Added big data test and updating to reader

This commit is contained in:
obscuren 2014-07-29 23:31:33 +02:00
parent d1d2b660dc
commit 6e94c024e4
2 changed files with 87 additions and 67 deletions

View File

@ -55,8 +55,7 @@ func DecodeWithReader(reader *bytes.Buffer) interface{} {
return reader.Next(int(char - 0x80))
case char <= 0xbf:
buff := bytes.NewReader(reader.Next(int(char - 0xb8)))
length := ReadVarint(buff)
length := ReadVarInt(reader.Next(int(char - 0xb7)))
return reader.Next(int(length))
@ -72,76 +71,22 @@ func DecodeWithReader(reader *bytes.Buffer) interface{} {
}
return slice
case char <= 0xff:
length := ReadVarInt(reader.Next(int(char - 0xf7)))
for i := uint64(0); i < length; i++ {
obj := DecodeWithReader(reader)
if obj != nil {
slice = append(slice, obj)
} else {
break
}
}
default:
}
return slice
}
// TODO Use a bytes.Buffer instead of a raw byte slice.
// Cleaner code, and use draining instead of seeking the next bytes to read
func Decode(data []byte, pos uint64) (interface{}, uint64) {
var slice []interface{}
char := int(data[pos])
switch {
case char <= 0x7f:
return data[pos], pos + 1
case char <= 0xb7:
b := uint64(data[pos]) - 0x80
return data[pos+1 : pos+1+b], pos + 1 + b
case char <= 0xbf:
b := uint64(data[pos]) - 0xb7
b2 := ReadVarint(bytes.NewReader(data[pos+1 : pos+1+b]))
return data[pos+1+b : pos+1+b+b2], pos + 1 + b + b2
case char <= 0xf7:
b := uint64(data[pos]) - 0xc0
prevPos := pos
pos++
for i := uint64(0); i < b; {
var obj interface{}
// Get the next item in the data list and append it
obj, prevPos = Decode(data, pos)
slice = append(slice, obj)
// Increment i by the amount bytes read in the previous
// read
i += (prevPos - pos)
pos = prevPos
}
return slice, pos
case char <= 0xff:
l := uint64(data[pos]) - 0xf7
b := ReadVarint(bytes.NewReader(data[pos+1 : pos+1+l]))
pos = pos + l + 1
prevPos := b
for i := uint64(0); i < uint64(b); {
var obj interface{}
obj, prevPos = Decode(data, pos)
slice = append(slice, obj)
i += (prevPos - pos)
pos = prevPos
}
return slice, pos
default:
panic(fmt.Sprintf("byte not supported: %q", char))
}
return slice, 0
}
var (
directRlp = big.NewInt(0x7f)
numberRlp = big.NewInt(0xb7)
@ -223,3 +168,67 @@ func Encode(object interface{}) []byte {
return buff.Bytes()
}
// TODO Use a bytes.Buffer instead of a raw byte slice.
// Cleaner code, and use draining instead of seeking the next bytes to read
func Decode(data []byte, pos uint64) (interface{}, uint64) {
var slice []interface{}
char := int(data[pos])
switch {
case char <= 0x7f:
return data[pos], pos + 1
case char <= 0xb7:
b := uint64(data[pos]) - 0x80
return data[pos+1 : pos+1+b], pos + 1 + b
case char <= 0xbf:
b := uint64(data[pos]) - 0xb7
b2 := ReadVarInt(data[pos+1 : pos+1+b])
return data[pos+1+b : pos+1+b+b2], pos + 1 + b + b2
case char <= 0xf7:
b := uint64(data[pos]) - 0xc0
prevPos := pos
pos++
for i := uint64(0); i < b; {
var obj interface{}
// Get the next item in the data list and append it
obj, prevPos = Decode(data, pos)
slice = append(slice, obj)
// Increment i by the amount bytes read in the previous
// read
i += (prevPos - pos)
pos = prevPos
}
return slice, pos
case char <= 0xff:
l := uint64(data[pos]) - 0xf7
b := ReadVarInt(data[pos+1 : pos+1+l])
pos = pos + l + 1
prevPos := b
for i := uint64(0); i < uint64(b); {
var obj interface{}
obj, prevPos = Decode(data, pos)
slice = append(slice, obj)
i += (prevPos - pos)
pos = prevPos
}
return slice, pos
default:
panic(fmt.Sprintf("byte not supported: %q", char))
}
return slice, 0
}

View File

@ -44,6 +44,17 @@ func TestValueSlice(t *testing.T) {
}
}
func TestLargeData(t *testing.T) {
data := make([]byte, 100000)
enc := Encode(data)
value := NewValue(enc)
value.Decode()
if value.Len() != len(data) {
t.Error("Expected data to be", len(data), "got", value.Len())
}
}
func TestValue(t *testing.T) {
value := NewValueFromBytes([]byte("\xcd\x83dog\x83god\x83cat\x01"))
if value.Get(0).Str() != "dog" {