Quellcode-Bibliothek mod.rs
Sprache: unbekannt
|
|
Columbo aufrufen.rs Download desUnknown {[0] [0] [0]}Datei anzeigen
use crate::alloc::alloc::{handle_alloc_error, Layout};
use crate::scopeguard::{guard, ScopeGuard};
use crate::TryReserveError;
use core::iter::FusedIterator;
use core::marker::PhantomData;
use core::mem;
use core::mem::MaybeUninit;
use core::ptr::NonNull;
use core::{hint, ptr};
cfg_if! {
// Use the SSE2 implementation if possible: it allows us to scan 16 buckets
// at once instead of 8. We don't bother with AVX since it would require
// runtime dispatch and wouldn't gain us much anyways: the probability of
// finding a match drops off drastically after the first few buckets.
//
// I attempted an implementation on ARM using NEON instructions, but it
// turns out that most NEON instructions have multi-cycle latency, which in
// the end outweighs any gains over the generic implementation.
if #[cfg(all(
target_feature = "sse2",
any(target_arch = "x86", target_arch = "x86_64"),
not(miri),
))] {
mod sse2;
use sse2 as imp;
} else if #[cfg(all(
target_arch = "aarch64",
target_feature = "neon",
// NEON intrinsics are currently broken on big-endian targets.
// See https://github.com/rust-lang/stdarch/issues/1484.
target_endian = "little",
not(miri),
))] {
mod neon;
use neon as imp;
} else {
mod generic;
use generic as imp;
}
}
mod alloc;
pub(crate) use self::alloc::{do_alloc, Allocator, Global};
mod bitmask;
use self::bitmask::BitMaskIter;
use self::imp::Group;
// Branch prediction hint. This is currently only available on nightly but it
// consistently improves performance by 10-15%.
#[cfg(not(feature = "nightly"))]
use core::convert::identity as likely;
#[cfg(not(feature = "nightly"))]
use core::convert::identity as unlikely;
#[cfg(feature = "nightly")]
use core::intrinsics::{likely, unlikely};
// FIXME: use strict provenance functions once they are stable.
// Implement it with a transmute for now.
#[inline(always)]
#[allow(clippy::useless_transmute)] // clippy is wrong, cast and transmute are different h ere
fn invalid_mut<T>(addr: usize) -> *mut T {
unsafe { core::mem::transmute(addr) }
}
#[inline]
unsafe fn offset_from<T>(to: *const T, from: *const T) -> usize {
to.offset_from(from) as usize
}
/// Whether memory allocation errors should return an error or abort.
#[derive(Copy, Clone)]
enum Fallibility {
Fallible,
Infallible,
}
impl Fallibility {
/// Error to return on capacity overflow.
#[cfg_attr(feature = "inline-more", inline)]
fn capacity_overflow(self) -> TryReserveError {
match self {
Fallibility::Fallible => TryReserveError::CapacityOverflow,
Fallibility::Infallible => panic!("Hash table capacity overflow"),
}
}
/// Error to return on allocation error.
#[cfg_attr(feature = "inline-more", inline)]
fn alloc_err(self, layout: Layout) -> TryReserveError {
match self {
Fallibility::Fallible => TryReserveError::AllocError { layout },
Fallibility::Infallible => handle_alloc_error(layout),
}
}
}
trait SizedTypeProperties: Sized {
const IS_ZERO_SIZED: bool = mem::size_of::<Self>() == 0;
const NEEDS_DROP: bool = mem::needs_drop::<Self>();
}
impl<T> SizedTypeProperties for T {}
/// Control byte value for an empty bucket.
const EMPTY: u8 = 0b1111_1111;
/// Control byte value for a deleted bucket.
const DELETED: u8 = 0b1000_0000;
/// Checks whether a control byte represents a full bucket (top bit is clear).
#[inline]
fn is_full(ctrl: u8) -> bool {
ctrl & 0x80 == 0
}
/// Checks whether a control byte represents a special value (top bit is set).
#[inline]
fn is_special(ctrl: u8) -> bool {
ctrl & 0x80 != 0
}
/// Checks whether a special control value is EMPTY (just check 1 bit).
#[inline]
fn special_is_empty(ctrl: u8) -> bool {
debug_assert!(is_special(ctrl));
ctrl & 0x01 != 0
}
/// Primary hash function, used to select the initial bucket to probe from.
#[inline]
#[allow(clippy::cast_possible_truncation)]
fn h1(hash: u64) -> usize {
// On 32-bit platforms we simply ignore the higher hash bits.
hash as usize
}
// Constant for h2 function that grabing the top 7 bits of the hash.
const MIN_HASH_LEN: usize = if mem::size_of::<usize>() < mem::size_of::<u64>() {
mem::size_of::<usize>()
} else {
mem::size_of::<u64>()
};
/// Secondary hash function, saved in the low 7 bits of the control byte.
#[inline]
#[allow(clippy::cast_possible_truncation)]
fn h2(hash: u64) -> u8 {
// Grab the top 7 bits of the hash. While the hash is normally a full 64-bit
// value, some hash functions (such as FxHash) produce a usize result
// instead, which means that the top 32 bits are 0 on 32-bit platforms.
// So we use MIN_HASH_LEN constant to handle this.
let top7 = hash >> (MIN_HASH_LEN * 8 - 7);
(top7 & 0x7f) as u8 // truncation
}
/// Probe sequence based on triangular numbers, which is guaranteed (since our
/// table size is a power of two) to visit every group of elements exactly once.
///
/// A triangular probe has us jump by 1 more group every time. So first we
/// jump by 1 group (meaning we just continue our linear scan), then 2 groups
/// (skipping over 1 group), then 3 groups (skipping over 2 groups), and so on.
///
/// Proof that the probe will visit every group in the table:
/// <https://fgiesen.wordpress.com/2015/02/22/triangular-numbers-mod-2n/>
struct ProbeSeq {
pos: usize,
stride: usize,
}
impl ProbeSeq {
#[inline]
fn move_next(&mut self, bucket_mask: usize) {
// We should have found an empty bucket by now and ended the probe.
debug_assert!(
self.stride <= bucket_mask,
"Went past end of probe sequence"
);
self.stride += Group::WIDTH;
self.pos += self.stride;
self.pos &= bucket_mask;
}
}
/// Returns the number of buckets needed to hold the given number of items,
/// taking the maximum load factor into account.
///
/// Returns `None` if an overflow occurs.
// Workaround for emscripten bug emscripten-core/emscripten-fastcomp#258
#[cfg_attr(target_os = "emscripten", inline(never))]
#[cfg_attr(not(target_os = "emscripten"), inline)]
fn capacity_to_buckets(cap: usize) -> Option<usize> {
debug_assert_ne!(cap, 0);
// For small tables we require at least 1 empty bucket so that lookups are
// guaranteed to terminate if an element doesn't exist in the table.
if cap < 8 {
// We don't bother with a table size of 2 buckets since that can only
// hold a single element. Instead we skip directly to a 4 bucket table
// which can hold 3 elements.
return Some(if cap < 4 { 4 } else { 8 });
}
// Otherwise require 1/8 buckets to be empty (87.5% load)
//
// Be careful when modifying this, calculate_layout relies on the
// overflow check here.
let adjusted_cap = cap.checked_mul(8)? / 7;
// Any overflows will have been caught by the checked_mul. Also, any
// rounding errors from the division above will be cleaned up by
// next_power_of_two (which can't overflow because of the previous division).
Some(adjusted_cap.next_power_of_two())
}
/// Returns the maximum effective capacity for the given bucket mask, taking
/// the maximum load factor into account.
#[inline]
fn bucket_mask_to_capacity(bucket_mask: usize) -> usize {
if bucket_mask < 8 {
// For tables with 1/2/4/8 buckets, we always reserve one empty slot.
// Keep in mind that the bucket mask is one less than the bucket count.
bucket_mask
} else {
// For larger tables we reserve 12.5% of the slots as empty.
((bucket_mask + 1) / 8) * 7
}
}
/// Helper which allows the max calculation for ctrl_align to be statically computed for each T
/// while keeping the rest of `calculate_layout_for` independent of `T`
#[derive(Copy, Clone)]
struct TableLayout {
size: usize,
ctrl_align: usize,
}
impl TableLayout {
#[inline]
const fn new<T>() -> Self {
let layout = Layout::new::<T>();
Self {
size: layout.size(),
ctrl_align: if layout.align() > Group::WIDTH {
layout.align()
} else {
Group::WIDTH
},
}
}
#[inline]
fn calculate_layout_for(self, buckets: usize) -> Option<(Layout, usize)> {
debug_assert!(buckets.is_power_of_two());
let TableLayout { size, ctrl_align } = self;
// Manual layout calculation since Layout methods are not yet stable.
let ctrl_offset =
size.checked_mul(buckets)?.checked_add(ctrl_align - 1)? & !(ctrl_align - 1);
let len = ctrl_offset.checked_add(buckets + Group::WIDTH)?;
// We need an additional check to ensure that the allocation doesn't
// exceed `isize::MAX` (https://github.com/rust-lang/rust/pull/95295).
if len > isize::MAX as usize - (ctrl_align - 1) {
return None;
}
Some((
unsafe { Layout::from_size_align_unchecked(len, ctrl_align) },
ctrl_offset,
))
}
}
/// A reference to an empty bucket into which an can be inserted.
pub struct InsertSlot {
index: usize,
}
/// A reference to a hash table bucket containing a `T`.
///
/// This is usually just a pointer to the element itself. However if the element
/// is a ZST, then we instead track the index of the element in the table so
/// that `erase` works properly.
pub struct Bucket<T> {
// Actually it is pointer to next element than element itself
// this is needed to maintain pointer arithmetic invariants
// keeping direct pointer to element introduces difficulty.
// Using `NonNull` for variance and niche layout
ptr: NonNull<T>,
}
// This Send impl is needed for rayon support. This is safe since Bucket is
// never exposed in a public API.
unsafe impl<T> Send for Bucket<T> {}
impl<T> Clone for Bucket<T> {
#[inline]
fn clone(&self) -> Self {
Self { ptr: self.ptr }
}
}
impl<T> Bucket<T> {
/// Creates a [`Bucket`] that contain pointer to the data.
/// The pointer calculation is performed by calculating the
/// offset from given `base` pointer (convenience for
/// `base.as_ptr().sub(index)`).
///
/// `index` is in units of `T`; e.g., an `index` of 3 represents a pointer
/// offset of `3 * size_of::<T>()` bytes.
///
/// If the `T` is a ZST, then we instead track the index of the element
/// in the table so that `erase` works properly (return
/// `NonNull::new_unchecked((index + 1) as *mut T)`)
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived
/// from the safety rules for [`<*mut T>::sub`] method of `*mut T` and the safety
/// rules of [`NonNull::new_unchecked`] function.
///
/// Thus, in order to uphold the safety contracts for the [`<*mut T>::sub`] method
/// and [`NonNull::new_unchecked`] function, as well as for the correct
/// logic of the work of this crate, the following rules are necessary and
/// sufficient:
///
/// * the `base` pointer must not be `dangling` and must points to the
/// end of the first `value element` from the `data part` of the table, i.e.
/// must be the pointer that returned by [`RawTable::data_end`] or by
/// [`RawTableInner::data_end<T>`];
///
/// * `index` must not be greater than `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)`
/// must be no greater than the number returned by the function
/// [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the
/// `index` must not be greater than `RawTableInner.bucket_mask`, i.e.
/// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)`
/// must be no greater than the number returned by the function
/// [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// [`Bucket`]: crate::raw::Bucket
/// [`<*mut T>::sub`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.sub-1
/// [`NonNull::new_unchecked`]: https://doc.rust-lang.org/stable/std/ptr/struct.NonNull.html#method.new_unchecked
/// [`RawTable::data_end`]: crate::raw::RawTable::data_end
/// [`RawTableInner::data_end<T>`]: RawTableInner::data_end<T>
/// [`RawTable::buckets`]: crate::raw::RawTable::buckets
/// [`RawTableInner::buckets`]: RawTableInner::buckets
#[inline]
unsafe fn from_base_index(base: NonNull<T>, index: usize) -> Self {
// If mem::size_of::<T>() != 0 then return a pointer to an `element` in
// the data part of the table (we start counting from "0", so that
// in the expression T[last], the "last" index actually one less than the
// "buckets" number in the table, i.e. "last = RawTableInner.bucket_mask"):
//
// `from_base_index(base, 1).as_ptr()` returns a pointer that
// points here in the data part of the table
// (to the start of T1)
// |
// | `base: NonNull<T>` must point here
// | (to the end of T0 or to the start of C0)
// v v
// [Padding], Tlast, ..., |T1|, T0, |C0, C1, ..., Clast
// ^
// `from_base_index(base, 1)` returns a pointer
// that points here in the data part of the table
// (to the end of T1)
//
// where: T0...Tlast - our stored data; C0...Clast - control bytes
// or metadata for data.
let ptr = if T::IS_ZERO_SIZED {
// won't overflow because index must be less than length (bucket_mask)
// and bucket_mask is guaranteed to be less than `isize::MAX`
// (see TableLayout::calculate_layout_for method)
invalid_mut(index + 1)
} else {
base.as_ptr().sub(index)
};
Self {
ptr: NonNull::new_unchecked(ptr),
}
}
/// Calculates the index of a [`Bucket`] as distance between two pointers
/// (convenience for `base.as_ptr().offset_from(self.ptr.as_ptr()) as usize`).
/// The returned value is in units of T: the distance in bytes divided by
/// [`core::mem::size_of::<T>()`].
///
/// If the `T` is a ZST, then we return the index of the element in
/// the table so that `erase` works properly (return `self.ptr.as_ptr() as usize - 1`).
///
/// This function is the inverse of [`from_base_index`].
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived
/// from the safety rules for [`<*const T>::offset_from`] method of `*const T`.
///
/// Thus, in order to uphold the safety contracts for [`<*const T>::offset_from`]
/// method, as well as for the correct logic of the work of this crate, the
/// following rules are necessary and sufficient:
///
/// * `base` contained pointer must not be `dangling` and must point to the
/// end of the first `element` from the `data part` of the table, i.e.
/// must be a pointer that returns by [`RawTable::data_end`] or by
/// [`RawTableInner::data_end<T>`];
///
/// * `self` also must not contain dangling pointer;
///
/// * both `self` and `base` must be created from the same [`RawTable`]
/// (or [`RawTableInner`]).
///
/// If `mem::size_of::<T>() == 0`, this function is always safe.
///
/// [`Bucket`]: crate::raw::Bucket
/// [`from_base_index`]: crate::raw::Bucket::from_base_index
/// [`RawTable::data_end`]: crate::raw::RawTable::data_end
/// [`RawTableInner::data_end<T>`]: RawTableInner::data_end<T>
/// [`RawTable`]: crate::raw::RawTable
/// [`RawTableInner`]: RawTableInner
/// [`<*const T>::offset_from`]: https://doc.rust-lang.org/nightly/core/primitive.pointer.html#method.offset_from
#[inline]
unsafe fn to_base_index(&self, base: NonNull<T>) -> usize {
// If mem::size_of::<T>() != 0 then return an index under which we used to store the
// `element` in the data part of the table (we start counting from "0", so
// that in the expression T[last], the "last" index actually is one less than the
// "buckets" number in the table, i.e. "last = RawTableInner.bucket_mask").
// For example for 5th element in table calculation is performed like this:
//
// mem::size_of::<T>()
// |
// | `self = from_base_index(base, 5)` that returns pointer
// | that points here in tha data part of the table
// | (to the end of T5)
// | | `base: NonNull<T>` must point here
// v | (to the end of T0 or to the start of C0)
// /???\ v v
// [Padding], Tlast, ..., |T10|, ..., T5|, T4, T3, T2, T1, T0, |C0, C1, C2, C3, C4, C5, ..., C10, ..., Clast
// \__________ __________/
// \/
// `bucket.to_base_index(base)` = 5
// (base.as_ptr() as usize - self.ptr.as_ptr() as usize) / mem::size_of::<T>()
//
// where: T0...Tlast - our stored data; C0...Clast - control bytes or metadata for data.
if T::IS_ZERO_SIZED {
// this can not be UB
self.ptr.as_ptr() as usize - 1
} else {
offset_from(base.as_ptr(), self.ptr.as_ptr())
}
}
/// Acquires the underlying raw pointer `*mut T` to `data`.
///
/// # Note
///
/// If `T` is not [`Copy`], do not use `*mut T` methods that can cause calling the
/// destructor of `T` (for example the [`<*mut T>::drop_in_place`] method), because
/// for properly dropping the data we also need to clear `data` control bytes. If we
/// drop data, but do not clear `data control byte` it leads to double drop when
/// [`RawTable`] goes out of scope.
///
/// If you modify an already initialized `value`, so [`Hash`] and [`Eq`] on the new
/// `T` value and its borrowed form *must* match those for the old `T` value, as the map
/// will not re-evaluate where the new value should go, meaning the value may become
/// "lost" if their location does not reflect their state.
///
/// [`RawTable`]: crate::raw::RawTable
/// [`<*mut T>::drop_in_place`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.drop_in_place
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
///
/// # Examples
///
/// ```
/// # #[cfg(feature = "raw")]
/// # fn test() {
/// use core::hash::{BuildHasher, Hash};
/// use hashbrown::raw::{Bucket, RawTable};
///
/// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>;
///
/// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 {
/// use core::hash::Hasher;
/// let mut state = hash_builder.build_hasher();
/// key.hash(&mut state);
/// state.finish()
/// }
///
/// let hash_builder = NewHashBuilder::default();
/// let mut table = RawTable::new();
///
/// let value = ("a", 100);
/// let hash = make_hash(&hash_builder, &value.0);
///
/// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0));
///
/// let bucket: Bucket<(&str, i32)> = table.find(hash, |(k1, _)| k1 == &value.0).unwrap();
///
/// assert_eq!(unsafe { &*bucket.as_ptr() }, &("a", 100));
/// # }
/// # fn main() {
/// # #[cfg(feature = "raw")]
/// # test()
/// # }
/// ```
#[inline]
pub fn as_ptr(&self) -> *mut T {
if T::IS_ZERO_SIZED {
// Just return an arbitrary ZST pointer which is properly aligned
// invalid pointer is good enough for ZST
invalid_mut(mem::align_of::<T>())
} else {
unsafe { self.ptr.as_ptr().sub(1) }
}
}
/// Create a new [`Bucket`] that is offset from the `self` by the given
/// `offset`. The pointer calculation is performed by calculating the
/// offset from `self` pointer (convenience for `self.ptr.as_ptr().sub(offset)`).
/// This function is used for iterators.
///
/// `offset` is in units of `T`; e.g., a `offset` of 3 represents a pointer
/// offset of `3 * size_of::<T>()` bytes.
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived
/// from the safety rules for [`<*mut T>::sub`] method of `*mut T` and safety
/// rules of [`NonNull::new_unchecked`] function.
///
/// Thus, in order to uphold the safety contracts for [`<*mut T>::sub`] method
/// and [`NonNull::new_unchecked`] function, as well as for the correct
/// logic of the work of this crate, the following rules are necessary and
/// sufficient:
///
/// * `self` contained pointer must not be `dangling`;
///
/// * `self.to_base_index() + ofset` must not be greater than `RawTableInner.bucket_mask`,
/// i.e. `(self.to_base_index() + ofset) <= RawTableInner.bucket_mask` or, in other
/// words, `self.to_base_index() + ofset + 1` must be no greater than the number returned
/// by the function [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the
/// `self.to_base_index() + ofset` must not be greater than `RawTableInner.bucket_mask`,
/// i.e. `(self.to_base_index() + ofset) <= RawTableInner.bucket_mask` or, in other words,
/// `self.to_base_index() + ofset + 1` must be no greater than the number returned by the
/// function [`RawTable::buckets`] or [`RawTableInner::buckets`].
///
/// [`Bucket`]: crate::raw::Bucket
/// [`<*mut T>::sub`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.sub-1
/// [`NonNull::new_unchecked`]: https://doc.rust-lang.org/stable/std/ptr/struct.NonNull.html#method.new_unchecked
/// [`RawTable::buckets`]: crate::raw::RawTable::buckets
/// [`RawTableInner::buckets`]: RawTableInner::buckets
#[inline]
unsafe fn next_n(&self, offset: usize) -> Self {
let ptr = if T::IS_ZERO_SIZED {
// invalid pointer is good enough for ZST
invalid_mut(self.ptr.as_ptr() as usize + offset)
} else {
self.ptr.as_ptr().sub(offset)
};
Self {
ptr: NonNull::new_unchecked(ptr),
}
}
/// Executes the destructor (if any) of the pointed-to `data`.
///
/// # Safety
///
/// See [`ptr::drop_in_place`] for safety concerns.
///
/// You should use [`RawTable::erase`] instead of this function,
/// or be careful with calling this function directly, because for
/// properly dropping the data we need also clear `data` control bytes.
/// If we drop data, but do not erase `data control byte` it leads to
/// double drop when [`RawTable`] goes out of scope.
///
/// [`ptr::drop_in_place`]: https://doc.rust-lang.org/core/ptr/fn.drop_in_place.html
/// [`RawTable`]: crate::raw::RawTable
/// [`RawTable::erase`]: crate::raw::RawTable::erase
#[cfg_attr(feature = "inline-more", inline)]
pub(crate) unsafe fn drop(&self) {
self.as_ptr().drop_in_place();
}
/// Reads the `value` from `self` without moving it. This leaves the
/// memory in `self` unchanged.
///
/// # Safety
///
/// See [`ptr::read`] for safety concerns.
///
/// You should use [`RawTable::remove`] instead of this function,
/// or be careful with calling this function directly, because compiler
/// calls its destructor when readed `value` goes out of scope. It
/// can cause double dropping when [`RawTable`] goes out of scope,
/// because of not erased `data control byte`.
///
/// [`ptr::read`]: https://doc.rust-lang.org/core/ptr/fn.read.html
/// [`RawTable`]: crate::raw::RawTable
/// [`RawTable::remove`]: crate::raw::RawTable::remove
#[inline]
pub(crate) unsafe fn read(&self) -> T {
self.as_ptr().read()
}
/// Overwrites a memory location with the given `value` without reading
/// or dropping the old value (like [`ptr::write`] function).
///
/// # Safety
///
/// See [`ptr::write`] for safety concerns.
///
/// # Note
///
/// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match
/// those for the old `T` value, as the map will not re-evaluate where the new
/// value should go, meaning the value may become "lost" if their location
/// does not reflect their state.
///
/// [`ptr::write`]: https://doc.rust-lang.org/core/ptr/fn.write.html
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
#[inline]
pub(crate) unsafe fn write(&self, val: T) {
self.as_ptr().write(val);
}
/// Returns a shared immutable reference to the `value`.
///
/// # Safety
///
/// See [`NonNull::as_ref`] for safety concerns.
///
/// [`NonNull::as_ref`]: https://doc.rust-lang.org/core/ptr/struct.NonNull.html#method.as_ref
///
/// # Examples
///
/// ```
/// # #[cfg(feature = "raw")]
/// # fn test() {
/// use core::hash::{BuildHasher, Hash};
/// use hashbrown::raw::{Bucket, RawTable};
///
/// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>;
///
/// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 {
/// use core::hash::Hasher;
/// let mut state = hash_builder.build_hasher();
/// key.hash(&mut state);
/// state.finish()
/// }
///
/// let hash_builder = NewHashBuilder::default();
/// let mut table = RawTable::new();
///
/// let value: (&str, String) = ("A pony", "is a small horse".to_owned());
/// let hash = make_hash(&hash_builder, &value.0);
///
/// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0));
///
/// let bucket: Bucket<(&str, String)> = table.find(hash, |(k, _)| k == &value.0).unwrap();
///
/// assert_eq!(
/// unsafe { bucket.as_ref() },
/// &("A pony", "is a small horse".to_owned())
/// );
/// # }
/// # fn main() {
/// # #[cfg(feature = "raw")]
/// # test()
/// # }
/// ```
#[inline]
pub unsafe fn as_ref<'a>(&self) -> &'a T {
&*self.as_ptr()
}
/// Returns a unique mutable reference to the `value`.
///
/// # Safety
///
/// See [`NonNull::as_mut`] for safety concerns.
///
/// # Note
///
/// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match
/// those for the old `T` value, as the map will not re-evaluate where the new
/// value should go, meaning the value may become "lost" if their location
/// does not reflect their state.
///
/// [`NonNull::as_mut`]: https://doc.rust-lang.org/core/ptr/struct.NonNull.html#method.as_mut
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
///
/// # Examples
///
/// ```
/// # #[cfg(feature = "raw")]
/// # fn test() {
/// use core::hash::{BuildHasher, Hash};
/// use hashbrown::raw::{Bucket, RawTable};
///
/// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>;
///
/// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 {
/// use core::hash::Hasher;
/// let mut state = hash_builder.build_hasher();
/// key.hash(&mut state);
/// state.finish()
/// }
///
/// let hash_builder = NewHashBuilder::default();
/// let mut table = RawTable::new();
///
/// let value: (&str, String) = ("A pony", "is a small horse".to_owned());
/// let hash = make_hash(&hash_builder, &value.0);
///
/// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0));
///
/// let bucket: Bucket<(&str, String)> = table.find(hash, |(k, _)| k == &value.0).unwrap();
///
/// unsafe {
/// bucket
/// .as_mut()
/// .1
/// .push_str(" less than 147 cm at the withers")
/// };
/// assert_eq!(
/// unsafe { bucket.as_ref() },
/// &(
/// "A pony",
/// "is a small horse less than 147 cm at the withers".to_owned()
/// )
/// );
/// # }
/// # fn main() {
/// # #[cfg(feature = "raw")]
/// # test()
/// # }
/// ```
#[inline]
pub unsafe fn as_mut<'a>(&self) -> &'a mut T {
&mut *self.as_ptr()
}
/// Copies `size_of<T>` bytes from `other` to `self`. The source
/// and destination may *not* overlap.
///
/// # Safety
///
/// See [`ptr::copy_nonoverlapping`] for safety concerns.
///
/// Like [`read`], `copy_nonoverlapping` creates a bitwise copy of `T`, regardless of
/// whether `T` is [`Copy`]. If `T` is not [`Copy`], using *both* the values
/// in the region beginning at `*self` and the region beginning at `*other` can
/// [violate memory safety].
///
/// # Note
///
/// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match
/// those for the old `T` value, as the map will not re-evaluate where the new
/// value should go, meaning the value may become "lost" if their location
/// does not reflect their state.
///
/// [`ptr::copy_nonoverlapping`]: https://doc.rust-lang.org/core/ptr/fn.copy_nonoverlapping.html
/// [`read`]: https://doc.rust-lang.org/core/ptr/fn.read.html
/// [violate memory safety]: https://doc.rust-lang.org/std/ptr/fn.read.html#ownership-of-the-returned-value
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
#[cfg(feature = "raw")]
#[inline]
pub unsafe fn copy_from_nonoverlapping(&self, other: &Self) {
self.as_ptr().copy_from_nonoverlapping(other.as_ptr(), 1);
}
}
/// A raw hash table with an unsafe API.
pub struct RawTable<T, A: Allocator = Global> {
table: RawTableInner,
alloc: A,
// Tell dropck that we own instances of T.
marker: PhantomData<T>,
}
/// Non-generic part of `RawTable` which allows functions to be instantiated only once regardless
/// of how many different key-value types are used.
struct RawTableInner {
// Mask to get an index from a hash value. The value is one less than the
// number of buckets in the table.
bucket_mask: usize,
// [Padding], T1, T2, ..., Tlast, C1, C2, ...
// ^ points here
ctrl: NonNull<u8>,
// Number of elements that can be inserted before we need to grow the table
growth_left: usize,
// Number of elements in the table, only really used by len()
items: usize,
}
impl<T> RawTable<T, Global> {
/// Creates a new empty hash table without allocating any memory.
///
/// In effect this returns a table with exactly 1 bucket. However we can
/// leave the data pointer dangling since that bucket is never written to
/// due to our load factor forcing us to always have at least 1 free bucket.
#[inline]
pub const fn new() -> Self {
Self {
table: RawTableInner::NEW,
alloc: Global,
marker: PhantomData,
}
}
/// Attempts to allocate a new hash table with at least enough capacity
/// for inserting the given number of elements without reallocating.
#[cfg(feature = "raw")]
pub fn try_with_capacity(capacity: usize) -> Result<Self, TryReserveError> {
Self::try_with_capacity_in(capacity, Global)
}
/// Allocates a new hash table with at least enough capacity for inserting
/// the given number of elements without reallocating.
pub fn with_capacity(capacity: usize) -> Self {
Self::with_capacity_in(capacity, Global)
}
}
impl<T, A: Allocator> RawTable<T, A> {
const TABLE_LAYOUT: TableLayout = TableLayout::new::<T>();
/// Creates a new empty hash table without allocating any memory, using the
/// given allocator.
///
/// In effect this returns a table with exactly 1 bucket. However we can
/// leave the data pointer dangling since that bucket is never written to
/// due to our load factor forcing us to always have at least 1 free bucket.
#[inline]
pub const fn new_in(alloc: A) -> Self {
Self {
table: RawTableInner::NEW,
alloc,
marker: PhantomData,
}
}
/// Allocates a new hash table with the given number of buckets.
///
/// The control bytes are left uninitialized.
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn new_uninitialized(
alloc: A,
buckets: usize,
fallibility: Fallibility,
) -> Result<Self, TryReserveError> {
debug_assert!(buckets.is_power_of_two());
Ok(Self {
table: RawTableInner::new_uninitialized(
&alloc,
Self::TABLE_LAYOUT,
buckets,
fallibility,
)?,
alloc,
marker: PhantomData,
})
}
/// Attempts to allocate a new hash table using the given allocator, with at least enough
/// capacity for inserting the given number of elements without reallocating.
#[cfg(feature = "raw")]
pub fn try_with_capacity_in(capacity: usize, alloc: A) -> Result<Self, TryReserveError> {
Ok(Self {
table: RawTableInner::fallible_with_capacity(
&alloc,
Self::TABLE_LAYOUT,
capacity,
Fallibility::Fallible,
)?,
alloc,
marker: PhantomData,
})
}
/// Allocates a new hash table using the given allocator, with at least enough capacity for
/// inserting the given number of elements without reallocating.
pub fn with_capacity_in(capacity: usize, alloc: A) -> Self {
Self {
table: RawTableInner::with_capacity(&alloc, Self::TABLE_LAYOUT, capacity),
alloc,
marker: PhantomData,
}
}
/// Returns a reference to the underlying allocator.
#[inline]
pub fn allocator(&self) -> &A {
&self.alloc
}
/// Returns pointer to one past last `data` element in the table as viewed from
/// the start point of the allocation.
///
/// The caller must ensure that the `RawTable` outlives the returned [`NonNull<T>`],
/// otherwise using it may result in [`undefined behavior`].
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
pub fn data_end(&self) -> NonNull<T> {
// `self.table.ctrl.cast()` returns pointer that
// points here (to the end of `T0`)
// ∨
// [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, CTa_0, CTa_1, ..., CTa_m
// \________ ________/
// \/
// `n = buckets - 1`, i.e. `RawTable::buckets() - 1`
//
// where: T0...T_n - our stored data;
// CT0...CT_n - control bytes or metadata for `data`.
// CTa_0...CTa_m - additional control bytes, where `m = Group::WIDTH - 1` (so that the search
// with loading `Group` bytes from the heap works properly, even if the result
// of `h1(hash) & self.bucket_mask` is equal to `self.bucket_mask`). See also
// `RawTableInner::set_ctrl` function.
//
// P.S. `h1(hash) & self.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
// of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`.
self.table.ctrl.cast()
}
/// Returns pointer to start of data table.
#[inline]
#[cfg(any(feature = "raw", feature = "nightly"))]
pub unsafe fn data_start(&self) -> NonNull<T> {
NonNull::new_unchecked(self.data_end().as_ptr().wrapping_sub(self.buckets()))
}
/// Return the information about memory allocated by the table.
///
/// `RawTable` allocates single memory block to store both data and metadata.
/// This function returns allocation size and alignment and the beginning of the area.
/// These are the arguments which will be passed to `dealloc` when the table is dropped.
///
/// This function might be useful for memory profiling.
#[inline]
#[cfg(feature = "raw")]
pub fn allocation_info(&self) -> (NonNull<u8>, Layout) {
// SAFETY: We use the same `table_layout` that was used to allocate
// this table.
unsafe { self.table.allocation_info_or_zero(Self::TABLE_LAYOUT) }
}
/// Returns the index of a bucket from a `Bucket`.
#[inline]
pub unsafe fn bucket_index(&self, bucket: &Bucket<T>) -> usize {
bucket.to_base_index(self.data_end())
}
/// Returns a pointer to an element in the table.
///
/// The caller must ensure that the `RawTable` outlives the returned [`Bucket<T>`],
/// otherwise using it may result in [`undefined behavior`].
///
/// # Safety
///
/// If `mem::size_of::<T>() != 0`, then the caller of this function must observe the
/// following safety rules:
///
/// * The table must already be allocated;
///
/// * The `index` must not be greater than the number returned by the [`RawTable::buckets`]
/// function, i.e. `(index + 1) <= self.buckets()`.
///
/// It is safe to call this function with index of zero (`index == 0`) on a table that has
/// not been allocated, but using the returned [`Bucket`] results in [`undefined behavior`].
///
/// If `mem::size_of::<T>() == 0`, then the only requirement is that the `index` must
/// not be greater than the number returned by the [`RawTable::buckets`] function, i.e.
/// `(index + 1) <= self.buckets()`.
///
/// [`RawTable::buckets`]: RawTable::buckets
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[inline]
pub unsafe fn bucket(&self, index: usize) -> Bucket<T> {
// If mem::size_of::<T>() != 0 then return a pointer to the `element` in the `data part` of the table
// (we start counting from "0", so that in the expression T[n], the "n" index actually one less than
// the "buckets" number of our `RawTable`, i.e. "n = RawTable::buckets() - 1"):
//
// `table.bucket(3).as_ptr()` returns a pointer that points here in the `data`
// part of the `RawTable`, i.e. to the start of T3 (see `Bucket::as_ptr`)
// |
// | `base = self.data_end()` points here
// | (to the start of CT0 or to the end of T0)
// v v
// [Pad], T_n, ..., |T3|, T2, T1, T0, |CT0, CT1, CT2, CT3, ..., CT_n, CTa_0, CTa_1, ..., CTa_m
// ^ \__________ __________/
// `table.bucket(3)` returns a pointer that points \/
// here in the `data` part of the `RawTable` (to additional control bytes
// the end of T3) `m = Group::WIDTH - 1`
//
// where: T0...T_n - our stored data;
// CT0...CT_n - control bytes or metadata for `data`;
// CTa_0...CTa_m - additional control bytes (so that the search with loading `Group` bytes from
// the heap works properly, even if the result of `h1(hash) & self.table.bucket_mask`
// is equal to `self.table.bucket_mask`). See also `RawTableInner::set_ctrl` function.
//
// P.S. `h1(hash) & self.table.bucket_mask` is the same as `hash as usize % self.buckets()` because the number
// of buckets is a power of two, and `self.table.bucket_mask = self.buckets() - 1`.
debug_assert_ne!(self.table.bucket_mask, 0);
debug_assert!(index < self.buckets());
Bucket::from_base_index(self.data_end(), index)
}
/// Erases an element from the table without dropping it.
#[cfg_attr(feature = "inline-more", inline)]
unsafe fn erase_no_drop(&mut self, item: &Bucket<T>) {
let index = self.bucket_index(item);
self.table.erase(index);
}
/// Erases an element from the table, dropping it in place.
#[cfg_attr(feature = "inline-more", inline)]
#[allow(clippy::needless_pass_by_value)]
pub unsafe fn erase(&mut self, item: Bucket<T>) {
// Erase the element from the table first since drop might panic.
self.erase_no_drop(&item);
item.drop();
}
/// Finds and erases an element from the table, dropping it in place.
/// Returns true if an element was found.
#[cfg(feature = "raw")]
#[cfg_attr(feature = "inline-more", inline)]
pub fn erase_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> bool {
// Avoid `Option::map` because it bloats LLVM IR.
if let Some(bucket) = self.find(hash, eq) {
unsafe {
self.erase(bucket);
}
true
} else {
false
}
}
/// Removes an element from the table, returning it.
///
/// This also returns an `InsertSlot` pointing to the newly free bucket.
#[cfg_attr(feature = "inline-more", inline)]
#[allow(clippy::needless_pass_by_value)]
pub unsafe fn remove(&mut self, item: Bucket<T>) -> (T, InsertSlot) {
self.erase_no_drop(&item);
(
item.read(),
InsertSlot {
index: self.bucket_index(&item),
},
)
}
/// Finds and removes an element from the table, returning it.
#[cfg_attr(feature = "inline-more", inline)]
pub fn remove_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<T> {
// Avoid `Option::map` because it bloats LLVM IR.
match self.find(hash, eq) {
Some(bucket) => Some(unsafe { self.remove(bucket).0 }),
None => None,
}
}
/// Marks all table buckets as empty without dropping their contents.
#[cfg_attr(feature = "inline-more", inline)]
pub fn clear_no_drop(&mut self) {
self.table.clear_no_drop();
}
/// Removes all elements from the table without freeing the backing memory.
#[cfg_attr(feature = "inline-more", inline)]
pub fn clear(&mut self) {
if self.is_empty() {
// Special case empty table to avoid surprising O(capacity) time.
return;
}
// Ensure that the table is reset even if one of the drops panic
let mut self_ = guard(self, |self_| self_.clear_no_drop());
unsafe {
// SAFETY: ScopeGuard sets to zero the `items` field of the table
// even in case of panic during the dropping of the elements so
// that there will be no double drop of the elements.
self_.table.drop_elements::<T>();
}
}
/// Shrinks the table to fit `max(self.len(), min_size)` elements.
#[cfg_attr(feature = "inline-more", inline)]
pub fn shrink_to(&mut self, min_size: usize, hasher: impl Fn(&T) -> u64) {
// Calculate the minimal number of elements that we need to reserve
// space for.
let min_size = usize::max(self.table.items, min_size);
if min_size == 0 {
let mut old_inner = mem::replace(&mut self.table, RawTableInner::NEW);
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If any elements' drop function panics, then there will only be a memory leak,
// because we have replaced the inner table with a new one.
old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
return;
}
// Calculate the number of buckets that we need for this number of
// elements. If the calculation overflows then the requested bucket
// count must be larger than what we have right and nothing needs to be
// done.
let min_buckets = match capacity_to_buckets(min_size) {
Some(buckets) => buckets,
None => return,
};
// If we have more buckets than we need, shrink the table.
if min_buckets < self.buckets() {
// Fast path if the table is empty
if self.table.items == 0 {
let new_inner =
RawTableInner::with_capacity(&self.alloc, Self::TABLE_LAYOUT, min_size);
let mut old_inner = mem::replace(&mut self.table, new_inner);
unsafe {
// SAFETY:
// 1. We call the function only once;
// 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`]
// and [`TableLayout`] that were used to allocate this table.
// 3. If any elements' drop function panics, then there will only be a memory leak,
// because we have replaced the inner table with a new one.
old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT);
}
} else {
// Avoid `Result::unwrap_or_else` because it bloats LLVM IR.
unsafe {
// SAFETY:
// 1. We know for sure that `min_size >= self.table.items`.
// 2. The [`RawTableInner`] must already have properly initialized control bytes since
// we will never expose RawTable::new_uninitialized in a public API.
if self
.resize(min_size, hasher, Fallibility::Infallible)
.is_err()
{
// SAFETY: The result of calling the `resize` function cannot be an error
// because `fallibility == Fallibility::Infallible.
hint::unreachable_unchecked()
}
}
}
}
}
/// Ensures that at least `additional` items can be inserted into the table
/// without reallocation.
#[cfg_attr(feature = "inline-more", inline)]
pub fn reserve(&mut self, additional: usize, hasher: impl Fn(&T) -> u64) {
if unlikely(additional > self.table.growth_left) {
// Avoid `Result::unwrap_or_else` because it bloats LLVM IR.
unsafe {
// SAFETY: The [`RawTableInner`] must already have properly initialized control
// bytes since we will never expose RawTable::new_uninitialized in a public API.
if self
.reserve_rehash(additional, hasher, Fallibility::Infallible)
.is_err()
{
// SAFETY: All allocation errors will be caught inside `RawTableInner::reserve_rehash`.
hint::unreachable_unchecked()
}
}
}
}
/// Tries to ensure that at least `additional` items can be inserted into
/// the table without reallocation.
#[cfg_attr(feature = "inline-more", inline)]
pub fn try_reserve(
&mut self,
additional: usize,
hasher: impl Fn(&T) -> u64,
) -> Result<(), TryReserveError> {
if additional > self.table.growth_left {
// SAFETY: The [`RawTableInner`] must already have properly initialized control
// bytes since we will never expose RawTable::new_uninitialized in a public API.
unsafe { self.reserve_rehash(additional, hasher, Fallibility::Fallible) }
} else {
Ok(())
}
}
/// Out-of-line slow path for `reserve` and `try_reserve`.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes,
/// otherwise calling this function results in [`undefined behavior`]
///
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
#[cold]
#[inline(never)]
unsafe fn reserve_rehash(
&mut self,
additional: usize,
hasher: impl Fn(&T) -> u64,
fallibility: Fallibility,
) -> Result<(), TryReserveError> {
unsafe {
// SAFETY:
// 1. We know for sure that `alloc` and `layout` matches the [`Allocator`] and
// [`TableLayout`] that were used to allocate this table.
// 2. The `drop` function is the actual drop function of the elements stored in
// the table.
// 3. The caller ensures that the control bytes of the `RawTableInner`
// are already initialized.
self.table.reserve_rehash_inner(
&self.alloc,
additional,
&|table, index| hasher(table.bucket::<T>(index).as_ref()),
fallibility,
Self::TABLE_LAYOUT,
if T::NEEDS_DROP {
Some(mem::transmute(ptr::drop_in_place::<T> as unsafe fn(*mut T)))
} else {
None
},
)
}
}
/// Allocates a new table of a different size and moves the contents of the
/// current table into it.
///
/// # Safety
///
/// The [`RawTableInner`] must have properly initialized control bytes,
/// otherwise calling this function results in [`undefined behavior`]
///
/// The caller of this function must ensure that `capacity >= self.table.items`
/// otherwise:
///
/// * If `self.table.items != 0`, calling of this function with `capacity`
/// equal to 0 (`capacity == 0`) results in [`undefined behavior`].
///
/// * If `capacity_to_buckets(capacity) < Group::WIDTH` and
/// `self.table.items > capacity_to_buckets(capacity)`
/// calling this function results in [`undefined behavior`].
///
/// * If `capacity_to_buckets(capacity) >= Group::WIDTH` and
/// `self.table.items > capacity_to_buckets(capacity)`
/// calling this function are never return (will go into an
/// infinite loop).
///
/// See [`RawTableInner::find_insert_slot`] for more information.
///
/// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot
/// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html
unsafe fn resize(
&mut self,
capacity: usize,
hasher: impl Fn(&T) -> u64,
fallibility: Fallibility,
) -> Result<(), TryReserveError> {
// SAFETY:
// 1. The caller of this function guarantees that `capacity >= self.table.items`.
// 2. We know for sure that `alloc` and `layout` matches the [`Allocator`] and
// [`TableLayout`] that were used to allocate this table.
// 3. The caller ensures that the control bytes of the `RawTableInner`
// are already initialized.
self.table.resize_inner(
&self.alloc,
capacity,
&|table, index| hasher(table.bucket::<T>(index).as_ref()),
fallibility,
Self::TABLE_LAYOUT,
)
}
/// Inserts a new element into the table, and returns its raw bucket.
///
/// This does not check if the given element already exists in the table.
#[cfg_attr(feature = "inline-more", inline)]
pub fn insert(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> Bucket<T> {
unsafe {
// SAFETY:
// 1. The [`RawTableInner`] must already have properly initialized control bytes since
// we will never expose `RawTable::new_uninitialized` in a public API.
//
// 2. We reserve additional space (if necessary) right after calling this function.
let mut slot = self.table.find_insert_slot(hash);
// We can avoid growing the table once we have reached our load factor if we are replacing
// a tombstone. This works since the number of EMPTY slots does not change in this case.
//
// SAFETY: The function is guaranteed to return [`InsertSlot`] that contains an index
// in the range `0..=self.buckets()`.
let old_ctrl = *self.table.ctrl(slot.index);
if unlikely(self.table.growth_left == 0 && special_is_empty(old_ctrl)) {
self.reserve(1, hasher);
// SAFETY: We know for sure that `RawTableInner` has control bytes
// initialized and that there is extra space in the table.
slot = self.table.find_insert_slot(hash);
}
self.insert_in_slot(hash, slot, value)
}
}
/// Attempts to insert a new element without growing the table and return its raw bucket.
///
/// Returns an `Err` containing the given element if inserting it would require growing the
/// table.
///
/// This does not check if the given element already exists in the table.
#[cfg(feature = "raw")]
#[cfg_attr(feature = "inline-more", inline)]
pub fn try_insert_no_grow(&mut self, hash: u64, value: T) -> Result<Bucket<T>, T> {
unsafe {
match self.table.prepare_insert_no_grow(hash) {
Ok(index) => {
let bucket = self.bucket(index);
bucket.write(value);
Ok(bucket)
}
Err(()) => Err(value),
}
}
}
/// Inserts a new element into the table, and returns a mutable reference to it.
///
/// This does not check if the given element already exists in the table.
#[cfg_attr(feature = "inline-more", inline)]
pub fn insert_entry(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> &mut T {
unsafe { self.insert(hash, value, hasher).as_mut() }
}
/// Inserts a new element into the table, without growing the table.
///
/// There must be enough space in the table to insert the new element.
///
/// This does not check if the given element already exists in the table.
#[cfg_attr(feature = "inline-more", inline)]
#[cfg(any(feature = "raw", feature = "rustc-internal-api"))]
pub unsafe fn insert_no_grow(&mut self, hash: u64, value: T) -> Bucket<T> {
let (index, old_ctrl) = self.table.prepare_insert_slot(hash);
let bucket = self.table.bucket(index);
// If we are replacing a DELETED entry then we don't need to update
// the load counter.
self.table.growth_left -= special_is_empty(old_ctrl) as usize;
bucket.write(value);
self.table.items += 1;
bucket
}
/// Temporary removes a bucket, applying the given function to the removed
/// element and optionally put back the returned value in the same bucket.
///
/// Returns `true` if the bucket still contains an element
///
/// This does not check if the given bucket is actually occupied.
#[cfg_attr(feature = "inline-more", inline)]
pub unsafe fn replace_bucket_with<F>(&mut self, bucket: Bucket<T>, f: F) -> bool
where
F: FnOnce(T) -> Option<T>,
{
let index = self.bucket_index(&bucket);
let old_ctrl = *self.table.ctrl(index);
debug_assert!(self.is_bucket_full(index));
let old_growth_left = self.table.growth_left;
let item = self.remove(bucket).0;
if let Some(new_item) = f(item) {
self.table.growth_left = old_growth_left;
self.table.set_ctrl(index, old_ctrl);
self.table.items += 1;
self.bucket(index).write(new_item);
true
} else {
false
}
}
/// Searches for an element in the table. If the element is not found,
/// returns `Err` with the position of a slot where an element with the
/// same hash could be inserted.
///
/// This function may resize the table if additional space is required for
/// inserting an element.
#[inline]
pub fn find_or_find_insert_slot(
&mut self,
hash: u64,
mut eq: impl FnMut(&T) -> bool,
hasher: impl Fn(&T) -> u64,
) -> Result<Bucket<T>, InsertSlot> {
self.reserve(1, hasher);
unsafe {
// SAFETY:
// 1. We know for sure that there is at least one empty `bucket` in the table.
// 2. The [`RawTableInner`] must already have properly initialized control bytes since we will
// never expose `RawTable::new_uninitialized` in a public API.
// 3. The `find_or_find_insert_slot_inner` function returns the `index` of only the full bucket,
// which is in the range `0..self.buckets()` (since there is at least one empty `bucket` in
// the table), so calling `self.bucket(index)` and `Bucket::as_ref` is safe.
match self
.table
.find_or_find_insert_slot_inner(hash, &mut |index| eq(self.bucket(index).as_ref()))
{
// SAFETY: See explanation above.
Ok(index) => Ok(self.bucket(index)),
Err(slot) => Err(slot),
}
}
}
/// Inserts a new element into the table in the given slot, and returns its
/// raw bucket.
///
/// # Safety
///
/// `slot` must point to a slot previously returned by
/// `find_or_find_insert_slot`, and no mutation of the table must have
/// occurred since that call.
#[inline]
pub unsafe fn insert_in_slot(&mut self, hash: u64, slot: InsertSlot, value: T) -> Bucket<T> {
let old_ctrl = *self.table.ctrl(slot.index);
self.table.record_item_insert_at(slot.index, old_ctrl, hash);
let bucket = self.bucket(slot.index);
bucket.write(value);
bucket
}
/// Searches for an element in the table.
#[inline]
pub fn find(&self, hash: u64, mut eq: impl FnMut(&T) -> bool) -> Option<Bucket<T>> {
unsafe {
// SAFETY:
// 1. The [`RawTableInner`] must already have properly initialized control bytes since we
// will never expose `RawTable::new_uninitialized` in a public API.
// 1. The `find_inner` function returns the `index` of only the full bucket, which is in
// the range `0..self.buckets()`, so calling `self.bucket(index)` and `Bucket::as_ref`
// is safe.
let result = self
.table
.find_inner(hash, &mut |index| eq(self.bucket(index).as_ref()));
// Avoid `Option::map` because it bloats LLVM IR.
match result {
// SAFETY: See explanation above.
Some(index) => Some(self.bucket(index)),
None => None,
}
}
}
/// Gets a reference to an element in the table.
#[inline]
pub fn get(&self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<&T> {
// Avoid `Option::map` because it bloats LLVM IR.
match self.find(hash, eq) {
Some(bucket) => Some(unsafe { bucket.as_ref() }),
None => None,
}
}
/// Gets a mutable reference to an element in the table.
#[inline]
pub fn get_mut(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<&mut T> {
// Avoid `Option::map` because it bloats LLVM IR.
match self.find(hash, eq) {
Some(bucket) => Some(unsafe { bucket.as_mut() }),
None => None,
}
}
/// Attempts to get mutable references to `N` entries in the table at once.
///
/// Returns an array of length `N` with the results of each query.
///
/// At most one mutable reference will be returned to any entry. `None` will be returned if any
/// of the hashes are duplicates. `None` will be returned if the hash is not found.
///
/// The `eq` argument should be a closure such that `eq(i, k)` returns true if `k` is equal to
/// the `i`th key to be looked up.
pub fn get_many_mut<const N: usize>(
&mut self,
hashes: [u64; N],
eq: impl FnMut(usize, &T) -> bool,
) -> Option<[&'_ mut T; N]> {
unsafe {
let ptrs = self.get_many_mut_pointers(hashes, eq)?;
for (i, &cur) in ptrs.iter().enumerate() {
if ptrs[..i].iter().any(|&prev| ptr::eq::<T>(prev, cur)) {
return None;
}
}
// All bucket are distinct from all previous buckets so we're clear to return the result
// of the lookup.
// TODO use `MaybeUninit::array_assume_init` here instead once that's stable.
Some(mem::transmute_copy(&ptrs))
}
}
pub unsafe fn get_many_unchecked_mut<const N: usize>(
&mut self,
hashes: [u64; N],
eq: impl FnMut(usize, &T) -> bool,
) -> Option<[&'_ mut T; N]> {
let ptrs = self.get_many_mut_pointers(hashes, eq)?;
Some(mem::transmute_copy(&ptrs))
}
unsafe fn get_many_mut_pointers<const N: usize>(
&mut self,
hashes: [u64; N],
mut eq: impl FnMut(usize, &T) -> bool,
) -> Option<[*mut T; N]> {
// TODO use `MaybeUninit::uninit_array` here instead once that's stable.
let mut outs: MaybeUninit<[*mut T; N]> = MaybeUninit::uninit();
let outs_ptr = outs.as_mut_ptr();
for (i, &hash) in hashes.iter().enumerate() {
let cur = self.find(hash, |k| eq(i, k))?;
*(*outs_ptr).get_unchecked_mut(i) = cur.as_mut();
}
// TODO use `MaybeUninit::array_assume_init` here instead once that's stable.
Some(outs.assume_init())
}
/// Returns the number of elements the map can hold without reallocating.
///
/// This number is a lower bound; the table might be able to hold
/// more, but is guaranteed to be able to hold at least this many.
#[inline]
pub fn capacity(&self) -> usize {
self.table.items + self.table.growth_left
}
/// Returns the number of elements in the table.
#[inline]
pub fn len(&self) -> usize {
self.table.items
}
/// Returns `true` if the table contains no elements.
#[inline]
pub fn is_empty(&self) -> bool {
self.len() == 0
}
/// Returns the number of buckets in the table.
#[inline]
pub fn buckets(&self) -> usize {
self.table.bucket_mask + 1
}
/// Checks whether the bucket at `index` is full.
///
/// # Safety
///
/// The caller must ensure `index` is less than the number of buckets.
#[inline]
pub unsafe fn is_bucket_full(&self, index: usize) -> bool {
self.table.is_bucket_full(index)
}
/// Returns an iterator over every element in the table. It is up to
/// the caller to ensure that the `RawTable` outlives the `RawIter`.
/// Because we cannot make the `next` method unsafe on the `RawIter`
/// struct, we have to make the `iter` method unsafe.
#[inline]
pub unsafe fn iter(&self) -> RawIter<T> {
// SAFETY:
// 1. The caller must uphold the safety contract for `iter` method.
// 2. The [`RawTableInner`] must already have properly initialized control bytes since
// we will never expose RawTable::new_uninitialized in a public API.
self.table.iter()
}
--> --------------------
--> maximum size reached
--> --------------------
[ 0.80Quellennavigators
]
|
2026-04-02
|