)