dyn_vec/src/lib.rs

637 lines
21 KiB
Rust

//! # A [`Vec<T: ?Sized>`]
//!
//! A dynamic length collection of unsized elements, akin to [`std::vec::Vec`].
//!
//! This crate depends on a number of unstable features, namely:
//! - `#![feature(ptr_metadata)]`, to allow manipulating the metadata of pointers.
//! - `#![feature(layout_for_ptr)]`, to allow getting the size/alignment of a value with only a pointer to it.
//! - `#![feature(coerce_unsized)]`, to add trait bounds for types which can be unsized to another type.
//!
//! # Examples
//!
//! You can create a vector with [`Vec::new`]:
//!
//! ```
//! # use std::fmt::Debug;
//! # use dyn_vec::prelude::*;
//! let vec: Vec<dyn Debug> = Vec::new();
//! # assert_eq!(format!("{:?}", vec), "[]");
//! ```
//!
//! or with the [`vec!`] macro:
//!
//! ```
//! # use dyn_vec::prelude::{*, vec};
//! # use std::fmt::Debug;
//! let vec: Vec<i32> = vec![1, 2, 3];
//! // check the docs for `vec!` for more info on this syntax
//! let vec_boxed: Vec<dyn Debug> = vec![box:
//! Box::new(1) as _,
//! Box::new("foo") as _,
//! Box::new(true) as _
//! ];
//! let vec_unsized: Vec<dyn Debug> = vec![unsized: 1, "foo", true];
//! let vec_from_elem: Vec<i32> = vec![3; 5];
//! # assert_eq!(vec, [1, 2, 3]);
//! # assert_eq!(format!("{:?}", vec_boxed), r#"[1, "foo", true]"#);
//! # assert_eq!(format!("{:?}", vec_unsized), r#"[1, "foo", true]"#);
//! # assert_eq!(vec_from_elem, [3; 5]);
//! ```
//!
//! A vector can be pushed to with [`Vec::push`]:
//!
//! ```
//! # use dyn_vec::prelude::{*, vec};
//! let mut vec: Vec<i32> = vec![];
//! vec.push(1);
//! vec.push(2);
//! vec.push(3);
//! # assert_eq!(format!("{:?}", vec), "[1, 2, 3]");
//! ```
//!
//! ...and with [`push_box`] and [`push_unsize`]:
//!
//! ```
//! # use dyn_vec::prelude::{*, vec};
//! # use std::fmt::Debug;
//! let mut vec: Vec<dyn Debug> = vec![];
//! vec.push_box(Box::new(1));
//! vec.push_box(Box::new("foo"));
//! vec.push_box(Box::new(true));
//!
//! vec.push_unsize(2);
//! vec.push_unsize("bar");
//! vec.push_unsize(false);
//! # assert_eq!(format!("{:?}", vec), r#"[1, "foo", true, 2, "bar", false]"#);
//! ```
//!
//! Finally, a vector can be `unsize`d to another vector:
//!
//! ```
//! # use dyn_vec::prelude::{*, vec};
//! # use std::fmt::Debug;
//! let vec: Vec<i32> = vec![1, 2, 3];
//! // vec.push_unsize("foo"); // not yet...
//! let mut vec: Vec<dyn Debug> = vec.unsize();
//! vec.push_unsize("foo"); // now we can!
//! # assert_eq!(format!("{:?}", vec), r#"[1, 2, 3, "foo"]"#);
//! ```
//!
//! # Data Layout
//!
//! ```text
//! Vec<T>
//! ┌────┬────┬────┬────┐
//! │ptr │len │cap │end │
//! └─┬──┴────┴─┬──┴─┬──┘
//! │ │ │
//! │ └────┼───────────────────────────────────────────────┐
//! ┌─┘ └───────────────────┐ │
//! │ │ │
//! ▼ ▼ ▼
//! ┌────┬────┬─────┬──────────┬───┬─────┬───────────────┬───┬───┬───┐
//! │pad │elem│pad │elem │pad│elem │ │ptr│ptr│ptr│
//! └────┴────┴─────┴──────────┴───┴─────┴───────────────┴─┬─┴─┬─┴─┬─┘
//! ▲ ▲ ▲ ▲ │ │ │ ▲
//! │ │ │ └───────────────────────┘ │ │ │
//! │ │ └──────────────────────────────────────────┘ │ │
//! │ └─────────────────────────────────────────────────────────┘ │
//! │ │
//! └─ aligned to 8 also aligned to 8 ─┘
//! ```
//!
//! [`Vec<T: ?Sized>`]: Vec
//! [`push_box`]: Vec::push_box
//! [`push_unsize`]: Vec::push_unsize
#![feature(ptr_metadata)]
#![feature(layout_for_ptr)]
#![feature(coerce_unsized)]
#![warn(missing_docs)]
#![warn(clippy::pedantic)]
#![allow(clippy::must_use_candidate)]
#[cfg(test)]
mod test;
// TODO: maybe remove this?
// Too small to put in its own file
/// Prelude, suitable for glob imports.
///
/// Using `prelude::*` will cause a conflict for the `vec!` macro. `prelude::{*, vec}` is recommended.
pub mod prelude {
pub use super::{Vec, vec};
}
use core::panic;
use std::{
ptr::{NonNull, self, drop_in_place, metadata},
marker::PhantomData,
alloc::{alloc, Layout, dealloc},
mem::{size_of, size_of_val, align_of_val, self, size_of_val_raw},
slice,
ops::{CoerceUnsized, Index, IndexMut},
any::Any
};
/// Copy `size` bytes of memory from `src` to `dst`.
///
/// # Safety
///
/// `src` must be valid for reads, `dst` must be valid for writes, etc, you get the idea.
// TODO: inline me! i didnt realize it was avaliable as `copy_to` until the code was mostly complete.
unsafe fn memcpy(src: *const u8, dst: *mut u8, size: usize) {
src.copy_to(dst, size);
}
fn align_up<T: ?Sized>(ptr: *const T, align: usize) -> *const T {
let (mut data, meta) = ptr.to_raw_parts();
data = ((data as usize + align - 1) & !(align - 1)) as _;
ptr::from_raw_parts(data, meta)
}
fn align_up_mut<T: ?Sized>(ptr: *mut T, align: usize) -> *mut T {
align_up(ptr as _, align) as _
}
/// A heap allocated, dynamic length collection of `?Sized` elements.
///
/// See [`std::vec::Vec`] (the standard library `Vec` type) for more information.
pub struct Vec<T: ?Sized> {
ptr: NonNull<u8>,
len: usize,
capacity: usize,
end_ptr: NonNull<u8>,
_phantom: PhantomData<T>
}
// keeps this file cleaner
mod impls;
mod iter;
pub use iter::*;
/// The extra data stored at the end of the allocation.
type Extra<T> = *const T;
impl<T: ?Sized> Vec<T> {
/// Creates a new, empty vector.
pub fn new() -> Self {
let ptr = NonNull::dangling();
Self {
ptr,
len: 0,
capacity: 0,
end_ptr: ptr,
_phantom: PhantomData
}
}
/// Creates a new vector that holds `len` copies of `v`.
///
/// Only avaliable when `T: Sized`.
pub fn from_elem(v: T, len: usize) -> Self where T: Sized + Clone {
let mut vec = Self::with_capacity(len);
for _ in 0..len {
vec.push(v.clone());
}
vec
}
/// Creates a new vector that can hold the given amount of `T`s.
///
/// Only avaliable when `T: Sized`.
pub fn with_capacity(cap: usize) -> Self where T: Sized {
Self::with_capacity_for::<T>(cap)
}
/// Creates a new vector with the given capacity, measured in bytes.\
pub fn with_capacity_bytes(cap: usize) -> Self {
let mut vec = Self::new();
unsafe { vec.realloc(cap); }
vec
}
/// Creates a new vector with enough capacity to hold the given amount of `U`s.
pub fn with_capacity_for<U>(cap: usize) -> Self {
Self::with_capacity_bytes(cap * (size_of::<U>() + size_of::<Extra<U>>()))
}
/// Appends an element to the end of the vector.
///
/// Only avaliable if `T: Sized`.
pub fn push(&mut self, v: T) where T: Sized {
unsafe { self.push_raw(&v) }
mem::forget(v);
}
/// Appends an (possibly unsized) boxed element to the end of the vector.
pub fn push_box(&mut self, v: Box<T>) {
let ptr = Box::into_raw(v);
unsafe {
self.push_raw(ptr);
dealloc(ptr.cast(), Layout::for_value_raw(ptr));
}
}
/// Appends a sized element of type `U` to the end of the vector, given that it can be `CoerceUnsized` to a `T`.
pub fn push_unsize<U>(&mut self, v: U) where for<'a> &'a U: CoerceUnsized<&'a T> {
let v_unsized: &T = &v;
unsafe { self.push_raw(v_unsized) };
mem::forget(v);
}
unsafe fn push_raw(&mut self, v: *const T) {
let size = size_of_val(&*v);
if !self.will_fit(&*v) {
// oh no! allocation too small!
// make sure we have enough space for a new element, but also space for future elements
// this bit is tricky, we must make sure we have enough space for padding too, so its probably UB somehow
// FIXME: ^^^
let new_alloc_size = self.capacity * 2 + size * 2 + size_of::<Extra<T>>();
self.realloc(new_alloc_size);
}
self.push_raw_unchecked(v);
}
/// Given an element, returns a pointer to where it would be written if it was pushed, assuming no reallocation is needed.
///
/// The pointer will be aligned, but writing to it may overwrite data belonging to the Vec.
/// To check for this, call `will_fit`.
/// In addition, the extra data for the element must be set using `set_extra_from_ptr`.
fn get_next_elem_ptr(&self, v: &T) -> *mut u8 {
align_up_mut(self.end_ptr.as_ptr(), align_of_val(v))
}
/// Checks if a given element will fit in the vector without reallocations.
pub fn will_fit(&self, v: &T) -> bool {
let remaining_space = self.get_ptr_to_extra(self.len) as usize - self.end_ptr.as_ptr() as usize;
let needed_space = size_of_val(v) + size_of::<Extra<T>>();
remaining_space >= needed_space
}
unsafe fn push_raw_unchecked(&mut self, v: *const T) {
let size = size_of_val(&*v);
let dest = self.get_next_elem_ptr(&*v); // this is mentioned by the `// SAFETY:` in `as_slice_flatten`
memcpy(v.cast(), dest, size);
let new_ptr = ptr::from_raw_parts::<T>(dest.cast(), metadata(v));
self.set_extra_from_ptr(self.len, new_ptr);
self.end_ptr = NonNull::new_unchecked(dest.wrapping_add(size));
self.len += 1;
}
/// Pops an element off the end of the vector, putting it in a [`Box`].
pub fn pop(&mut self) -> Option<Box<T>> {
unsafe {
self.len = self.len.checked_sub(1)?;
let el = self.get_ptr(self.len);
let alloc = alloc(Layout::for_value_raw(el));
memcpy(el.cast(), alloc, size_of_val_raw(el));
Some(Box::from_raw(ptr::from_raw_parts_mut(alloc.cast(), metadata(el))))
}
}
unsafe fn realloc(&mut self, size: usize) {
let layout = Layout::from_size_align_unchecked(size, 8).pad_to_align();
if self.capacity == 0 {
// will panic if OOM
self.ptr = NonNull::new(alloc(layout)).unwrap();
self.end_ptr = self.ptr;
} else {
// cannot use realloc here
let new_alloc = NonNull::new(alloc(layout)).unwrap();
// data
let mut ptr = new_alloc.as_ptr();
for i in 0..self.len {
let v = self.get_unchecked(i);
let size = size_of_val(v);
ptr = align_up_mut(ptr, align_of_val(v));
memcpy((v as *const T).cast(), ptr, size);
let meta = self.get_ptr(i).to_raw_parts().1;
self.set_extra_from_ptr(i, ptr::from_raw_parts(ptr.cast(), meta));
ptr = ptr.wrapping_add(size);
}
self.end_ptr = NonNull::new_unchecked(ptr);
// metadata
let meta_src = self.get_ptr_to_extra(self.len);
let meta_dst = {
let current_alloc_end = self.ptr.as_ptr().wrapping_add(self.capacity);
let new_alloc_end = new_alloc.as_ptr().wrapping_add(layout.size());
let meta_len = current_alloc_end as usize - meta_src as usize;
new_alloc_end.wrapping_sub(meta_len)
};
let meta_size = self.len * size_of::<Extra<T>>();
memcpy(meta_src.cast(), meta_dst, meta_size);
dealloc(self.ptr.as_ptr(), Layout::from_size_align_unchecked(self.capacity, 8));
self.ptr = new_alloc;
}
self.capacity = layout.size();
}
/// for internal use
///
/// NOTE: 1-indexed, to allow getting a pointer to the end of the alloc easily
fn get_ptr_to_extra(&self, index: usize) -> *mut Extra<T> {
self.ptr.as_ptr()
.wrapping_add(self.capacity)
.cast::<Extra<T>>()
.wrapping_sub(index)
}
unsafe fn set_extra_from_ptr(&self, index: usize, ptr: *const T) {
self.get_ptr_to_extra(index + 1).write(ptr);
}
unsafe fn get_ptr(&self, index: usize) -> *const T {
*self.get_ptr_to_extra(index + 1)
}
/// Gets a reference to the element at the specified index.
///
/// Returns `None` if the index is out-of-bounds.
pub fn get(&self, index: usize) -> Option<&T> {
if index < self.len {
Some(unsafe { self.get_unchecked(index) })
} else {
None
}
}
/// Gets a reference to the element at the specified index.
///
/// # Safety
///
/// Immediate UB if the index is out-of-bounds.
pub unsafe fn get_unchecked(&self, index: usize) -> &T {
&*self.get_ptr(index)
}
/// Gets a mutable reference to the element at the specified index.
///
/// Returns `None` if the index is out-of-bounds.
pub fn get_mut(&mut self, index: usize) -> Option<&mut T> {
if index < self.len {
Some(unsafe { self.get_unchecked_mut(index) })
} else {
None
}
}
/// Gets a mutable reference to the element at the specified index.
///
/// # Safety
///
/// Immediate UB if the index is out-of-bounds.
pub unsafe fn get_unchecked_mut(&mut self, index: usize) -> &mut T {
&mut *(self.get_ptr(index) as *mut _)
}
/// Returns the length of the vector, which is how many items it contains.
pub fn len(&self) -> usize {
self.len
}
/// Returns `true` if the vector holds no elements.
pub fn is_empty(&self) -> bool {
self.len == 0
}
/// Returns the capacity, which is the size of the allocation in bytes.
pub fn capacity(&self) -> usize {
self.capacity
}
/// Returns a pointer to the allocation of the vector.
pub fn as_ptr(&self) -> *const u8 {
self.ptr.as_ptr()
}
/// Returns a mutable pointer to the allocation of the vector.
pub fn as_mut_ptr(&mut self) -> *mut u8 {
self.ptr.as_ptr()
}
/// Iterates over the vector by-ref.
pub fn iter(&self) -> Iter<T> {
Iter::new(self)
}
/// Iterates over the vector by-mut.
pub fn iter_mut(&mut self) -> IterMut<T> {
IterMut::new(self)
}
/// Converts a `Vec<T: Sized>` into a `Vec<U: ?Sized>`, given that `T` can be `CoerceUnsized` into `U`.
pub fn unsize<U: ?Sized>(mut self) -> Vec<U> where for<'a> &'a T: CoerceUnsized<&'a U> {
if size_of::<Extra<U>>() > size_of::<Extra<T>>() {
let elem_size = self.end_ptr.as_ptr() as usize - self.ptr.as_ptr() as usize;
let extra_size = self.len * size_of::<Extra<U>>();
let needed_size = elem_size + extra_size;
if needed_size > self.capacity {
unsafe { self.realloc(needed_size); }
}
}
let new_vec = Vec::<U> {
ptr: self.ptr,
len: self.len,
capacity: self.capacity,
end_ptr: self.end_ptr,
_phantom: PhantomData,
};
if size_of::<Extra<U>>() > size_of::<Extra<T>>() {
// new extra larger than old extra, must go from back to front
for i in (0..self.len).rev() {
let current = unsafe { &*self.get_ptr(i) };
unsafe { new_vec.set_extra_from_ptr(i, current as &U) }
}
} else {
// new extra smaller or same size as old extra, must go from front to back
for i in 0..self.len {
let current = unsafe { &*self.get_ptr(i) };
unsafe { new_vec.set_extra_from_ptr(i, current as &U) }
}
}
mem::forget(self);
new_vec
}
unsafe fn dealloc(&self) {
if self.capacity != 0 {
dealloc(self.ptr.as_ptr(), Layout::from_size_align_unchecked(self.capacity, 8));
}
}
/// Extends this vector with an iterator.
///
/// Similar to [`Extend::extend`], but seperate to prevent conflicting implementations.
pub fn extend_unsize<U, I: IntoIterator<Item = U>>(&mut self, iter: I) where for<'a> &'a U: CoerceUnsized<&'a T> {
for item in iter {
self.push_unsize(item);
}
}
}
impl<T> Vec<[T]> {
/// Returns a slice over all the elements in the vector.
///
/// Only avaliable for `Vec<[T]>`.
pub fn as_slice_flatten(&self) -> &[T] {
if self.len == 0 {
return unsafe { slice::from_raw_parts(NonNull::dangling().as_ptr(), 0) }
}
// SAFETY: the slices should be contiguous by the logic of `push_raw_unchecked`
unsafe {
slice::from_raw_parts(self.get_ptr(0).to_raw_parts().0.cast(), {
let start = self.get_ptr(0).to_raw_parts().0 as usize;
let end = self.end_ptr.as_ptr() as usize;
debug_assert_eq!((end - start) % size_of::<T>(), 0);
(end - start) / size_of::<T>() // integer division!
})
}
}
/// Returns a mutable slice over all the elements in the vector.
///
/// Only avaliable for `Vec<[T]>`.
pub fn as_mut_slice_flatten(&mut self) -> &mut [T] {
if self.len == 0 {
return unsafe { slice::from_raw_parts_mut(NonNull::dangling().as_ptr(), 0) }
}
// SAFETY: the slices should be contiguous by the logic of `push_raw_unchecked`
unsafe {
slice::from_raw_parts_mut(self.get_ptr(0).to_raw_parts().0 as _, {
let start = self.get_ptr(0).to_raw_parts().0 as usize;
let end = self.end_ptr.as_ptr() as usize;
debug_assert_eq!((end - start) % size_of::<T>(), 0);
(end - start) / size_of::<T>() // integer division!
})
}
}
}
impl Vec<dyn Any> {
/// Gets a reference to the element at then specified index, downcasting it to the specified type.
///
/// Same as `.get().map(|v| v.downcast()).flatten()`.
pub fn downcast_get<T: Any>(&self, index: usize) -> Option<&T> {
self.get(index)?.downcast_ref()
}
/// Gets a mutable reference to the element at then specified index, downcasting it to the specified type.
///
/// Same as `.get_mut().map(|v| v.downcast_mut()).flatten()`.
pub fn downcast_get_mut<T: Any>(&mut self, index: usize) -> Option<&mut T> {
self.get_mut(index)?.downcast_mut()
}
/// Pops an element off the end of the vector, downcasting it to the specified type.
///
/// If the element is not of type `T`, the element will not be popped.
///
/// Similiar to `.pop().map(|v| v.downcast()).flatten()`, but without an intermediate allocation.
pub fn downcast_pop<T: Any>(&mut self) -> Option<T> {
unsafe {
let el = self.get_ptr(self.len.checked_sub(1)?) as *mut dyn Any;
let v = Some(((&mut *el).downcast_mut()? as *mut T).read());
self.len -= 1;
v
}
}
}
impl<T: ?Sized> Drop for Vec<T> {
fn drop(&mut self) {
unsafe {
for i in 0..self.len {
drop_in_place(self.get_unchecked_mut(i));
}
self.dealloc();
}
}
}
impl<T: ?Sized> Index<usize> for Vec<T> {
type Output = T;
#[track_caller]
fn index(&self, index: usize) -> &Self::Output {
match self.get(index) {
Some(v) => v,
None => panic!("index out of bounds: the len is {} but the index is {}", self.len, index),
}
}
}
impl<T: ?Sized> IndexMut<usize> for Vec<T> {
#[track_caller]
fn index_mut(&mut self, index: usize) -> &mut Self::Output {
let len = self.len;
match self.get_mut(index) {
Some(v) => v,
None => panic!("index out of bounds: the len is {} but the index is {}", len, index),
}
}
}
/// Creates a [`Vec`].
///
/// # Examples
///
/// ```
/// # use dyn_vec::prelude::{vec, Vec};
/// # use std::fmt::Debug;
/// let vec1: Vec<i32> = vec![1, 2, 3];
/// let vec2: Vec<dyn Debug> = vec![box:
/// Box::new(1) as _,
/// Box::new(String::from("foo")) as _,
/// Box::new(true) as _
/// ];
/// let vec3: Vec<dyn Debug> = vec![unsized: 1, String::from("foo"), true];
/// ```
#[macro_export]
macro_rules! vec {
() => {
$crate::Vec::new();
};
(box: $($elem:expr),+ $(,)?) => {{
let mut vec = $crate::Vec::new();
$(vec.push_box($elem);)+
vec
}};
(unsized: $($elem:expr),+ $(,)?) => {{
let mut vec = $crate::Vec::new();
$(vec.push_unsize($elem);)+
vec
}};
($elem:expr; $n:expr) => {
$crate::Vec::from_elem($elem, $n)
};
($($elem:expr),+ $(,)?) => {{
let mut vec = $crate::Vec::new();
$(vec.push($elem);)+
vec
}};
}