Add a dedicated length-prefixing method to `Hasher`
This accomplishes two main goals:
- Make it clear who is responsible for prefix-freedom, including how they should do it
- Make it feasible for a `Hasher` that *doesn't* care about Hash-DoS resistance to get better performance by not hashing lengths
This does not change rustc-hash, since that's in an external crate, but that could potentially use it in future.
Fixes#94026
r? rust-lang/libs
---
The core of this change is the following two new methods on `Hasher`:
```rust
pub trait Hasher {
/// Writes a length prefix into this hasher, as part of being prefix-free.
///
/// If you're implementing [`Hash`] for a custom collection, call this before
/// writing its contents to this `Hasher`. That way
/// `(collection![1, 2, 3], collection![4, 5])` and
/// `(collection![1, 2], collection![3, 4, 5])` will provide different
/// sequences of values to the `Hasher`
///
/// The `impl<T> Hash for [T]` includes a call to this method, so if you're
/// hashing a slice (or array or vector) via its `Hash::hash` method,
/// you should **not** call this yourself.
///
/// This method is only for providing domain separation. If you want to
/// hash a `usize` that represents part of the *data*, then it's important
/// that you pass it to [`Hasher::write_usize`] instead of to this method.
///
/// # Examples
///
/// ```
/// #![feature(hasher_prefixfree_extras)]
/// # // Stubs to make the `impl` below pass the compiler
/// # struct MyCollection<T>(Option<T>);
/// # impl<T> MyCollection<T> {
/// # fn len(&self) -> usize { todo!() }
/// # }
/// # impl<'a, T> IntoIterator for &'a MyCollection<T> {
/// # type Item = T;
/// # type IntoIter = std::iter::Empty<T>;
/// # fn into_iter(self) -> Self::IntoIter { todo!() }
/// # }
///
/// use std:#️⃣:{Hash, Hasher};
/// impl<T: Hash> Hash for MyCollection<T> {
/// fn hash<H: Hasher>(&self, state: &mut H) {
/// state.write_length_prefix(self.len());
/// for elt in self {
/// elt.hash(state);
/// }
/// }
/// }
/// ```
///
/// # Note to Implementers
///
/// If you've decided that your `Hasher` is willing to be susceptible to
/// Hash-DoS attacks, then you might consider skipping hashing some or all
/// of the `len` provided in the name of increased performance.
#[inline]
#[unstable(feature = "hasher_prefixfree_extras", issue = "88888888")]
fn write_length_prefix(&mut self, len: usize) {
self.write_usize(len);
}
/// Writes a single `str` into this hasher.
///
/// If you're implementing [`Hash`], you generally do not need to call this,
/// as the `impl Hash for str` does, so you can just use that.
///
/// This includes the domain separator for prefix-freedom, so you should
/// **not** call `Self::write_length_prefix` before calling this.
///
/// # Note to Implementers
///
/// The default implementation of this method includes a call to
/// [`Self::write_length_prefix`], so if your implementation of `Hasher`
/// doesn't care about prefix-freedom and you've thus overridden
/// that method to do nothing, there's no need to override this one.
///
/// This method is available to be overridden separately from the others
/// as `str` being UTF-8 means that it never contains `0xFF` bytes, which
/// can be used to provide prefix-freedom cheaper than hashing a length.
///
/// For example, if your `Hasher` works byte-by-byte (perhaps by accumulating
/// them into a buffer), then you can hash the bytes of the `str` followed
/// by a single `0xFF` byte.
///
/// If your `Hasher` works in chunks, you can also do this by being careful
/// about how you pad partial chunks. If the chunks are padded with `0x00`
/// bytes then just hashing an extra `0xFF` byte doesn't necessarily
/// provide prefix-freedom, as `"ab"` and `"ab\u{0}"` would likely hash
/// the same sequence of chunks. But if you pad with `0xFF` bytes instead,
/// ensuring at least one padding byte, then it can often provide
/// prefix-freedom cheaper than hashing the length would.
#[inline]
#[unstable(feature = "hasher_prefixfree_extras", issue = "88888888")]
fn write_str(&mut self, s: &str) {
self.write_length_prefix(s.len());
self.write(s.as_bytes());
}
}
```
With updates to the `Hash` implementations for slices and containers to call `write_length_prefix` instead of `write_usize`.
`write_str` defaults to using `write_length_prefix` since, as was pointed out in the issue, the `write_u8(0xFF)` approach is insufficient for hashers that work in chunks, as those would hash `"a\u{0}"` and `"a"` to the same thing. But since `SipHash` works byte-wise (there's an internal buffer to accumulate bytes until a full chunk is available) it overrides `write_str` to continue to use the add-non-UTF-8-byte approach.
---
Compatibility:
Because the default implementation of `write_length_prefix` calls `write_usize`, the changed hash implementation for slices will do the same thing the old one did on existing `Hasher`s.
This accomplishes two main goals:
- Make it clear who is responsible for prefix-freedom, including how they should do it
- Make it feasible for a `Hasher` that *doesn't* care about Hash-DoS resistance to get better performance by not hashing lengths
This does not change rustc-hash, since that's in an external crate, but that could potentially use it in future.
Avoid using `rand::thread_rng` in the stdlib benchmarks.
This is kind of an anti-pattern because it introduces extra nondeterminism for no real reason. In thread_rng's case this comes both from the random seed and also from the reseeding operations it does, which occasionally does syscalls (which adds additional nondeterminism). The impact of this would be pretty small in most cases, but it's a good practice to avoid (particularly because avoiding it was not hard).
Anyway, several of our benchmarks already did the right thing here anyway, so the change was pretty easy and mostly just applying it more universally. That said, the stdlib benchmarks aren't particularly stable (nor is our benchmark framework particularly great), so arguably this doesn't matter that much in practice.
~~Anyway, this also bumps the `rand` dev-dependency to 0.8, since it had fallen somewhat out of date.~~ Nevermind, too much of a headache.
Tweak the vec-calloc runtime check to only apply to shortish-arrays
r? `@Mark-Simulacrum`
`@nbdd0121` pointed out in https://github.com/rust-lang/rust/pull/95362#issuecomment-1114085395 that LLVM currently doesn't constant-fold the `IsZero` check for long arrays, so that seems like a reasonable justification for limiting it.
It appears that it's based on length, not byte size, (https://godbolt.org/z/4s48Y81dP), so that's what I used in the PR. Maybe it's a ["the number of inlining shall be three"](https://youtu.be/s4wnuiCwTGU?t=320) sort of situation.
Certainly there's more that could be done here -- that generated code that checks long arrays byte-by-byte is highly suboptimal, for example -- but this is an easy, low-risk tweak.
std::fmt: Various fixes and improvements to documentation
This PR contains the following changes:
- **Added argument index comments to examples for specifying precision**
The examples for specifying the precision have comments explaining which
argument the specifier is referring to. However, for implicit positional
arguments, the examples simply refer to "next arg". To simplify following the
comments, "next arg" was supplemented with the actual resulting argument index.
- **Fixed documentation for specifying precision via `.*`**
The documentation stated that in case of the syntax `{<arg>:<spec>.*}`, "the
`<arg>` part refers to the value to print, and the precision must come in the
input preceding `<arg>`". This is not correct: the <arg> part does indeed refer
to the value to print, but the precision does not come in the input preciding
arg, but in the next implicit input (as if specified with {}).
Fixes#96413.
- **Fix the grammar documentation**
According to the grammar documented, the format specifier `{: }` should not be
legal because of the whitespace it contains. However, in reality, this is
perfectly fine because the actual implementation allows spaces before the
closing brace. Fixes#71088.
Also, the exact meaning of most of the terminal symbols was not specified, for
example the meaning of `identifier`.
- **Removed reference to Formatter::buf and other private fields**
Formatter::buf is not a public field and therefore isn't very helpful in user-
facing documentation. Also, the other public fields of Formatter were removed
during stabilization of std::fmt (4af3494bb0) and can only be accessed via
getters.
- **Improved list of formatting macros**
Two improvements:
1. write! can not only receive a `io::Write`, but also a `fmt::Write` as first argument.
2. The description texts now contain links to the actual macros for easier
navigation.
Clarify docs for `from_raw_parts` on `Vec` and `String`
Closes#95427
Original safety explanation for `from_raw_parts` was unclear on safety for consuming a C string. This clarifies when doing so is safe.
Classify BinaryHeap & LinkedList unit tests as such
All but one of these so-called integration test case are unit tests, just like btree's were (#75531). In addition, reunite the unit tests of linked_list that were split off during #23104 because they needed to remain unit tests (they were later moved to the separate file they are in during #63207). The two sets could remain separate files, but I opted to merge them back together, more or less in the order they used to be, apart from one duplicate name `test_split_off` and one duplicate tiny function `list_from`.
Two improvements:
1. write! can not only receive a `io::Write`, but also a `fmt::Write` as first argument.
2. The description texts now contain links to the actual macros for easier
navigation.
Formatter::buf is not a public field and therefore isn't very helpful in user-
facing documentation. Also, the other public fields of Formatter were made
private during stabilization of std::fmt (4af3494bb0) and can now only be read
via accessor methods.
According to the grammar documented, the format specifier `{: }` should not be
legal because of the whitespace it contains. However, in reality, this is
perfectly fine because the actual implementation allows spaces before the
closing brace. Fixes#71088.
Also, the exact meaning of most of the terminal symbols was not specified, for
example the meaning of `identifier`.
The examples for specifying the precision have comments explaining which
argument the specifier is referring to. However, for implicit positional
arguments, the examples simply talk about "next arg". To make it easier for
readers to follow the comments, "next arg" was supplemented with the actual
resulting argument index.
The documentation stated that in case of the syntax `{<arg>:<spec>.*}`, "the
`<arg>` part refers to the value to print, and the precision must come in the
input preceding `<arg>`". This is not correct: the <arg> part does indeed refer
to the value to print, but the precision does not come in the input preciding
arg, but in the next implicit input (as if specified with {}).
Fixes#96413.
Implement str to [u8] conversion for refcounted containers
This seems motivated to complete the APIs for shared containers since we already have similar allocation-free conversions for strings like `From<Box<[u8]>> for Box<str>`.
Insta-stable since it's a new trait impl?
Revert "Re-export core::ffi types from std::ffi"
This reverts commit 9aed829fe6.
Fixes https://github.com/rust-lang/rust/issues/96435 , a regression
in crates doing `use std::ffi::*;` and `use std::os::raw::*;`.
We can re-add this re-export once the `core::ffi` types
are stable, and thus the `std::os::raw` types can become re-exports as
well, which will avoid the conflict. (Type aliases to the same type
still conflict, but re-exports of the same type don't.)
Clarify that `Cow::into_owned` returns owned data
Two sections of the `Cow::into_owned` docs imply that `into_owned` returns a `Cow`. Clarify that it returns the underlying owned object, either cloned or extracted from the `Cow`.
Fix some confusing wording and improve slice-search-related docs
This adds more links between `contains` and `binary_search` because I do think they have some relevant connections. If your (big) slice happens to be sorted and you know it, surely you should be using `[3; 100].binary_search(&5).is_ok()` over `[3; 100].contains(&5)`?
This also fixes the confusing "searches this sorted X" wording which just sounds really weird because it doesn't know whether it's actually sorted. It should be but it may not be. The new wording should make it clearer that you will probably want to sort it and in the same sentence it also mentions the related function `contains`.
Similarly, this mentions `binary_search` on `contains`' docs.
This also fixes some other minor stuff and inconsistencies.
[test] Add test cases for untested functions for VecDeque
Added test cases of the following functions
- get
- get_mut
- swap
- reserve_exact
- try_reserve_exact
- try_reserve
- contains
- rotate_left
- rotate_right
- binary_search
- binary_search_by
- binary_search_by_key
`alloc`: make `vec!` unavailable under `no_global_oom_handling`
`alloc`: make `vec!` unavailable under `no_global_oom_handling`
The `vec!` macro has 3 rules, but two are not usable under
`no_global_oom_handling` builds of the standard library
(even with a zero size):
```rust
let _ = vec![42]; // Error: requires `exchange_malloc` lang_item.
let _ = vec![42; 0]; // Error: cannot find function `from_elem`.
```
Thus those two rules should not be available to begin with.
The remaining one, with an empty matcher, is just a shorthand for
`new()` and may not make as much sense to have alone, since the
idea behind `vec!` is to enable `Vec`s to be defined with the same
syntax as array expressions. Furthermore, the documentation can be
confusing since it shows the other rules.
Thus perhaps it is better and simpler to disable `vec!` entirely
under `no_global_oom_handling` environments, and let users call
`new()` instead:
```rust
let _: Vec<i32> = vec![];
let _: Vec<i32> = Vec::new();
```
Notwithstanding this, a `try_vec!` macro would be useful, such as
the one introduced in https://github.com/rust-lang/rust/pull/95051.
If the shorthand for `new()` is deemed worth keeping on its own,
then it may be interesting to have a separate `vec!` macro with
a single rule and different, simpler documentation.
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
Speed up Vec::clear().
Currently it just calls `truncate(0)`. `truncate()` is (a) not marked as
`#[inline]`, and (b) more general than needed for `clear()`.
This commit changes `clear()` to do the work itself. This modest change
was first proposed in rust-lang#74172, where the reviewer rejected it because
there was insufficient evidence that `Vec::clear()`'s performance
mattered enough to justify the change. Recent changes within rustc have
made `Vec::clear()` hot within `macro_parser.rs`, so the change is now
clearly worthwhile.
Although it doesn't show wins on CI perf runs, this seems to be because they
use PGO. But not all platforms currently use PGO. Also, local builds don't use
PGO, and `truncate` sometimes shows up in an over-represented fashion in local
profiles. So local profiling will be made easier by this change.
Note that this will also benefit `String::clear()`, because it just
calls `Vec::clear()`.
Finally, the commit removes the `vec-clear.rs` codegen test. It was
added in #52908. From before then until now, `Vec::clear()` just called
`Vec::truncate()` with a zero length. The body of Vec::truncate() has
changed a lot since then. Now that `Vec::clear()` is doing actual work
itself, and not just calling `Vec::truncate()`, it's not surprising that
its generated code includes a load and an icmp. I think it's reasonable
to remove this test.
r? `@m-ou-se`
The `vec!` macro has 3 rules, but two are not usable under
`no_global_oom_handling` builds of the standard library
(even with a zero size):
```rust
let _ = vec![42]; // Error: requires `exchange_malloc` lang_item.
let _ = vec![42; 0]; // Error: cannot find function `from_elem`.
```
Thus those two rules should not be available to begin with.
The remaining one, with an empty matcher, is just a shorthand for
`new()` and may not make as much sense to have alone, since the
idea behind `vec!` is to enable `Vec`s to be defined with the same
syntax as array expressions. Furthermore, the documentation can be
confusing since it shows the other rules.
Thus perhaps it is better and simpler to disable `vec!` entirely
under `no_global_oom_handling` environments, and let users call
`new()` instead:
```rust
let _: Vec<i32> = vec![];
let _: Vec<i32> = Vec::new();
```
Notwithstanding this, a `try_vec!` macro would be useful, such as
the one introduced in https://github.com/rust-lang/rust/pull/95051.
If the shorthand for `new()` is deemed worth keeping on its own,
then it may be interesting to have a separate `vec!` macro with
a single rule and different, simpler documentation.
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
Optimize RcInnerPtr::inc_strong()/inc_weak() instruction count
Inspired by this internals thread: https://internals.rust-lang.org/t/rc-optimization-on-64-bit-targets/16362
[The generated assembly is a bit smaller](https://rust.godbolt.org/z/TeTnf6144) and is a more efficient usage of the CPU's instruction cache. `unlikely` doesn't impact any of the small artificial tests I've done, but I've included it in case it might help more complex scenarios when this is inlined.
Currently it just calls `truncate(0)`. `truncate()` is (a) not marked as
`#[inline]`, and (b) more general than needed for `clear()`.
This commit changes `clear()` to do the work itself. This modest change
was first proposed in rust-lang#74172, where the reviewer rejected it because
there was insufficient evidence that `Vec::clear()`'s performance
mattered enough to justify the change. Recent changes within rustc have
made `Vec::clear()` hot within `macro_parser.rs`, so the change is now
clearly worthwhile.
Although it doesn't show wins on CI perf runs, this seems to be because they
use PGO. But not all platforms currently use PGO. Also, local builds don't use
PGO, and `truncate` sometimes shows up in an over-represented fashion in local
profiles. So local profiling will be made easier by this change.
Note that this will also benefit `String::clear()`, because it just
calls `Vec::clear()`.
Finally, the commit removes the `vec-clear.rs` codegen test. It was
added in #52908. From before then until now, `Vec::clear()` just called
`Vec::truncate()` with a zero length. The body of Vec::truncate() has
changed a lot since then. Now that `Vec::clear()` is doing actual work
itself, and not just calling `Vec::truncate()`, it's not surprising that
its generated code includes a load and an icmp. I think it's reasonable
to remove this test.