This commit deletes the `alloc_system` crate from the standard
distribution. This unstable crate is no longer needed in the modern
stable global allocator world, but rather its functionality is folded
directly into the standard library. The standard library was already the
only stable location to access this crate, and as a result this should
not affect any stable code.
Incorporate a stray test
`liballoc/repeat-generic-slice.rs` doesn't seem to be tested (I think it was intended to be placed in `run-pass`). This PR incorporates the test into `liballoc/tests`.
Add slice::sort_by_cached_key as a memoised sort_by_key
At present, `slice::sort_by_key` calls its key function twice for each comparison that is made. When the key function is expensive (which can often be the case when `sort_by_key` is chosen over `sort_by`), this can lead to very suboptimal behaviour.
To address this, I've introduced a new slice method, `sort_by_cached_key`, which has identical semantic behaviour to `sort_by_key`, except that it guarantees the key function will only be called once per element.
Where there are `n` elements and the key function is `O(m)`:
- `slice::sort_by_cached_key` time complexity is `O(m n log m n)`, compared to `slice::sort_by_key`'s `O(m n + n log n)`.
- `slice::sort_by_cached_key` space complexity remains at `O(n + m)`. (Technically, it now reserves a slice of size `n`, whereas before it reserved a slice of size `n/2`.)
`slice::sort_unstable_by_key` has not been given an analogue, as it is important that unstable sorts are in-place, which is not a property that is guaranteed here. However, this also means that `slice::sort_unstable_by_key` is likely to be slower than `slice::sort_by_cached_key` when the key function does not have negligible complexity. We might want to explore this trade-off further in the future.
Benchmarks (for a vector of 100 `i32`s):
```
# Lexicographic: `|x| x.to_string()`
test bench_sort_by_key ... bench: 112,638 ns/iter (+/- 19,563)
test bench_sort_by_cached_key ... bench: 15,038 ns/iter (+/- 4,814)
# Identity: `|x| *x`
test bench_sort_by_key ... bench: 1,346 ns/iter (+/- 238)
test bench_sort_by_cached_key ... bench: 1,839 ns/iter (+/- 765)
# Power: `|x| x.pow(31)`
test bench_sort_by_key ... bench: 3,624 ns/iter (+/- 738)
test bench_sort_by_cached_key ... bench: 1,997 ns/iter (+/- 311)
# Abs: `|x| x.abs()`
test bench_sort_by_key ... bench: 1,546 ns/iter (+/- 174)
test bench_sort_by_cached_key ... bench: 1,668 ns/iter (+/- 790)
```
(So it seems functions that are single operations do perform slightly worse with this method, but for pretty much any more complex key, you're better off with this optimisation.)
I've definitely found myself using expensive keys in the past and wishing this optimisation was made (e.g. for https://github.com/rust-lang/rust/pull/47415). This feels like both desirable and expected behaviour, at the small cost of slightly more stack allocation and minute degradation in performance for extremely trivial keys.
Resolves#34447.
Stabilize inclusive range (`..=`)
Stabilize the followings:
* `inclusive_range` — The `std::ops::RangeInclusive` and `std::ops::RangeInclusiveTo` types, except its fields (tracked by #49022 separately).
* `inclusive_range_syntax` — The `a..=b` and `..=b` expression syntax
* `dotdoteq_in_patterns` — Using `a..=b` in a pattern
cc #28237
r? @rust-lang/lang
The GNU C library (glibc) is documented to always allocate with an alignment
of at least 8 or 16 bytes, on 32-bit or 64-bit platforms:
https://www.gnu.org/software/libc/manual/html_node/Aligned-Memory-Blocks.html
This matches our use of `MIN_ALIGN` before this commit.
However, even when libc is glibc, the program might be linked
with another allocator that redefines the `malloc` symbol and friends.
(The `alloc_jemalloc` crate does, in some cases.)
So `alloc_system` doesn’t know which allocator it calls,
and needs to be conservative in assumptions it makes.
The C standard says:
https://port70.net/%7Ensz/c/c11/n1570.html#7.22.3
> The pointer returned if the allocation succeeds is suitably aligned
> so that it may be assigned to a pointer to any type of object
> with a fundamental alignment requirement
https://port70.net/~nsz/c/c11/n1570.html#6.2.8p2
> A fundamental alignment is represented by an alignment less than
> or equal to the greatest alignment supported by the implementation
> in all contexts, which is equal to `_Alignof (max_align_t)`.
`_Alignof (max_align_t)` depends on the ABI and doesn’t seem to have
a clear definition, but it seems to match our `MIN_ALIGN` in practice.
However, the size of objects is rounded up to the next multiple
of their alignment (since that size is also the stride used in arrays).
Conversely, the alignment of a non-zero-size object is at most its size.
So for example it seems ot be legal for `malloc(8)` to return a pointer
that’s only 8-bytes-aligned, even if `_Alignof (max_align_t)` is 16.
This commit removes the `rand` crate from the standard library facade as
well as the `__rand` module in the standard library. Neither of these
were used in any meaningful way in the standard library itself. The only
need for randomness in libstd is to initialize the thread-local keys of
a `HashMap`, and that unconditionally used `OsRng` defined in the
standard library anyway.
The cruft of the `rand` crate and the extra `rand` support in the
standard library makes libstd slightly more difficult to port to new
platforms, namely WebAssembly which doesn't have any randomness at all
(without interfacing with JS). The purpose of this commit is to clarify
and streamline randomness in libstd, focusing on how it's only required
in one location, hashmap seeds.
Note that the `rand` crate out of tree has almost always been a drop-in
replacement for the `rand` crate in-tree, so any usage (accidental or
purposeful) of the crate in-tree should switch to the `rand` crate on
crates.io. This then also has the further benefit of avoiding
duplication (mostly) between the two crates!
Add Vec::drain_filter
This implements the API proposed in #43244.
So I spent like half a day figuring out how to implement this in some awesome super-optimized unsafe way, which had me very confident this was worth putting into the stdlib.
Then I looked at the impl for `retain`, and was like "oh dang". I compared the two and they basically ended up being the same speed. And the `retain` impl probably translates to DoubleEndedIter a lot more cleanly if we ever want that.
So now I'm not totally confident this needs to go in the stdlib, but I've got two implementations and an amazingly robust test suite, so I figured I might as well toss it over the fence for discussion.
Add method `String::retain`
Behaves like `Vec::retain`, accepting a predicate `FnMut(char) -> bool`
and reducing the string to only characters for which the predicate
returns `true`.
Behaves like `Vec::retain`, accepting a predicate `FnMut(char) -> bool`
and reducing the string to only characters for which the predicate
returns `true`.
Alex Crichton