[][src]Crate restruct_derive

restruct is used to interpret binary data stored in files or other sources or convert between C structs and Rust types and when using a parser-generator is considered disproportionate. It is a brainchild of Python's struct-module.

The library uses Format Strings as compact descriptions of the binary data and the intended conversion to/from Rust-types. The Format Strings are interpreted at compile-time to generate a type whose functions can be used to convert between unstructured and structured data.

// Generate a parser in little-endian for two 32bit integers, a float and a bool.
struct FooParser;

// Pack a tuple of two integers, a float an a bool into a [u8; _]-buffer.
let packed = FooParser::pack((1, 2, 3.0, false));
assert_eq!(packed.len(), FooParser::SIZE);
// Packing and unpacking can't fail at runtime.
let unpacked = FooParser::unpack(packed);
assert_eq!(unpacked, (1, 2, 3.0, false));

// Packing and unpacking is const
const FOOBAR: [u8; FooParser::SIZE] = FooParser::pack((987, 412, std::f32::consts::PI, false));
const BARFOO: <FooParser as restruct::Struct>::Unpacked = FooParser::unpack(FOOBAR);
assert_eq!(BARFOO, (987, 412, std::f32::consts::PI, false));

// Read/Write data in the given format to a `io::Read/Write`
let mut buffer = Vec::new();
let inp = (123, 456, -2.521, false);
FooParser::write_to(inp, &mut buffer)?;
let outp = FooParser::read_from(&mut &buffer[..])?;
assert_eq!(outp, inp);
This example is not tested
// As the packing- and unpacking-functions are const, they can initialize other constants.
// Read some file from disk and directly unpack it into a const during compilation.
struct Tea;

const TEAPOT: <Tea as restruct::Struct>::Unpacked = Tea::unpack(*include_bytes!("teapot.bin"));
const TEAPOT_ACTIVE: bool = TEAPOT.3;

The type-layout is determined entireley at compile-time. The "packed" representation is always a fixed-length [u8; ...]-array. The "unpacked" representation is a (possibly nested) tuple of primitive types.

It is not possible to describe variable-sized types like String or Vec<T> (see the Examples-sections for more information) or to generate/modify parsers at runtime; parsers can be generated via macros (including proc-macros), though.

The conversion functions are const and may therefor be used in a const-context. As long as endianess does not need to be converted and copying can be elided, packing and unpacking should usually be free of any runtime cost.

Note that this crate is currently nightly-only; the following feature-gates need to be unsealed:



Parsers are derived on types using the Struct-proc-macro from the restruct_derive crate. The Format String is passed via the fmt-attribute.

struct FrameHeader;

The fmt-attribute can be used multiple times and all fragments are concatenated before being interpreted.

The proc-macro will add the following items to the given type, among others:

  • An implementation of restruct::Struct, which will hold the type aliases for the packed and unpacked representation. For example, <Foo as restruct::Struct>::Packed will be a type alias for [u8; N], where N is some const, and ...::Unpacked will be a tuple.
  • An associated constant SIZE, which gives the size in bytes of the packed form.
  • An associated constant FIELDS, an array of tuples of the form (&'static str, usize, usize, usize) for the name of the type, the offset, the alignment and the total size of each field.
  • A const fn pack() to convert from unpacked (tuple) into packed (array) form.
  • A const fn unpack() to convert from packed (array) into unpacked (tuple) form.
  • A fn unpack_slice() that takes a &[u8]-slice and unpacks it's content. This method will panic if the given slice is smaller than Self::SIZE.
  • A fn read_from() to read one unpacked instance from an any io::Read.
  • A fn write_to() to write one unpacked instance to any io::Write.
  • A unsafe fn from_raw<T>(ptr: *const T) to read one unpacked instance from a raw pointer.
  • An implementation of std::fmt::Debug.

Format Strings

Format Strings are used to specify the exact byte-layout when packing and unpacking data.

The first character in a Format String may be used to control the byte order, size and internal alignment for all following Format Characters. For example, a Format String starting with "<..." specifies that all following Format Characters shall be interpreted as little-endian, shall use primitive types and shall not add alignment while packing/unpacking. This can be set only once in a Format String.

Zero or more Format Characters may be given to specify the type of data being packed/unpacked. Format Characters map to type aliases defined by the libc crate when using native mode (@) or primitive types when using standard mode (=, <, > and !). For example, "@l" refers to libc::c_long, which is a type alias for either i32 or i64 depending on the current platform; "=l" always refers to i32 and so does "<l", ">l" and "!l".

Byte Order, Size, and Alignment

CharacterByte orderSizeAlignment
!network (= big-endian)standardnone

If the first character is not one of these, @ is assumed.

Alignment between types is added only in native mode (@). For example, the Format String "@bL" (usually) describes a (i8, u64), which will result in a [u8; 16] when packed: 1 byte for the i8, seven alignment bytes and then eight bytes for the u64. Alignment is never added at the start or end of the packed data; add a type with a repeat count of zero to add alignment for that type.

As a general rule, you should use standard types when dealing with data from IO (e.g. file-formats, protocols, anything persisted and transfered to other platforms, etc.) and native types when reading data structures from memory.

Format Characters

FormatNative typeStandard type
xno valueno value
s[u8; _][u8; _]
`ident`<ident as restruct::Struct>::Packed<ident as restruct::Struct>::Packed

A Format Character may be preceded by an repeat count. For example, the format string "3x4h2`Foo`" means exactly the same as "xxx hhhh `Foo` `Foo`".

Whitespace characters between formats are ignored; a count and its format must not contain whitespace.

Native types are indirected via the libc crate to Rust's primitive-types. Therefor libc must be available in the final crate when using native Format Strings. See the Examples-section for caveats.

For the s Format Character, the count is interpreted as the length of a [u8; _]-array, not a repeat count like for the other format characters. For example, "3s?" means ([u8; 3], bool) while "3f?" means (f32, f32, f32, bool).

For the ? Format Character, values not equal to 0 are interpreted as true when unpacking. When packing a bool, true is represented as 1, false as 0; it's size is alway one byte.

The special Format "`...`" allows to refer to another type which was derived using this crate. In it's packed form, a nested tuple is expected. For example, after deriving a type Foo with Format String "<b2i", another Format String on type Bar can refer to this as "<?2`Foo`", resulting in Bar expectecting a packed type (bool, (i8, i32, i32), (i8, i32, i32)).

The x Format Character denotes padding bytes. While they contribute to the size of the packed form, they are not present in the unpacked representation. For example, "?2x?" will be a (bool, bool) in unpacked and a [u8; 4] in packed form. Padding bytes are always set to 0 in the packed form.

The repeat count 0 has special meaning in the sense that the field will not be present in the unpacked representation and only it's alignment contributes to the size of the packed representation (if alignment is considered at all, see above). For example, the Format String "b0q" describes a (i8, ) while the end of the packed representation is aligned to a i64; the packed representation is therefor a [u8; 8]. Using a repeat count of 0 is effectively a no-op when using Format Strings where alignment is not taken into account.


Packing three integers using standard sizes in big-endian:

#[fmt = ">2hl"]
struct Foobar;

// The derived type implements std::fmt::Debug

let input = (1, 2, 3);
let expected = [0, 1, 0, 2, 0, 0, 0, 3];
let packed = Foobar::pack(input);
assert_eq!(packed, expected);
let unpacked = Foobar::unpack(packed);
assert_eq!(unpacked, input);

The Format String can passed in multiple parts, simplifying construction by macros:

#[fmt = ">"]
#[fmt = "2h"]
#[fmt = "10x"]
#[fmt = "3i"]
struct Foobar;

Slices can be unpacked at the cost of a copy (which may get elided):

#[fmt = ">2h"]
struct Foobar;

let buf = vec![0, 1, 0, 2, 255, 255];
let unpacked = Foobar::unpack_slice(&buf);
assert_eq!(unpacked, (1, 2));

The derived types can be referred to via the Struct trait:

use std::io::{self, Read};

#[fmt = ">2hl"]
struct FoobarHeader;

impl FoobarHeader {
    pub fn read_header<R>(r: &mut R) -> io::Result<<Self as restruct::Struct>::Unpacked>
     where R: io::Read

As packing and unpacking is const, these functions can be used to initialize other constants.

This example is not tested
#[fmt = ">2hl"]
struct Header;

const DEFAULT_HEADER: [u8; Header::SIZE] = Header::pack((0x0001, 0xff00, 0xdeadc0de));
const HEADER: <Header as restruct::Struct>::Unpacked = Header::unpack(*include_bytes!("header.bin"));

Format Strings always describe fixed-sized data stuctures. When dealing when variable-sized formats, two steps are necessary:

use std::io::{self, Write};

#[fmt = "<IN"]
struct Frame;

impl Frame {
    /// Read one frame from the given reader, returning it's payload
    pub fn read<R>(r: &mut R) -> io::Result<Vec<u8>>
     where R: io::Read {
        // Read the fixed-size header
        let (magic, size) = Frame::read_from(r)?;
        if magic != 0xdeadc0de {
            panic!("Unknown frame-format!");
        // We know the size of the now following data
        let mut buf = vec![0; size];
        r.read_exact(&mut buf).and(Ok(buf))

    /// Write the given data as a new frame to the given writer.
    pub fn write<W>(buf: &[u8], w: &mut W) -> io::Result<()>
     where W: io::Write {
        // Write the header
        Frame::write_to((0xdeadc0de, buf.len()), w)?;
        // ... and the rest

let mut buf = Vec::new();
let content = String::from("The quick brown fox");
Frame::write(content.as_ref(), &mut buf).expect("Writing failed");
// ...
let new_buf = Frame::read(&mut buf.as_slice()).expect("Reading failed");
let new_content = String::from_utf8(new_buf).expect("UTF-8 decoding failed");
assert_eq!(new_content, content);

Alignment rules apply in native mode:

#[fmt = "b0ib"]
struct Foobar;
// The second `i8` will be aligned to the boundary of an `libc::c_int`, the final
// size is therefor larger than two bytes.
assert!(Foobar::SIZE > 2);

#[fmt = "b0Q"]
struct Barfoo;
// The end of the packed representation is aligned to a `libc::c_ulonglong`, which
// means it will be 8 bytes in total.
assert_eq!(Barfoo::SIZE, 8);

Formats can refer to previous definitions:

#[fmt = "=i3s"]
struct Foo;

#[fmt = "=2`Foo`"]
struct Bar;

assert_eq!(Bar::SIZE, Foo::SIZE * 2);
assert_eq!(Bar::pack(((1, [0, 1, 2]), (2, [10, 20, 30]))).len(), Bar::SIZE);

When using native types, caveats may appear regarding the actual type used:

This example is not tested
extern "C" {
    fn read_header() -> *const u8;

#[fmt = "@Lhb"]
struct Header;

impl Header {
    pub fn read_magic() -> u64 {
        let header = unsafe { Self::from_raw(read_header()) };

The read_magic() function is defined to return a u64, which needs to match the "L" used in the Format String. This will work fine on platforms where c_ulong is a u64 but fail to compile e.g. on i586-platforms where c_ulong is a u32. Either use std::convert::TryFrom or make sure to use the type aliases from libc when using native mode.

Also note that the "@...b" in the Format String above is aliased via libc::c_char; it resolves to i8 on x86-platforms but u8 on ARM because c_char is unsigned on that platform. A line like header.2 < 0 will - rightfully so - cause a compile-error on ARM.

When converting from native structs, you must be sure that your layout description matches the actual layout:

This example is not tested
#[fmt = "@2dl"]
struct Header;

let head = unsafe { Header::from_raw(...) };

Let assume that the C-struct we try to match above uses int32_t as it's third element. The layout above will match on 32bit-platforms where "@...l" is i32. On 64bit-platforms however "@...l" is i64, so from_raw() will cause an out-of-bounds memory access by four bytes on those platforms! The correct Format String would have been "@2di".

Derive Macros


Derive packing/unpacking on a given type. See the main documentation on this crate for details.