RecognizingDecoder didn't handle ambiguous input

YAML is not guaranteed to be recognizable, so we need to bump JSON and
protobuf above it in the decoding order. Add a unit test.
This commit is contained in:
Clayton Coleman 2016-04-23 14:54:54 -04:00
parent 7e1089bb75
commit 4ad5565c41
No known key found for this signature in database
GPG Key ID: 3D16906B4F1C5CB3
4 changed files with 101 additions and 36 deletions

View File

@ -17,6 +17,7 @@ limitations under the License.
package api_test
import (
"bytes"
"encoding/hex"
"math/rand"
"testing"
@ -41,6 +42,28 @@ func init() {
})
}
func TestUniversalDeserializer(t *testing.T) {
expected := &v1.Pod{ObjectMeta: v1.ObjectMeta{Name: "test"}}
d := api.Codecs.UniversalDeserializer()
for _, mediaType := range []string{"application/json", "application/yaml", "application/vnd.kubernetes.protobuf"} {
e, ok := api.Codecs.SerializerForMediaType(mediaType, nil)
if !ok {
t.Fatal(mediaType)
}
buf := &bytes.Buffer{}
if err := e.EncodeToStream(expected, buf); err != nil {
t.Fatalf("%s: %v", mediaType, err)
}
obj, _, err := d.Decode(buf.Bytes(), &unversioned.GroupVersionKind{Kind: "Pod", Version: "v1"}, nil)
if err != nil {
t.Fatalf("%s: %v", mediaType, err)
}
if !api.Semantic.DeepEqual(expected, obj) {
t.Fatalf("%s: %#v", mediaType, obj)
}
}
}
func TestProtobufRoundTrip(t *testing.T) {
obj := &v1.Pod{}
apitesting.FuzzerFor(t, v1.SchemeGroupVersion, rand.NewSource(benchmarkSeed)).Fuzz(obj)

View File

@ -186,12 +186,13 @@ func (s *Serializer) EncodeToStream(obj runtime.Object, w io.Writer, overrides .
}
// RecognizesData implements the RecognizingDecoder interface.
func (s *Serializer) RecognizesData(peek io.Reader) (bool, error) {
_, ok := utilyaml.GuessJSONStream(peek, 2048)
func (s *Serializer) RecognizesData(peek io.Reader) (ok, unknown bool, err error) {
if s.yaml {
return !ok, nil
// we could potentially look for '---'
return false, true, nil
}
return ok, nil
_, ok = utilyaml.GuessJSONStream(peek, 2048)
return ok, false, nil
}
// EncodesAsText returns true because both JSON and YAML are considered textual representations

View File

@ -419,12 +419,6 @@ func (s *RawSerializer) EncodeToStream(obj runtime.Object, w io.Writer, override
}
}
// RecognizesData implements the RecognizingDecoder interface - objects encoded with this serializer
// have no innate identifying information and so cannot be recognized.
func (s *RawSerializer) RecognizesData(peek io.Reader) (bool, error) {
return false, nil
}
var LengthDelimitedFramer = lengthDelimitedFramer{}
type lengthDelimitedFramer struct{}

View File

@ -27,51 +27,98 @@ import (
type RecognizingDecoder interface {
runtime.Decoder
RecognizesData(peek io.Reader) (bool, error)
// RecognizesData should return true if the input provided in the provided reader
// belongs to this decoder, or an error if the data could not be read or is ambiguous.
// Unknown is true if the data could not be determined to match the decoder type.
// Decoders should assume that they can read as much of peek as they need (as the caller
// provides) and may return unknown if the data provided is not sufficient to make a
// a determination. When peek returns EOF that may mean the end of the input or the
// end of buffered input - recognizers should return the best guess at that time.
RecognizesData(peek io.Reader) (ok, unknown bool, err error)
}
// NewDecoder creates a decoder that will attempt multiple decoders in an order defined
// by:
//
// 1. The decoder implements RecognizingDecoder and identifies the data
// 2. All other decoders, and any decoder that returned true for unknown.
//
// The order passed to the constructor is preserved within those priorities.
func NewDecoder(decoders ...runtime.Decoder) runtime.Decoder {
recognizing, blind := []RecognizingDecoder{}, []runtime.Decoder{}
for _, d := range decoders {
if r, ok := d.(RecognizingDecoder); ok {
recognizing = append(recognizing, r)
} else {
blind = append(blind, d)
}
}
return &decoder{
recognizing: recognizing,
blind: blind,
decoders: decoders,
}
}
type decoder struct {
recognizing []RecognizingDecoder
blind []runtime.Decoder
decoders []runtime.Decoder
}
var _ RecognizingDecoder = &decoder{}
func (d *decoder) RecognizesData(peek io.Reader) (bool, bool, error) {
var (
lastErr error
anyUnknown bool
)
data, _ := bufio.NewReaderSize(peek, 1024).Peek(1024)
for _, r := range d.decoders {
switch t := r.(type) {
case RecognizingDecoder:
ok, unknown, err := t.RecognizesData(bytes.NewBuffer(data))
if err != nil {
lastErr = err
continue
}
anyUnknown = anyUnknown || unknown
if !ok {
continue
}
return true, false, nil
}
}
return false, anyUnknown, lastErr
}
func (d *decoder) Decode(data []byte, gvk *unversioned.GroupVersionKind, into runtime.Object) (runtime.Object, *unversioned.GroupVersionKind, error) {
var lastErr error
for _, r := range d.recognizing {
buf := bytes.NewBuffer(data)
ok, err := r.RecognizesData(buf)
if err != nil {
lastErr = err
continue
var (
lastErr error
skipped []runtime.Decoder
)
// try recognizers, record any decoders we need to give a chance later
for _, r := range d.decoders {
switch t := r.(type) {
case RecognizingDecoder:
buf := bytes.NewBuffer(data)
ok, unknown, err := t.RecognizesData(buf)
if err != nil {
lastErr = err
continue
}
if unknown {
skipped = append(skipped, t)
continue
}
if !ok {
continue
}
return r.Decode(data, gvk, into)
default:
skipped = append(skipped, t)
}
if !ok {
continue
}
return r.Decode(data, gvk, into)
}
for _, d := range d.blind {
out, actual, err := d.Decode(data, gvk, into)
// try recognizers that returned unknown or didn't recognize their data
for _, r := range skipped {
out, actual, err := r.Decode(data, gvk, into)
if err != nil {
lastErr = err
continue
}
return out, actual, nil
}
if lastErr == nil {
lastErr = fmt.Errorf("no serialization format matched the provided data")
}