go/types, types2: add Sizes computation to match gc behavior

Fixes #60431
Fixes #60734
Fixes #61035

Change-Id: I82513da3e1714e8271fae220fe242bf2bfb4eb9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/506856
Auto-Submit: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Reviewed-by: Robert Griesemer <gri@google.com>
Run-TryBot: Robert Griesemer <gri@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Auto-Submit: Robert Griesemer <gri@google.com>

src/cmd/compile/internal/types2/gcsizes.go

@@ -0,0 +1,169 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package types2
+
+type gcSizes struct {
+	WordSize int64 // word size in bytes - must be >= 4 (32bits)
+	MaxAlign int64 // maximum alignment in bytes - must be >= 1
+}
+
+func (s *gcSizes) Alignof(T Type) (result int64) {
+	defer func() {
+		assert(result >= 1)
+	}()
+
+	// For arrays and structs, alignment is defined in terms
+	// of alignment of the elements and fields, respectively.
+	switch t := under(T).(type) {
+	case *Array:
+		// spec: "For a variable x of array type: unsafe.Alignof(x)
+		// is the same as unsafe.Alignof(x[0]), but at least 1."
+		return s.Alignof(t.elem)
+	case *Struct:
+		if len(t.fields) == 0 && IsSyncAtomicAlign64(T) {
+			// Special case: sync/atomic.align64 is an
+			// empty struct we recognize as a signal that
+			// the struct it contains must be
+			// 64-bit-aligned.
+			//
+			// This logic is equivalent to the logic in
+			// cmd/compile/internal/types/size.go:calcStructOffset
+			return 8
+		}
+
+		// spec: "For a variable x of struct type: unsafe.Alignof(x)
+		// is the largest of the values unsafe.Alignof(x.f) for each
+		// field f of x, but at least 1."
+		max := int64(1)
+		for _, f := range t.fields {
+			if a := s.Alignof(f.typ); a > max {
+				max = a
+			}
+		}
+		return max
+	case *Slice, *Interface:
+		// Multiword data structures are effectively structs
+		// in which each element has size WordSize.
+		// Type parameters lead to variable sizes/alignments;
+		// StdSizes.Alignof won't be called for them.
+		assert(!isTypeParam(T))
+		return s.WordSize
+	case *Basic:
+		// Strings are like slices and interfaces.
+		if t.Info()&IsString != 0 {
+			return s.WordSize
+		}
+	case *TypeParam, *Union:
+		unreachable()
+	}
+	a := s.Sizeof(T) // may be 0 or negative
+	// spec: "For a variable x of any type: unsafe.Alignof(x) is at least 1."
+	if a < 1 {
+		return 1
+	}
+	// complex{64,128} are aligned like [2]float{32,64}.
+	if isComplex(T) {
+		a /= 2
+	}
+	if a > s.MaxAlign {
+		return s.MaxAlign
+	}
+	return a
+}
+
+func (s *gcSizes) Offsetsof(fields []*Var) []int64 {
+	offsets := make([]int64, len(fields))
+	var offs int64
+	for i, f := range fields {
+		if offs < 0 {
+			// all remaining offsets are too large
+			offsets[i] = -1
+			continue
+		}
+		// offs >= 0
+		a := s.Alignof(f.typ)
+		offs = align(offs, a) // possibly < 0 if align overflows
+		offsets[i] = offs
+		if d := s.Sizeof(f.typ); d >= 0 && offs >= 0 {
+			offs += d // ok to overflow to < 0
+		} else {
+			offs = -1 // f.typ or offs is too large
+		}
+	}
+	return offsets
+}
+
+func (s *gcSizes) Sizeof(T Type) int64 {
+	switch t := under(T).(type) {
+	case *Basic:
+		assert(isTyped(T))
+		k := t.kind
+		if int(k) < len(basicSizes) {
+			if s := basicSizes[k]; s > 0 {
+				return int64(s)
+			}
+		}
+		if k == String {
+			return s.WordSize * 2
+		}
+	case *Array:
+		n := t.len
+		if n <= 0 {
+			return 0
+		}
+		// n > 0
+		esize := s.Sizeof(t.elem)
+		if esize < 0 {
+			return -1 // element too large
+		}
+		if esize == 0 {
+			return 0 // 0-size element
+		}
+		// esize > 0
+		// Final size is esize * n; and size must be <= maxInt64.
+		const maxInt64 = 1<<63 - 1
+		if esize > maxInt64/n {
+			return -1 // esize * n overflows
+		}
+		return esize * n
+	case *Slice:
+		return s.WordSize * 3
+	case *Struct:
+		n := t.NumFields()
+		if n == 0 {
+			return 0
+		}
+		offsets := s.Offsetsof(t.fields)
+		offs := offsets[n-1]
+		size := s.Sizeof(t.fields[n-1].typ)
+		if offs < 0 || size < 0 {
+			return -1 // type too large
+		}
+		// gc: The last field of a non-zero-sized struct is not allowed to
+		// have size 0.
+		if offs > 0 && size == 0 {
+			size = 1
+		}
+		// gc: Size includes alignment padding.
+		return align(offs+size, s.Alignof(t)) // may overflow to < 0 which is ok
+	case *Interface:
+		// Type parameters lead to variable sizes/alignments;
+		// StdSizes.Sizeof won't be called for them.
+		assert(!isTypeParam(T))
+		return s.WordSize * 2
+	case *TypeParam, *Union:
+		unreachable()
+	}
+	return s.WordSize // catch-all
+}
+
+// gcSizesFor returns the Sizes used by gc for an architecture.
+// The result is nil if a compiler/architecture pair is not known.
+func gcSizesFor(compiler, arch string) *gcSizes {
+	if compiler != "gc" {
+		return nil
+	}
+	return gcArchSizes[arch]
+}

src/cmd/compile/internal/types2/sizes.go

@@ -227,7 +227,7 @@ func (s *StdSizes) Sizeof(T Type) int64 {
 }
 
 // common architecture word sizes and alignments
-var gcArchSizes = map[string]*StdSizes{
+var gcArchSizes = map[string]*gcSizes{
 	"386":      {4, 4},
 	"amd64":    {8, 8},
 	"amd64p32": {4, 8},
@@ -255,20 +255,15 @@ var gcArchSizes = map[string]*StdSizes{
 // "386", "amd64", "amd64p32", "arm", "arm64", "loong64", "mips", "mipsle",
 // "mips64", "mips64le", "ppc64", "ppc64le", "riscv64", "s390x", "sparc64", "wasm".
 func SizesFor(compiler, arch string) Sizes {
-	var m map[string]*StdSizes
 	switch compiler {
 	case "gc":
-		m = gcArchSizes
+		return gcSizesFor(compiler, arch)
 	case "gccgo":
-		m = gccgoArchSizes
-	default:
-		return nil
+		if s, ok := gccgoArchSizes[arch]; ok {
+			return s
+		}
 	}
-	s, ok := m[arch]
-	if !ok {
-		return nil
-	}
-	return s
+	return nil
 }
 
 // stdSizes is used if Config.Sizes == nil.

src/cmd/compile/internal/types2/sizes_test.go

@@ -133,3 +133,62 @@ var s struct {
 		})
 	}
 }
+
+type gcSizeTest struct {
+	name string
+	src  string
+}
+
+var gcSizesTests = []gcSizeTest{
+	{
+		"issue60431",
+		`
+package main
+
+import "unsafe"
+
+// The foo struct size is expected to be rounded up to 16 bytes.
+type foo struct {
+	a int64
+	b bool
+}
+
+func main() {
+	assert(unsafe.Sizeof(foo{}) == 16)
+}`,
+	},
+	{
+		"issue60734",
+		`
+package main
+
+import (
+	"unsafe"
+)
+
+// The Data struct size is expected to be rounded up to 16 bytes.
+type Data struct {
+	Value  uint32   // 4 bytes
+	Label  [10]byte // 10 bytes
+	Active bool     // 1 byte
+	// padded with 1 byte to make it align
+}
+
+func main() {
+	assert(unsafe.Sizeof(Data{}) == 16)
+}
+`,
+	},
+}
+
+func TestGCSizes(t *testing.T) {
+	types2.DefPredeclaredTestFuncs()
+	for _, tc := range gcSizesTests {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			conf := types2.Config{Importer: defaultImporter(), Sizes: types2.SizesFor("gc", "amd64")}
+			mustTypecheck(tc.src, &conf, nil)
+		})
+	}
+}

src/go/types/gcsizes.go

@@ -0,0 +1,171 @@
+// Code generated by "go test -run=Generate -write=all"; DO NOT EDIT.
+
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package types
+
+type gcSizes struct {
+	WordSize int64 // word size in bytes - must be >= 4 (32bits)
+	MaxAlign int64 // maximum alignment in bytes - must be >= 1
+}
+
+func (s *gcSizes) Alignof(T Type) (result int64) {
+	defer func() {
+		assert(result >= 1)
+	}()
+
+	// For arrays and structs, alignment is defined in terms
+	// of alignment of the elements and fields, respectively.
+	switch t := under(T).(type) {
+	case *Array:
+		// spec: "For a variable x of array type: unsafe.Alignof(x)
+		// is the same as unsafe.Alignof(x[0]), but at least 1."
+		return s.Alignof(t.elem)
+	case *Struct:
+		if len(t.fields) == 0 && _IsSyncAtomicAlign64(T) {
+			// Special case: sync/atomic.align64 is an
+			// empty struct we recognize as a signal that
+			// the struct it contains must be
+			// 64-bit-aligned.
+			//
+			// This logic is equivalent to the logic in
+			// cmd/compile/internal/types/size.go:calcStructOffset
+			return 8
+		}
+
+		// spec: "For a variable x of struct type: unsafe.Alignof(x)
+		// is the largest of the values unsafe.Alignof(x.f) for each
+		// field f of x, but at least 1."
+		max := int64(1)
+		for _, f := range t.fields {
+			if a := s.Alignof(f.typ); a > max {
+				max = a
+			}
+		}
+		return max
+	case *Slice, *Interface:
+		// Multiword data structures are effectively structs
+		// in which each element has size WordSize.
+		// Type parameters lead to variable sizes/alignments;
+		// StdSizes.Alignof won't be called for them.
+		assert(!isTypeParam(T))
+		return s.WordSize
+	case *Basic:
+		// Strings are like slices and interfaces.
+		if t.Info()&IsString != 0 {
+			return s.WordSize
+		}
+	case *TypeParam, *Union:
+		unreachable()
+	}
+	a := s.Sizeof(T) // may be 0 or negative
+	// spec: "For a variable x of any type: unsafe.Alignof(x) is at least 1."
+	if a < 1 {
+		return 1
+	}
+	// complex{64,128} are aligned like [2]float{32,64}.
+	if isComplex(T) {
+		a /= 2
+	}
+	if a > s.MaxAlign {
+		return s.MaxAlign
+	}
+	return a
+}
+
+func (s *gcSizes) Offsetsof(fields []*Var) []int64 {
+	offsets := make([]int64, len(fields))
+	var offs int64
+	for i, f := range fields {
+		if offs < 0 {
+			// all remaining offsets are too large
+			offsets[i] = -1
+			continue
+		}
+		// offs >= 0
+		a := s.Alignof(f.typ)
+		offs = align(offs, a) // possibly < 0 if align overflows
+		offsets[i] = offs
+		if d := s.Sizeof(f.typ); d >= 0 && offs >= 0 {
+			offs += d // ok to overflow to < 0
+		} else {
+			offs = -1 // f.typ or offs is too large
+		}
+	}
+	return offsets
+}
+
+func (s *gcSizes) Sizeof(T Type) int64 {
+	switch t := under(T).(type) {
+	case *Basic:
+		assert(isTyped(T))
+		k := t.kind
+		if int(k) < len(basicSizes) {
+			if s := basicSizes[k]; s > 0 {
+				return int64(s)
+			}
+		}
+		if k == String {
+			return s.WordSize * 2
+		}
+	case *Array:
+		n := t.len
+		if n <= 0 {
+			return 0
+		}
+		// n > 0
+		esize := s.Sizeof(t.elem)
+		if esize < 0 {
+			return -1 // element too large
+		}
+		if esize == 0 {
+			return 0 // 0-size element
+		}
+		// esize > 0
+		// Final size is esize * n; and size must be <= maxInt64.
+		const maxInt64 = 1<<63 - 1
+		if esize > maxInt64/n {
+			return -1 // esize * n overflows
+		}
+		return esize * n
+	case *Slice:
+		return s.WordSize * 3
+	case *Struct:
+		n := t.NumFields()
+		if n == 0 {
+			return 0
+		}
+		offsets := s.Offsetsof(t.fields)
+		offs := offsets[n-1]
+		size := s.Sizeof(t.fields[n-1].typ)
+		if offs < 0 || size < 0 {
+			return -1 // type too large
+		}
+		// gc: The last field of a non-zero-sized struct is not allowed to
+		// have size 0.
+		if offs > 0 && size == 0 {
+			size = 1
+		}
+		// gc: Size includes alignment padding.
+		return align(offs+size, s.Alignof(t)) // may overflow to < 0 which is ok
+	case *Interface:
+		// Type parameters lead to variable sizes/alignments;
+		// StdSizes.Sizeof won't be called for them.
+		assert(!isTypeParam(T))
+		return s.WordSize * 2
+	case *TypeParam, *Union:
+		unreachable()
+	}
+	return s.WordSize // catch-all
+}
+
+// gcSizesFor returns the Sizes used by gc for an architecture.
+// The result is nil if a compiler/architecture pair is not known.
+func gcSizesFor(compiler, arch string) *gcSizes {
+	if compiler != "gc" {
+		return nil
+	}
+	return gcArchSizes[arch]
+}

src/go/types/generate_test.go

@@ -102,6 +102,7 @@ var filemap = map[string]action{
 	"context.go":      nil,
 	"context_test.go": nil,
 	"gccgosizes.go":   nil,
+	"gcsizes.go":      func(f *ast.File) { renameIdent(f, "IsSyncAtomicAlign64", "_IsSyncAtomicAlign64") },
 	"hilbert_test.go": func(f *ast.File) { renameImportPath(f, `"cmd/compile/internal/types2"`, `"go/types"`) },
 	"infer.go": func(f *ast.File) {
 		fixTokenPos(f)

src/go/types/sizes.go

@@ -229,7 +229,7 @@ func (s *StdSizes) Sizeof(T Type) int64 {
 }
 
 // common architecture word sizes and alignments
-var gcArchSizes = map[string]*StdSizes{
+var gcArchSizes = map[string]*gcSizes{
 	"386":      {4, 4},
 	"amd64":    {8, 8},
 	"amd64p32": {4, 8},
@@ -257,20 +257,15 @@ var gcArchSizes = map[string]*StdSizes{
 // "386", "amd64", "amd64p32", "arm", "arm64", "loong64", "mips", "mipsle",
 // "mips64", "mips64le", "ppc64", "ppc64le", "riscv64", "s390x", "sparc64", "wasm".
 func SizesFor(compiler, arch string) Sizes {
-	var m map[string]*StdSizes
 	switch compiler {
 	case "gc":
-		m = gcArchSizes
+		return gcSizesFor(compiler, arch)
 	case "gccgo":
-		m = gccgoArchSizes
-	default:
-		return nil
+		if s, ok := gccgoArchSizes[arch]; ok {
+			return s
+		}
 	}
-	s, ok := m[arch]
-	if !ok {
-		return nil
-	}
-	return s
+	return nil
 }
 
 // stdSizes is used if Config.Sizes == nil.

src/go/types/sizes_test.go

@@ -134,3 +134,62 @@ var s struct {
 		})
 	}
 }
+
+type gcSizeTest struct {
+	name string
+	src  string
+}
+
+var gcSizesTests = []gcSizeTest{
+	{
+		"issue60431",
+		`
+package main
+
+import "unsafe"
+
+// The foo struct size is expected to be rounded up to 16 bytes.
+type foo struct {
+	a int64
+	b bool
+}
+
+func main() {
+	assert(unsafe.Sizeof(foo{}) == 16)
+}`,
+	},
+	{
+		"issue60734",
+		`
+package main
+
+import (
+	"unsafe"
+)
+
+// The Data struct size is expected to be rounded up to 16 bytes.
+type Data struct {
+	Value  uint32   // 4 bytes
+	Label  [10]byte // 10 bytes
+	Active bool     // 1 byte
+	// padded with 1 byte to make it align
+}
+
+func main() {
+	assert(unsafe.Sizeof(Data{}) == 16)
+}
+`,
+	},
+}
+
+func TestGCSizes(t *testing.T) {
+	types.DefPredeclaredTestFuncs()
+	for _, tc := range gcSizesTests {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			conf := types.Config{Importer: importer.Default(), Sizes: types.SizesFor("gc", "amd64")}
+			mustTypecheck(tc.src, &conf, nil)
+		})
+	}
+}

src/internal/types/testdata/check/builtins0.go

@@ -792,16 +792,16 @@ type S2 struct{ // offset
 type S3 struct { // offset
 	a int64  //  0
 	b int32  //  8
-}                // 12
+}                // 16
 
 type S4 struct { // offset
 	S3       //  0
 	int32    // 12
-}                // 16
+}                // 24
 
 type S5 struct {   // offset
 	a [3]int32 //  0
-	b int32    // 12
+	b int32    // 16
 }                  // 16
 
 func (S2) m() {}
@@ -936,16 +936,16 @@ func Sizeof1() {
 	assert(unsafe.Sizeof(y2) == 8)
 
 	var y3 S3
-	assert(unsafe.Sizeof(y3) == 12)
+	assert(unsafe.Sizeof(y3) == 16)
 
 	var y4 S4
-	assert(unsafe.Sizeof(y4) == 16)
+	assert(unsafe.Sizeof(y4) == 24)
 
 	var y5 S5
 	assert(unsafe.Sizeof(y5) == 16)
 
 	var a3 [10]S3
-	assert(unsafe.Sizeof(a3) == 156)
+	assert(unsafe.Sizeof(a3) == 160)
 
 	// test case for issue 5670
 	type T struct {