Better binary packing for Go
Go
Switch branches/tags
Nothing to show
Clone or download
mewmew and lunixbochs Fix conversion of signed integers (#58)
* add test case for invalid conversion of signed integer

When converting signed 32-bit integers to 64-bit
integers, the sign is lost, and negative values are
converted to their unsigned counterpart.

E.g. converting -1 from int32 to int64 results in
4294967295, when the correct result should be -1.

* add expected and got output for failed test cases
* fix conversion of signed integers when unpacking from small to large bit-size
* add test case for down conversion of signed integer
Latest commit 02e4c2a Apr 8, 2018

README.md

Build Status

struc

Struc exists to pack and unpack C-style structures from bytes, which is useful for binary files and network protocols. It could be considered an alternative to encoding/binary, which requires massive boilerplate for some similar operations.

Take a look at an example comparing struc and encoding/binary

Struc considers usability first. That said, it does cache reflection data and aims to be competitive with encoding/binary struct packing in every way, including performance.

Example struct

type Example struct {
    Var   int `struc:"int32,sizeof=Str"`
    Str   string
    Weird []byte `struc:"[8]int64"`
    Var   []int `struc:"[]int32,little"`
}

Struct tag format

  • Var []int `struc:"[]int32,little,sizeof=StringField"` will pack Var as a slice of little-endian int32, and link it as the size of StringField.
  • sizeof=: Indicates this field is a number used to track the length of a another field. sizeof fields are automatically updated on Pack() based on the current length of the tracked field, and are used to size the target field during Unpack().
  • Bare values will be parsed as type and endianness.

Endian formats

  • big (default)
  • little

Recognized types

  • pad - this type ignores field contents and is backed by a [length]byte containing nulls
  • bool
  • byte
  • int8, uint8
  • int16, uint16
  • int32, uint32
  • int64, uint64
  • float32
  • float64

Types can be indicated as arrays/slices using [] syntax. Example: []int64, [8]int32.

Bare slice types (those with no [size]) must have a linked Sizeof field.

Private fields are ignored when packing and unpacking.

Example code

package main

import (
    "bytes"
    "github.com/lunixbochs/struc"
)

type Example struct {
    A int `struc:"big"`

    // B will be encoded/decoded as a 16-bit int (a "short")
    // but is stored as a native int in the struct
    B int `struc:"int16"`

    // the sizeof key links a buffer's size to any int field
    Size int `struc:"int8,little,sizeof=Str"`
    Str  string

    // you can get freaky if you want
    Str2 string `struc:"[5]int64"`
}

func main() {
    var buf bytes.Buffer
    t := &Example{1, 2, 0, "test", "test2"}
    err := struc.Pack(&buf, t)
    o := &Example{}
    err = struc.Unpack(&buf, o)
}

Benchmark

BenchmarkEncode uses struc. Stdlib benchmarks use equivalent encoding/binary code. Manual encodes without any reflection, and should be considered an upper bound on performance (which generated code based on struc definitions should be able to achieve).

BenchmarkEncode        1000000   1265 ns/op
BenchmarkStdlibEncode  1000000   1855 ns/op
BenchmarkManualEncode  5000000    284 ns/op
BenchmarkDecode        1000000   1259 ns/op
BenchmarkStdlibDecode  1000000   1656 ns/op
BenchmarkManualDecode  20000000  89.0 ns/op