cmd/cgo, debug/dwarf: fix translation of zero-size arrays

In cgo, now that recursive calls to typeConv.Type() always work,
we can more robustly calculate the array sizes based on the size
of our element type.

Also, in debug/dwarf, the decision to call zeroType is made
based on a type's usage	within a particular struct, but dwarf.Type
values are cached in typeCache, so the modification might affect
uses of the type in other structs.  Current compilers don't appear
to share DWARF type entries for "[]foo" and "[0]foo", but they also
don't consistently share type entries in other cases.  Arguably
modifying the types is an improvement in some cases, but varying
translated types according to compiler whims seems like a bad idea.

Lastly, also in debug/dwarf, zeroType only needs to rewrite the
top-level dimension, and only if the rest of the array size is
non-zero.

Fixes issue 8428.

LGTM=iant
R=iant
CC=golang-codereviews
https://codereview.appspot.com/127980043

Committer: Ian Lance Taylor <iant@golang.org>

3 files changed, 108 insertions, 57 deletions

diff --git a/misc/cgo/test/issue8428.go b/misc/cgo/test/issue8428.go
new file mode 100644
index 000000000..f5b0ee497
--- /dev/null
+++ b/misc/cgo/test/issue8428.go
@@ -0,0 +1,48 @@
+// Copyright 2014 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 cgotest
+
+// Issue 8428.  Cgo inconsistently translated zero size arrays.
+
+/*
+struct issue8428one {
+	char b;
+	char rest[];
+};
+
+struct issue8428two {
+	void *p;
+	char b;
+	char rest[0];
+};
+
+struct issue8428three {
+	char w[1][2][3][0];
+	char x[2][3][0][1];
+	char y[3][0][1][2];
+	char z[0][1][2][3];
+};
+*/
+import "C"
+
+import "unsafe"
+
+var _ = C.struct_issue8428one{
+	b:    C.char(0),
+	rest: [0]C.char{},
+}
+
+var _ = C.struct_issue8428two{
+	p:    unsafe.Pointer(nil),
+	b:    C.char(0),
+	rest: [0]C.char{},
+}
+
+var _ = C.struct_issue8428three{
+	w: [1][2][3][0]C.char{},
+	x: [2][3][0][1]C.char{},
+	y: [3][0][1][2]C.char{},
+	z: [0][1][2][3]C.char{},
+}
diff --git a/src/cmd/cgo/gcc.go b/src/cmd/cgo/gcc.go
index 26def654d..482b35197 100644
--- a/src/cmd/cgo/gcc.go
+++ b/src/cmd/cgo/gcc.go
@@ -1075,12 +1075,6 @@ func (c *typeConv) Type(dtype dwarf.Type, pos token.Pos) *Type {
 		return t
 	}
 
-	// clang won't generate DW_AT_byte_size for pointer types,
-	// so we have to fix it here.
-	if dt, ok := base(dtype).(*dwarf.PtrType); ok && dt.ByteSize == -1 {
-		dt.ByteSize = c.ptrSize
-	}
-
 	t := new(Type)
 	t.Size = dtype.Size() // note: wrong for array of pointers, corrected below
 	t.Align = -1
@@ -1104,12 +1098,20 @@ func (c *typeConv) Type(dtype dwarf.Type, pos token.Pos) *Type {
 			t.Go = c.Opaque(t.Size)
 			break
 		}
+		count := dt.Count
+		if count == -1 {
+			// Indicates flexible array member, which Go doesn't support.
+			// Translate to zero-length array instead.
+			count = 0
+		}
 		sub := c.Type(dt.Type, pos)
 		t.Align = sub.Align
 		t.Go = &ast.ArrayType{
-			Len: c.intExpr(dt.Count),
+			Len: c.intExpr(count),
 			Elt: sub.Go,
 		}
+		// Recalculate t.Size now that we know sub.Size.
+		t.Size = count * sub.Size
 		t.C.Set("__typeof__(%s[%d])", sub.C, dt.Count)
 
 	case *dwarf.BoolType:
@@ -1210,6 +1212,11 @@ func (c *typeConv) Type(dtype dwarf.Type, pos token.Pos) *Type {
 		}
 
 	case *dwarf.PtrType:
+		// Clang doesn't emit DW_AT_byte_size for pointer types.
+		if t.Size != c.ptrSize && t.Size != -1 {
+			fatalf("%s: unexpected: %d-byte pointer type - %s", lineno(pos), t.Size, dtype)
+		}
+		t.Size = c.ptrSize
 		t.Align = c.ptrSize
 
 		if _, ok := base(dt.Type).(*dwarf.VoidType); ok {
@@ -1381,34 +1388,24 @@ func (c *typeConv) Type(dtype dwarf.Type, pos token.Pos) *Type {
 		}
 	}
 
-	if t.Size <= 0 {
-		// Clang does not record the size of a pointer in its DWARF entry,
-		// so if dtype is an array, the call to dtype.Size at the top of the function
-		// computed the size as the array length * 0 = 0.
-		// The type switch called Type (this function) recursively on the pointer
-		// entry, and the code near the top of the function updated the size to
-		// be correct, so calling dtype.Size again will produce the correct value.
-		t.Size = dtype.Size()
-		if t.Size < 0 {
-			// Unsized types are [0]byte, unless they're typedefs of other types
-			// or structs with tags.
-			// if so, use the name we've already defined.
-			t.Size = 0
-			switch dt := dtype.(type) {
-			case *dwarf.TypedefType:
-				// ok
-			case *dwarf.StructType:
-				if dt.StructName != "" {
-					break
-				}
-				t.Go = c.Opaque(0)
-			default:
-				t.Go = c.Opaque(0)
-			}
-			if t.C.Empty() {
-				t.C.Set("void")
+	if t.Size < 0 {
+		// Unsized types are [0]byte, unless they're typedefs of other types
+		// or structs with tags.
+		// if so, use the name we've already defined.
+		t.Size = 0
+		switch dt := dtype.(type) {
+		case *dwarf.TypedefType:
+			// ok
+		case *dwarf.StructType:
+			if dt.StructName != "" {
+				break
 			}
-			return t
+			t.Go = c.Opaque(0)
+		default:
+			t.Go = c.Opaque(0)
+		}
+		if t.C.Empty() {
+			t.C.Set("void")
 		}
 	}
 
diff --git a/src/pkg/debug/dwarf/type.go b/src/pkg/debug/dwarf/type.go
index b64333ecc..fa40b2bef 100644
--- a/src/pkg/debug/dwarf/type.go
+++ b/src/pkg/debug/dwarf/type.go
@@ -118,7 +118,12 @@ func (t *ArrayType) String() string {
 	return "[" + strconv.FormatInt(t.Count, 10) + "]" + t.Type.String()
 }
 
-func (t *ArrayType) Size() int64 { return t.Count * t.Type.Size() }
+func (t *ArrayType) Size() int64 {
+	if t.Count == -1 {
+		return 0
+	}
+	return t.Count * t.Type.Size()
+}
 
 // A VoidType represents the C void type.
 type VoidType struct {
@@ -369,7 +374,7 @@ func (d *Data) readType(name string, r typeReader, off Offset, typeCache map[Off
 		t.StrideBitSize, _ = e.Val(AttrStrideSize).(int64)
 
 		// Accumulate dimensions,
-		ndim := 0
+		var dims []int64
 		for kid := next(); kid != nil; kid = next() {
 			// TODO(rsc): Can also be TagEnumerationType
 			// but haven't seen that in the wild yet.
@@ -381,26 +386,24 @@ func (d *Data) readType(name string, r typeReader, off Offset, typeCache map[Off
 					count, ok = kid.Val(AttrUpperBound).(int64)
 					if ok {
 						count++ // Length is one more than upper bound.
-					} else {
+					} else if len(dims) == 0 {
 						count = -1 // As in x[].
 					}
 				}
-				if ndim == 0 {
-					t.Count = count
-				} else {
-					// Multidimensional array.
-					// Create new array type underneath this one.
-					t.Type = &ArrayType{Type: t.Type, Count: count}
-				}
-				ndim++
+				dims = append(dims, count)
 			case TagEnumerationType:
 				err = DecodeError{name, kid.Offset, "cannot handle enumeration type as array bound"}
 				goto Error
 			}
 		}
-		if ndim == 0 {
+		if len(dims) == 0 {
 			// LLVM generates this for x[].
-			t.Count = -1
+			dims = []int64{-1}
+		}
+
+		t.Count = dims[0]
+		for i := len(dims) - 1; i >= 1; i-- {
+			t.Type = &ArrayType{Type: t.Type, Count: dims[i]}
 		}
 
 	case TagBaseType:
@@ -476,7 +479,7 @@ func (d *Data) readType(name string, r typeReader, off Offset, typeCache map[Off
 		t.StructName, _ = e.Val(AttrName).(string)
 		t.Incomplete = e.Val(AttrDeclaration) != nil
 		t.Field = make([]*StructField, 0, 8)
-		var lastFieldType Type
+		var lastFieldType *Type
 		var lastFieldBitOffset int64
 		for kid := next(); kid != nil; kid = next() {
 			if kid.Tag == TagMember {
@@ -518,7 +521,7 @@ func (d *Data) readType(name string, r typeReader, off Offset, typeCache map[Off
 					// (DWARF writes out 0-length arrays as if they were 1-length arrays.)
 					zeroArray(lastFieldType)
 				}
-				lastFieldType = f.Type
+				lastFieldType = &f.Type
 				lastFieldBitOffset = bito
 			}
 		}
@@ -667,13 +670,16 @@ Error:
 	return nil, err
 }
 
-func zeroArray(t Type) {
-	for {
-		at, ok := t.(*ArrayType)
-		if !ok {
-			break
-		}
-		at.Count = 0
-		t = at.Type
+func zeroArray(t *Type) {
+	if t == nil {
+		return
+	}
+	at, ok := (*t).(*ArrayType)
+	if !ok || at.Type.Size() == 0 {
+		return
 	}
+	// Make a copy to avoid invalidating typeCache.
+	tt := *at
+	tt.Count = 0
+	*t = &tt
 }