Function std::intrinsics::transmute

pub const unsafe extern "rust-intrinsic" fn transmute<T, U>(e: T) -> U

Reinterprets the bits of a value of one type as another type.

Both types must have the same size. Neither the original, nor the result, may be an invalid value.

transmute is semantically equivalent to a bitwise move of one type into another. It copies the bits from the source value into the destination value, then forgets the original. It’s equivalent to C’s memcpy under the hood, just like transmute_copy.

Because transmute is a by-value operation, alignment of the transmuted values themselves is not a concern. As with any other function, the compiler already ensures both T and U are properly aligned. However, when transmuting values that point elsewhere (such as pointers, references, boxes…), the caller has to ensure proper alignment of the pointed-to values.

transmute is incredibly unsafe. There are a vast number of ways to cause undefined behavior with this function. transmute should be the absolute last resort.

Transmuting pointers to integers in a const context is undefined behavior. Any attempt to use the resulting value for integer operations will abort const-evaluation.

The nomicon has additional documentation.

Examples

There are a few things that transmute is really useful for.

Turning a pointer into a function pointer. This is not portable to machines where function pointers and data pointers have different sizes.

fn foo() -> i32 {
    0
}
let pointer = foo as *const ();
let function = unsafe {
    std::mem::transmute::<*const (), fn() -> i32>(pointer)
};
assert_eq!(function(), 0);

Extending a lifetime, or shortening an invariant lifetime. This is advanced, very unsafe Rust!

struct R<'a>(&'a i32);
unsafe fn extend_lifetime<'b>(r: R<'b>) -> R<'static> {
    std::mem::transmute::<R<'b>, R<'static>>(r)
}

unsafe fn shorten_invariant_lifetime<'b, 'c>(r: &'b mut R<'static>)
                                             -> &'b mut R<'c> {
    std::mem::transmute::<&'b mut R<'static>, &'b mut R<'c>>(r)
}

Alternatives

Don’t despair: many uses of transmute can be achieved through other means. Below are common applications of transmute which can be replaced with safer constructs.

Turning raw bytes (&[u8]) into u32, f64, etc.:

let raw_bytes = [0x78, 0x56, 0x34, 0x12];

let num = unsafe {
    std::mem::transmute::<[u8; 4], u32>(raw_bytes)
};

// use `u32::from_ne_bytes` instead
let num = u32::from_ne_bytes(raw_bytes);
// or use `u32::from_le_bytes` or `u32::from_be_bytes` to specify the endianness
let num = u32::from_le_bytes(raw_bytes);
assert_eq!(num, 0x12345678);
let num = u32::from_be_bytes(raw_bytes);
assert_eq!(num, 0x78563412);

Turning a pointer into a usize:

let ptr = &0;
let ptr_num_transmute = unsafe {
    std::mem::transmute::<&i32, usize>(ptr)
};

// Use an `as` cast instead
let ptr_num_cast = ptr as *const i32 as usize;

Turning a *mut T into an &mut T:

let ptr: *mut i32 = &mut 0;
let ref_transmuted = unsafe {
    std::mem::transmute::<*mut i32, &mut i32>(ptr)
};

// Use a reborrow instead
let ref_casted = unsafe { &mut *ptr };

Turning an &mut T into an &mut U:

let ptr = &mut 0;
let val_transmuted = unsafe {
    std::mem::transmute::<&mut i32, &mut u32>(ptr)
};

// Now, put together `as` and reborrowing - note the chaining of `as`
// `as` is not transitive
let val_casts = unsafe { &mut *(ptr as *mut i32 as *mut u32) };

Turning an &str into a &[u8]:

// this is not a good way to do this.
let slice = unsafe { std::mem::transmute::<&str, &[u8]>("Rust") };
assert_eq!(slice, &[82, 117, 115, 116]);

// You could use `str::as_bytes`
let slice = "Rust".as_bytes();
assert_eq!(slice, &[82, 117, 115, 116]);

// Or, just use a byte string, if you have control over the string
// literal
assert_eq!(b"Rust", &[82, 117, 115, 116]);

Turning a Vec<&T> into a Vec<Option<&T>>.

To transmute the inner type of the contents of a container, you must make sure to not violate any of the container’s invariants. For Vec, this means that both the size and alignment of the inner types have to match. Other containers might rely on the size of the type, alignment, or even the TypeId, in which case transmuting wouldn’t be possible at all without violating the container invariants.

let store = [0, 1, 2, 3];
let v_orig = store.iter().collect::<Vec<&i32>>();

// clone the vector as we will reuse them later
let v_clone = v_orig.clone();

// Using transmute: this relies on the unspecified data layout of `Vec`, which is a
// bad idea and could cause Undefined Behavior.
// However, it is no-copy.
let v_transmuted = unsafe {
    std::mem::transmute::<Vec<&i32>, Vec<Option<&i32>>>(v_clone)
};

let v_clone = v_orig.clone();

// This is the suggested, safe way.
// It does copy the entire vector, though, into a new array.
let v_collected = v_clone.into_iter()
                         .map(Some)
                         .collect::<Vec<Option<&i32>>>();

let v_clone = v_orig.clone();

// This is the proper no-copy, unsafe way of "transmuting" a `Vec`, without relying on the
// data layout. Instead of literally calling `transmute`, we perform a pointer cast, but
// in terms of converting the original inner type (`&i32`) to the new one (`Option<&i32>`),
// this has all the same caveats. Besides the information provided above, also consult the
// [`from_raw_parts`] documentation.
let v_from_raw = unsafe {
    // Ensure the original vector is not dropped.
    let mut v_clone = std::mem::ManuallyDrop::new(v_clone);
    Vec::from_raw_parts(v_clone.as_mut_ptr() as *mut Option<&i32>,
                        v_clone.len(),
                        v_clone.capacity())
};

Implementing split_at_mut:

use std::{slice, mem};

// There are multiple ways to do this, and there are multiple problems
// with the following (transmute) way.
fn split_at_mut_transmute<T>(slice: &mut [T], mid: usize)
                             -> (&mut [T], &mut [T]) {
    let len = slice.len();
    assert!(mid <= len);
    unsafe {
        let slice2 = mem::transmute::<&mut [T], &mut [T]>(slice);
        // first: transmute is not type safe; all it checks is that T and
        // U are of the same size. Second, right here, you have two
        // mutable references pointing to the same memory.
        (&mut slice[0..mid], &mut slice2[mid..len])
    }
}

// This gets rid of the type safety problems; `&mut *` will *only* give
// you an `&mut T` from an `&mut T` or `*mut T`.
fn split_at_mut_casts<T>(slice: &mut [T], mid: usize)
                         -> (&mut [T], &mut [T]) {
    let len = slice.len();
    assert!(mid <= len);
    unsafe {
        let slice2 = &mut *(slice as *mut [T]);
        // however, you still have two mutable references pointing to
        // the same memory.
        (&mut slice[0..mid], &mut slice2[mid..len])
    }
}

// This is how the standard library does it. This is the best method, if
// you need to do something like this
fn split_at_stdlib<T>(slice: &mut [T], mid: usize)
                      -> (&mut [T], &mut [T]) {
    let len = slice.len();
    assert!(mid <= len);
    unsafe {
        let ptr = slice.as_mut_ptr();
        // This now has three mutable references pointing at the same
        // memory. `slice`, the rvalue ret.0, and the rvalue ret.1.
        // `slice` is never used after `let ptr = ...`, and so one can
        // treat it as "dead", and therefore, you only have two real
        // mutable slices.
        (slice::from_raw_parts_mut(ptr, mid),
         slice::from_raw_parts_mut(ptr.add(mid), len - mid))
    }
}