mirror of
https://github.com/superseriousbusiness/gotosocial.git
synced 2024-11-27 14:16:39 +00:00
274 lines
6.3 KiB
Go
274 lines
6.3 KiB
Go
|
package exifcommon
|
||
|
|
||
|
import (
|
||
|
"bytes"
|
||
|
"math"
|
||
|
"reflect"
|
||
|
"time"
|
||
|
|
||
|
"encoding/binary"
|
||
|
|
||
|
"github.com/dsoprea/go-logging"
|
||
|
)
|
||
|
|
||
|
var (
|
||
|
typeEncodeLogger = log.NewLogger("exif.type_encode")
|
||
|
)
|
||
|
|
||
|
// EncodedData encapsulates the compound output of an encoding operation.
|
||
|
type EncodedData struct {
|
||
|
Type TagTypePrimitive
|
||
|
Encoded []byte
|
||
|
|
||
|
// TODO(dustin): Is this really necessary? We might have this just to correlate to the incoming stream format (raw bytes and a unit-count both for incoming and outgoing).
|
||
|
UnitCount uint32
|
||
|
}
|
||
|
|
||
|
// ValueEncoder knows how to encode values of every type to bytes.
|
||
|
type ValueEncoder struct {
|
||
|
byteOrder binary.ByteOrder
|
||
|
}
|
||
|
|
||
|
// NewValueEncoder returns a new ValueEncoder.
|
||
|
func NewValueEncoder(byteOrder binary.ByteOrder) *ValueEncoder {
|
||
|
return &ValueEncoder{
|
||
|
byteOrder: byteOrder,
|
||
|
}
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeBytes(value []uint8) (ed EncodedData, err error) {
|
||
|
ed.Type = TypeByte
|
||
|
ed.Encoded = []byte(value)
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeAscii(value string) (ed EncodedData, err error) {
|
||
|
ed.Type = TypeAscii
|
||
|
|
||
|
ed.Encoded = []byte(value)
|
||
|
ed.Encoded = append(ed.Encoded, 0)
|
||
|
|
||
|
ed.UnitCount = uint32(len(ed.Encoded))
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
// encodeAsciiNoNul returns a string encoded as a byte-string without a trailing
|
||
|
// NUL byte.
|
||
|
//
|
||
|
// Note that:
|
||
|
//
|
||
|
// 1. This type can not be automatically encoded using `Encode()`. The default
|
||
|
// mode is to encode *with* a trailing NUL byte using `encodeAscii`. Only
|
||
|
// certain undefined-type tags using an unterminated ASCII string and these
|
||
|
// are exceptional in nature.
|
||
|
//
|
||
|
// 2. The presence of this method allows us to completely test the complimentary
|
||
|
// no-nul parser.
|
||
|
//
|
||
|
func (ve *ValueEncoder) encodeAsciiNoNul(value string) (ed EncodedData, err error) {
|
||
|
ed.Type = TypeAsciiNoNul
|
||
|
ed.Encoded = []byte(value)
|
||
|
ed.UnitCount = uint32(len(ed.Encoded))
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeShorts(value []uint16) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
ed.Encoded = make([]byte, ed.UnitCount*2)
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
ve.byteOrder.PutUint16(ed.Encoded[i*2:(i+1)*2], value[i])
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeShort
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeLongs(value []uint32) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
ed.Encoded = make([]byte, ed.UnitCount*4)
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
ve.byteOrder.PutUint32(ed.Encoded[i*4:(i+1)*4], value[i])
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeLong
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeFloats(value []float32) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
ed.Encoded = make([]byte, ed.UnitCount*4)
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
ve.byteOrder.PutUint32(ed.Encoded[i*4:(i+1)*4], math.Float32bits(value[i]))
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeFloat
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeDoubles(value []float64) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
ed.Encoded = make([]byte, ed.UnitCount*8)
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
ve.byteOrder.PutUint64(ed.Encoded[i*8:(i+1)*8], math.Float64bits(value[i]))
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeDouble
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeRationals(value []Rational) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
ed.Encoded = make([]byte, ed.UnitCount*8)
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
ve.byteOrder.PutUint32(ed.Encoded[i*8+0:i*8+4], value[i].Numerator)
|
||
|
ve.byteOrder.PutUint32(ed.Encoded[i*8+4:i*8+8], value[i].Denominator)
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeRational
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeSignedLongs(value []int32) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
|
||
|
b := bytes.NewBuffer(make([]byte, 0, 8*ed.UnitCount))
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
err := binary.Write(b, ve.byteOrder, value[i])
|
||
|
log.PanicIf(err)
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeSignedLong
|
||
|
ed.Encoded = b.Bytes()
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
func (ve *ValueEncoder) encodeSignedRationals(value []SignedRational) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
ed.UnitCount = uint32(len(value))
|
||
|
|
||
|
b := bytes.NewBuffer(make([]byte, 0, 8*ed.UnitCount))
|
||
|
|
||
|
for i := uint32(0); i < ed.UnitCount; i++ {
|
||
|
err := binary.Write(b, ve.byteOrder, value[i].Numerator)
|
||
|
log.PanicIf(err)
|
||
|
|
||
|
err = binary.Write(b, ve.byteOrder, value[i].Denominator)
|
||
|
log.PanicIf(err)
|
||
|
}
|
||
|
|
||
|
ed.Type = TypeSignedRational
|
||
|
ed.Encoded = b.Bytes()
|
||
|
|
||
|
return ed, nil
|
||
|
}
|
||
|
|
||
|
// Encode returns bytes for the given value, infering type from the actual
|
||
|
// value. This does not support `TypeAsciiNoNull` (all strings are encoded as
|
||
|
// `TypeAscii`).
|
||
|
func (ve *ValueEncoder) Encode(value interface{}) (ed EncodedData, err error) {
|
||
|
defer func() {
|
||
|
if state := recover(); state != nil {
|
||
|
err = log.Wrap(state.(error))
|
||
|
}
|
||
|
}()
|
||
|
|
||
|
switch t := value.(type) {
|
||
|
case []byte:
|
||
|
ed, err = ve.encodeBytes(t)
|
||
|
log.PanicIf(err)
|
||
|
case string:
|
||
|
ed, err = ve.encodeAscii(t)
|
||
|
log.PanicIf(err)
|
||
|
case []uint16:
|
||
|
ed, err = ve.encodeShorts(t)
|
||
|
log.PanicIf(err)
|
||
|
case []uint32:
|
||
|
ed, err = ve.encodeLongs(t)
|
||
|
log.PanicIf(err)
|
||
|
case []float32:
|
||
|
ed, err = ve.encodeFloats(t)
|
||
|
log.PanicIf(err)
|
||
|
case []float64:
|
||
|
ed, err = ve.encodeDoubles(t)
|
||
|
log.PanicIf(err)
|
||
|
case []Rational:
|
||
|
ed, err = ve.encodeRationals(t)
|
||
|
log.PanicIf(err)
|
||
|
case []int32:
|
||
|
ed, err = ve.encodeSignedLongs(t)
|
||
|
log.PanicIf(err)
|
||
|
case []SignedRational:
|
||
|
ed, err = ve.encodeSignedRationals(t)
|
||
|
log.PanicIf(err)
|
||
|
case time.Time:
|
||
|
// For convenience, if the user doesn't want to deal with translation
|
||
|
// semantics with timestamps.
|
||
|
|
||
|
s := ExifFullTimestampString(t)
|
||
|
|
||
|
ed, err = ve.encodeAscii(s)
|
||
|
log.PanicIf(err)
|
||
|
default:
|
||
|
log.Panicf("value not encodable: [%s] [%v]", reflect.TypeOf(value), value)
|
||
|
}
|
||
|
|
||
|
return ed, nil
|
||
|
}
|