gob: decode into nil, this time for sure.

Yesterday's change was too simple-minded and failed if an
interface value was being discarded.  We need to parse the
data stream and remember any type information that arrives.

Also fix a minor bug when ignoring an interface: toss only what
we know about, not everything.

R=rsc
CC=golang-dev
https://golang.org/cl/4179045

src/pkg/gob/decode.go

@@ -481,6 +481,19 @@ func (dec *Decoder) ignoreStruct(engine *decEngine) (err os.Error) {
 	return nil
 }
 
+func (dec *Decoder) ignoreSingle(engine *decEngine) (err os.Error) {
+	defer catchError(&err)
+	state := newDecodeState(dec, &dec.buf)
+	state.fieldnum = singletonField
+	delta := int(state.decodeUint())
+	if delta != 0 {
+		errorf("gob decode: corrupted data: non-zero delta for singleton")
+	}
+	instr := &engine.instr[singletonField]
+	instr.op(instr, state, unsafe.Pointer(nil))
+	return nil
+}
+
 func (dec *Decoder) decodeArrayHelper(state *decodeState, p uintptr, elemOp decOp, elemWid uintptr, length, elemIndir int, ovfl os.ErrorString) {
 	instr := &decInstr{elemOp, 0, elemIndir, 0, ovfl}
 	for i := 0; i < length; i++ {
@@ -653,8 +666,8 @@ func (dec *Decoder) ignoreInterface(state *decodeState) {
 	if id < 0 {
 		error(dec.err)
 	}
-	// At this point, the decoder buffer contains the value. Just toss it.
-	state.b.Reset()
+	// At this point, the decoder buffer contains a delimited value. Just toss it.
+	state.b.Next(int(state.decodeUint()))
 }
 
 // Index by Go types.
@@ -901,6 +914,16 @@ func (dec *Decoder) compileSingle(remoteId typeId, rt reflect.Type) (engine *dec
 	return
 }
 
+func (dec *Decoder) compileIgnoreSingle(remoteId typeId) (engine *decEngine, err os.Error) {
+	engine = new(decEngine)
+	engine.instr = make([]decInstr, 1) // one item
+	op := dec.decIgnoreOpFor(remoteId)
+	ovfl := overflow(dec.typeString(remoteId))
+	engine.instr[0] = decInstr{op, 0, 0, 0, ovfl}
+	engine.numInstr = 1
+	return
+}
+
 // Is this an exported - upper case - name?
 func isExported(name string) bool {
 	rune, _ := utf8.DecodeRuneInString(name)
@@ -984,7 +1007,12 @@ func (dec *Decoder) getIgnoreEnginePtr(wireId typeId) (enginePtr **decEngine, er
 		// To handle recursive types, mark this engine as underway before compiling.
 		enginePtr = new(*decEngine)
 		dec.ignorerCache[wireId] = enginePtr
+		wire := dec.wireType[wireId]
+		if wire != nil && wire.StructT != nil {
 			*enginePtr, err = dec.compileDec(wireId, emptyStructType)
+		} else {
+			*enginePtr, err = dec.compileIgnoreSingle(wireId)
+		}
 		if err != nil {
 			dec.ignorerCache[wireId] = nil, false
 		}
@@ -993,6 +1021,10 @@ func (dec *Decoder) getIgnoreEnginePtr(wireId typeId) (enginePtr **decEngine, er
 }
 
 func (dec *Decoder) decodeValue(wireId typeId, val reflect.Value) os.Error {
+	// If the value is nil, it means we should just ignore this item.
+	if val == nil {
+		return dec.decodeIgnoredValue(wireId)
+	}
 	// Dereference down to the underlying struct type.
 	rt, indir := indirect(val.Type())
 	enginePtr, err := dec.getDecEnginePtr(wireId, rt)
@@ -1010,6 +1042,18 @@ func (dec *Decoder) decodeValue(wireId typeId, val reflect.Value) os.Error {
 	return dec.decodeSingle(engine, rt, uintptr(val.Addr()), indir)
 }
 
+func (dec *Decoder) decodeIgnoredValue(wireId typeId) os.Error {
+	enginePtr, err := dec.getIgnoreEnginePtr(wireId)
+	if err != nil {
+		return err
+	}
+	wire := dec.wireType[wireId]
+	if wire != nil && wire.StructT != nil {
+		return dec.ignoreStruct(*enginePtr)
+	}
+	return dec.ignoreSingle(*enginePtr)
+}
+
 func init() {
 	var iop, uop decOp
 	switch reflect.Typeof(int(0)).Bits() {

src/pkg/gob/decoder.go

@@ -183,13 +183,9 @@ func (dec *Decoder) DecodeValue(value reflect.Value) os.Error {
 	dec.buf.Reset() // In case data lingers from previous invocation.
 	dec.err = nil
 	id := dec.decodeTypeSequence(false)
-	if id >= 0 {
-		// A nil value means "ignore the data".  Since it's already read into
-		// the decoder's buffer, all we need to do is not bother to decode it.
-		if value != nil {
+	if dec.err == nil {
 		dec.err = dec.decodeValue(id, value)
 	}
-	}
 	return dec.err
 }
 

src/pkg/gob/encoder_test.go

@@ -6,6 +6,7 @@ package gob
 
 import (
 	"bytes"
+	"fmt"
 	"io"
 	"os"
 	"reflect"
@@ -120,7 +121,7 @@ func corruptDataCheck(s string, err os.Error, t *testing.T) {
 	dec := NewDecoder(b)
 	err1 := dec.Decode(new(ET2))
 	if err1 != err {
-		t.Error("expected error", err, "got", err1)
+		t.Errorf("from %q expected error %s; got %s", s, err, err1)
 	}
 }
 
@@ -384,54 +385,72 @@ func TestInterfaceIndirect(t *testing.T) {
 	}
 }
 
-func TestDecodeIntoEmptyStruct(t *testing.T) {
-	type Empty struct{}
-	empty := &Empty{}
-	b := new(bytes.Buffer)
-	enc := NewEncoder(b)
-	err := enc.Encode(&struct{ A int }{23})
-	if err != nil {
-		t.Fatal("encode error:", err)
-	}
-	dec := NewDecoder(b)
-	err = dec.Decode(empty)
-	if err != nil {
-		t.Fatal("encode error:", err)
-	}
+// Now follow various tests that decode into things that can't represent the
+// encoded value, all of which should be legal.
+
+// Also, when the ignored object contains an interface value, it may define
+// types. Make sure that skipping the value still defines the types by using
+// the encoder/decoder pair to send a value afterwards.  If an interface
+// is sent, its type in the test is always NewType0, so this checks that the
+// encoder and decoder don't skew with respect to type definitions.
+
+type Struct0 struct {
+	I interface{}
 }
 
-func TestStructDecodeIntoNil(t *testing.T) {
-	nonempty := &struct{ A int }{23}
-	b := new(bytes.Buffer)
-	enc := NewEncoder(b)
-	err := enc.Encode(nonempty)
-	if err != nil {
-		t.Fatal("encode error:", err)
-	}
-	dec := NewDecoder(b)
-	err = dec.Decode(nil)
-	if err != nil {
-		t.Fatal("encode error:", err)
-	}
-	if b.Len() != 0 {
-		t.Fatalf("%d bytes remain after decode", b.Len())
-	}
+type NewType0 struct {
+	S string
 }
 
-func TestSingletonDecodeIntoNil(t *testing.T) {
+type ignoreTest struct {
+	in, out interface{}
+}
+
+var ignoreTests = []ignoreTest{
+	// Decode normal struct into an empty struct
+	{&struct{ A int }{23}, &struct{}{}},
+	// Decode normal struct into a nil.
+	{&struct{ A int }{23}, nil},
+	// Decode singleton string into a nil.
+	{"hello, world", nil},
+	// Decode singleton slice into a nil.
+	{[]int{1, 2, 3, 4}, nil},
+	// Decode struct containing an interface into a nil.
+	{&Struct0{&NewType0{"value0"}}, nil},
+	// Decode singleton slice of interfaces into a nil.
+	{[]interface{}{"hi", &NewType0{"value1"}, 23}, nil},
+}
+
+func TestDecodeIntoNothing(t *testing.T) {
+	Register(new(NewType0))
+	for i, test := range ignoreTests {
 		b := new(bytes.Buffer)
 		enc := NewEncoder(b)
-	err := enc.Encode("hello world")
+		err := enc.Encode(test.in)
 		if err != nil {
-		t.Fatal("encode error:", err)
+			t.Errorf("%d: encode error %s:", i, err)
+			continue
 		}
 		dec := NewDecoder(b)
-	err = dec.Decode(nil)
+		err = dec.Decode(test.out)
 		if err != nil {
-		t.Fatal("encode error:", err)
+			t.Errorf("%d: decode error: %s", i, err)
+			continue
+		}
+		// Now see if the encoder and decoder are in a consistent state.
+		str := fmt.Sprintf("Value %d", i)
+		err = enc.Encode(&NewType0{str})
+		if err != nil {
+			t.Fatalf("%d: NewType0 encode error: %s", i, err)
+		}
+		ns := new(NewType0)
+		err = dec.Decode(ns)
+		if err != nil {
+			t.Fatalf("%d: NewType0 decode error: %s", i, err)
+		}
+		if ns.S != str {
+			t.Fatalf("%d: expected %q got %q", i, str, ns.S)
 		}
-	if b.Len() != 0 {
-		t.Fatalf("%d bytes remain after decode", b.Len())
 	}
 }