Sometimes, you want to include data directly in your executable file, but
you don't want to translate that data into Rust code that does the static
initialization. This is very useful in embedded contexts, or if you have
some (usually relatively small) data that will always be needed, and you
don't want to deal with loading it from the filesystem and distributing it
as a separate file.
The Rust standard library (and core library) contains
include_bytes
for this purpose. This macro will give you a static reference to a binary array
containing the data from a
file: that is, a &'static [u8; N]
.
However, if you want to use your static data, you often want it to be of a
particular type, not just a &[u8]
. For example, you may know that your
included file is a sequence of f64
s, or a UTF-32 file, or of some
custom type. This crate provides macros for typed compile-time data
includes. This is provided by two main macros:
include_data
- outputs any type which is soundinclude_slice
- outputs a&'static [T]
slice for anyT
for which this is sound
This crate is no_std
and also no-alloc
.
This library will work out-of-the-box with any type that implements
bytemuck::AnyBitPattern
.
This includes:
- Primitive numerical types (
u16
,i32
,f64
, etc.) - Arrays of primitive numerical types (e.g.
[f32; N]
)
For example:
static MY_INTEGER: i32 = include_data!("../tests/test_data/file_exactly_4_bytes_long");
static SOME_TEXT: &[u32] = include_slice!(u32, "../tests/test_data/some_utf-32_file");
const FOUR_BYTES: [u8; 4] = include_data!("../tests/test_data/file_exactly_4_bytes_long");
Note that include_data
can assign to const
, while include_slice
cannot.
Aliases are provided for include_slice
for primitive number types, using
them is a matter of personal preference. For example:
static SOME_TEXT: &[u32] = include_u32s!("../tests/test_data/some_utf-32_file");
You can include data in any custom type you like. The best way of doing this
is if your custom type satisfies the requirements for
bytemuck::AnyBitPattern
,
in which case you can simply use include_data
.
#[repr(C)]
#[derive(Copy, Clone)]
struct Foo {
integer: u16,
pair: [u8; 2],
}
// Safety: the requirements for `AnyBitPattern` have been manually checked.
unsafe impl bytemuck::Zeroable for Foo {}
unsafe impl bytemuck::AnyBitPattern for Foo {}
static FOO_DATA: Foo = include_data!("../tests/test_data/file_exactly_4_bytes_long");
Alternatively, if your type cannot implement bytemuck::AnyBitPattern
(especially
if it is a foreign type over which you have no control), include_unsafe
can be used. In this case, you must guarantee that the file included is
valid for the target type. This may depend on host platform, compiler
version, and compiler profile (amongst other things): recall that Rust does
not have a stable ABI. Clearly, this is very unsafe and should be
avoided if possible.
#[repr(C)]
struct StructWithPadding {
byte: u8,
two_bytes: u16,
}
// Safety: we guarantee that the included file contains bytes which are
// a valid bit-pattern for our struct, when compiled on this host.
static BAR_DATA: StructWithPadding = unsafe { include_unsafe!("../tests/test_data/file_exactly_4_bytes_long") };
All macros exported by this crate are safe, except include_unsafe
(assuming, of course, that implementations of bytemuck::AnyBitPattern
are sound). If
the input file size does not match the target type (or is not divisible by
it, in the case of slices) or the file cannot be read, compilation will
fail.
include_unsafe
is very unsafe and should only be used with great care.
See the
documentation
for full details.
The interpretation of multi-byte sequences depends on a machine's endianness. In the case of these macros, multi-byte sequences will be interpreted into types according to the endianness of the compilation target, not the compilation host machine.
The interpreation of paths passed to these macros is host-platform specific
and identical to that of
include_bytes
.
The Minimum Supported Rust Version is 1.64.0.
Note that this crate is tested against a pinned version of the compiler,
simply because many tests check exact error messages. The current pinned
version for testing purposes can be found in rust-toolchain.toml
.
Depending on what you're trying to achieve, this crate might not be the best choice. Here are a few alternatives which may be more appropriate depending on the situation:
static_toml
: if the data you're including fits within atoml
spec, this crate parses it at compile time and includes the result as static, type-safe, data.const_gen
: tool that helps you usebuild.rs
to do compile-time code generation of constant values. More complicated and verbose thaninclude_data
, but also more flexible.- The standard library has
OnceLock
andLazyLock
both of which can be used to assign complex values tostatic
s, but do so at runtime and with a non-zero runtime cost. If your type needs runtime construction, these are very good chocies butinclude_data
is cheaper for "simple typed data" values.
If you know of any others, please drop an issue or open a PR.
The techniques used by this crate were published in a blog post by Jack Wrenn. Some of those techniques were original to Jack, while others were found in forum threads linked from that post. Please do reach out if you are somebody involved with these discussions, or have any prior work in this area. I am also grateful to Jack for comments on an earlier draft of this crate.