According to `[mem.poly.allocator.ctor]` the pointer contained in
`polymorphic_allocator` can never be null. The default constructor uses
`get_default_resource()`, which never returns null and the constructor
taking a pointer explicitly has a precondition that the pointer is
non-null.
This patch adds a warning and an assertion in case a user passes a null
pointer to `polymorphic_allocator` as well as marking `resource()` to
never return null.
This also fixes some tests which contained UB.
Fixes#148420
This was probably added to support https://wg21.link/n3644 but there's
no reason not to initialize the pointer in all standard modes. This is
technically an extension since n3644 only required value-initialized
iterators to be comparable, but supporting this as an extension should
be uncontroversial since it avoids potential reads of uninitialized
memory in C++03/C++11 without doing any harm.
Both Clang and GCC diagnose invalid calls to `__builtin_launder`, which
causes duplicate diagnostics when using `std::launder` in an invalid
way. While the diagnostic message for the builtin isn't perferct, it's
definitely good enough to understand the problem and adding our own
diagnostic doesn't really make things any clearer. Because of that, this
patch simply removes the `static_assert`s and lets the compiler handle
diagnosing incorrect arguments instead. This not only simplifies our
implementation, but also improves compile times a bit, since we avoid
instantiating some type traits.
Libc++'s policy is to support only the latest released Xcode, which is
Xcode 16.x. We did update our CI jobs to Xcode 16.x, but we forgot to
update the documentation, which still mentioned Xcode 15. This patch
updates the documentation and cleans up outdated mentions of
apple-clang-15 in the test suite.
Reverts llvm/llvm-project#148266
I'm reverting this temporarily, since the release branch is today and
this is ABI sensitive. Let's wait until after the branch so that we have
plenty time to discuss the patch.
GCC 15 also supports `__buitin_operator_{new,delete}` now, so the
`#else` cases are dead code. This patch inlines the calls to the wrapper
functions and simplifies some surrounding code.
Assertion semantics closely mimic C++26 Contracts evaluation semantics.
This brings our implementation closer in line with C++26 Library
Hardening (one particular benefit is that using the `observe` semantic
makes adopting hardening easier for projects).
Unlike `verbose_abort`, this function merely logs the error but does not
terminate execution. It is intended to make it possible to implement the
`observe` semantic for Hardening.
Currently, we try to instantiate the allocator on `__hash_value_type`,
which we don't define anymore. Instead, just use the
`map::allocator_type` to instantiate `__tree`, since that's what we
actually want anyways.
This fixes `insert()` calling the wrong `allocator_traits::construct` in
the associative containers by removing the special handling that lead to
the inconsistencty inside `__tree` and `__hash_table`.
Clang treats throwing/catching ObjC types differently from C++ types,
and omitting the `throw` in `std::make_exception_ptr` breaks ObjC
invariants about how types are thrown/caught.
Fixes#135089
Co-authored-by: Louis Dionne <ldionne.2@gmail.com>
Fixes#105260
This patch applies the change as a DR to C++20. The rationale is that
the paper is more like a bug fix. It does not introduce new features, it
simply changes an existing behaviour (as a bug fix). MSVC STL DRed this
paper to C++20 as well.
For these annotations to do anything you need `-Wthread-safety`, in
which case users most likely enable them anyways. This avoids that users
have to explictly define a macro just to use the feature they already
had to opt-in to.
The use-case for `__is_same_uncvref` seems rather dubious, since not a
single use-cases needed the `remove_cvref_t` to be applied to both of
the arguments. Removing the alias makes it clearer what actually
happens, since we're not using an internal name anymore and it's clear
what the `remove_cvref_t` should apply to.
`vector` has been used in a very simple way in `boyer_moore_searcher`.
We can instead just use `unique_ptr<T[]>`, which is a lot simpler,
allowing us to drop the `vector` dependency while not losing any
expressiveness in the code. As a nice side effect, this also reduces the
time it takes to instantiate the `boyer_moore_searcher` constructor from
26ms to 22ms on my machine.
libc++ doesn't run clang-tidy in the bootstrapping build, so it wasn't
run when `__builtin_invoke` was introduced. Once libc++ upgrades to a
newer clang version, clang-tidy will start to fail due to missing
`_LIBCPP_NODEBUG` annotations.
`<charconv>` doesn't properly export `std::errc` with locales disabled,
which causes the CI to fail. This fixed the modulemap to let
`from_chars_result` export `std::errc`, like it's supposed to do.
`std::invoke` is currently quite heavy compared to a function call,
since it involves quite heavy SFINAE. This can be done significantly
more efficient by the compiler, since most calls to `std::invoke` are
simple function calls and 6 out of the seven overloads for `std::invoke`
exist only to support member pointers. Even these boil down to a few
relatively simple checks.
Some real-world testing with this patch revealed some significant
results. For example, instantiating `std::format("Banane")` (and its
callees) went down from ~125ms on my system to ~104ms.
Fixes#108624
This allows `flat_map::insert(Iter, Iter)` to directly forward to
underlying containers' `insert(Iter, Iter)`, instead of inserting one
element at a time, when input models "product iterator". atm,
`flat_map::iterator` and `zip_view::iterator` are "product iterator"s.
This gives about almost 10x speed up in my benchmark with -03 (for both
before and after)
```cpp
Benchmark Time CPU Time Old Time New CPU Old CPU New
-----------------------------------------------------------------------------------------------------------------------------------------------
flat_map::insert_product_iterator_flat_map/32 -0.5028 -0.5320 149 74 149 70
flat_map::insert_product_iterator_flat_map/1024 -0.8617 -0.8618 3113 430 3112 430
flat_map::insert_product_iterator_flat_map/8192 -0.8877 -0.8877 26682 2995 26679 2995
flat_map::insert_product_iterator_flat_map/65536 -0.8769 -0.8769 226235 27844 226221 27841
flat_map::insert_product_iterator_zip/32 -0.5844 -0.5844 162 67 162 67
flat_map::insert_product_iterator_zip/1024 -0.8754 -0.8754 3427 427 3427 427
flat_map::insert_product_iterator_zip/8192 -0.8934 -0.8934 28134 3000 28132 3000
flat_map::insert_product_iterator_zip/65536 -0.8783 -0.8783 229783 27960 229767 27958
OVERALL_GEOMEAN -0.8319 -0.8332 0 0 0 0
```
---------
Co-authored-by: Louis Dionne <ldionne.2@gmail.com>
Most of the diff is replacing `__parent_pointer` with
`__end_node_pointer`. The most interesting diff is that the pointer
aliases are now defined directly inside `__tree` instead of a separate
traits class.
This patch is very similar to #134819 in nature. Before this patch, we
were dereferencing pointers to objects which were never constructed. Now
we always assume that nodes store `pair<const KeyT, ValueT>` for
unordered_maps instead, as they actually do.
If the creation of a thread fails, this causes an idle loop that will
never end because the thread wasn't started in the first place.
This also adds a test for the regression reported in #125433 to make
sure we're not reintroducing it later.
Fixes#125428
Refactor the function to streamline the logic so it matches the
specification in [filebuf.virtuals] more closely. In particular, avoid
modifying the put area pointers when we loop around after a partial
codecvt conversion.
Note that we're technically not up-to-spec in this implementation, since
the Standard says that we shouldn't try more than once after a partial
codecvt conversion. However, this refactoring attempts not to change any
functionality.
This patch addresses several implementation issues in `bitset`'s
conversion functions `to_ullong` and `to_ulong`, and refactors its
converting constructor `__bitset(unsigned long long __v)` to a more
generic and elegant implementation.
