This is a spin-off from D79555 review, that with this patch will be able to use `__libcpp_copysign` instead of adhoc `__copysign_constexpr` helper.
Reviewed By: ldionne, #libc
Differential Revision: https://reviews.llvm.org/D106364
This replaces _LIBCPP_INLINE_VISIBILITY with _LIBCPP_HIDE_FROM_ABI. It's
not intended to do for other parts of libc++. This change makes it easy
to search and replace all occurrences of the patches in review.
We've been forgetting to add those to most of the <ranges> review.
To avoid forgetting in the future, I added an item in the pre-commit
checklist.
Differential Revision: https://reviews.llvm.org/D106287
The new testing configuration did not turn off #pragma system_header,
which means we were not seeing warnings in system headers.
Differential Revision: https://reviews.llvm.org/D106187
libc++ has started splicing standard library headers into much more
fine-grained content for maintainability. It's very likely that outdated
and naive tooling (some of which is outside of LLVM's scope) will
suggest users include things such as `<__algorithm/find.h>` instead of
`<algorithm>`, and Hyrum's law suggests that users will eventually begin
to rely on this without the help of tooling. As such, this commit
intends to protect users from themselves, by making it a hard error for
anyone outside of the standard library to include libc++ detail headers.
This is the first of four patches. Patch #2 will solve the problem for
pre-processor `#include`s; patches #3 and #4 will solve the problem for
`<__tree>` and `<__hash_table>` (since I've never touched the test cases
that are failing for these two, I want to split them out into their own
commits to be extra careful). Patch #5 will concern itself with
`<__threading_support>`, which intersects with libcxxabi (which I know
even less about).
Differential Revision: https://reviews.llvm.org/D105932
The checks within the libc++experimental memory_resource class uses this
limit:
_MaxAlign = _LIBCPP_ALIGNOF(max_align_t);
Therefore, only use max_align_t for this limit instead of using
`__STDCPP_DEFAULT_NEW_ALIGNMENT__` if available.
Differential Revision: https://reviews.llvm.org/D105905
* <__algorithm/iter_swap.h>
* <__algorithm/swap_ranges.h>
* <__functional/is_transparent.h>
* <__memory/uses_allocator.h>
* <__ranges/drop_view.h>
* <__ranges/transform_view.h>
* <shared_mutex>
* <span>
Also updates header inclusions that were affected.
**NOTE:** This is a proper subset of D105932. Since the content has
already been LGTM'd, I intend to merge this patch without review,
pending green CI. I decided it would be better to move these changes
into their own commit since the former patch has undergone further
changes and will need yet another light review. In the event any of
that gets rolled back (for whatever reason), the changes in this patch
won't be affected.
Differential Revision: https://reviews.llvm.org/D106040
Instead of using TARGET_TRIPLE, which is always set to LLVM_DEFAULT_TARGET_TRIPLE,
use that variable directly to populate the various XXXX_TARGET_TRIPLE
variables in the runtimes.
This re-applies 77396bbc98 and 5099e01568, which were reverted in
850b57c5fb because they broke the build.
Differential Revision: https://reviews.llvm.org/D106009
This configuration is interesting because GCC has a different level of
strictness for some C++ rules. In particular, it implements the older
standards more stringently than Clang, which can help find places where
we are non-conforming (especially in the test suite).
Differential Revision: https://reviews.llvm.org/D105936
When a target triple is specified in CMake via XXX_TARGET_TRIPLE, we tried
passing the --target=<...> flag to the compiler. However, not all compilers
support that flag (e.g. GCC, which is not a cross-compiler). As a result,
setting e.g. LIBCXX_TARGET_TRIPLE=<host-triple> would end up trying to
pass --target=<host-triple> to GCC, which breaks everything because the
flag isn't even supported.
This commit only adds `--target=<...>` & friends to the flags if it is
supported by the compiler.
One could argue that it's confusing to pass LIBCXX_TARGET_TRIPLE=<...>
and have it be ignored. That's correct, and one possibility would be
to assert that the requested triple is the same as the host triple when
we know the compiler is unable to cross-compile. However, note that this
is a pre-existing issue (setting the TARGET_TRIPLE variable never had an
influence on the flags passed to the compiler), and also fixing that is
starting to look like reimplementing a lot of CMake logic that is already
handled with CMAKE_CXX_COMPILER_TARGET.
Differential Revision: https://reviews.llvm.org/D106082
The XFAIL comments about VCRuntime not providing aligned operator new
are outdated; these days VCRuntime does provide them.
However, the tests used to fail on Windows, as the pointers allocated
with an aligned operator new (which is implemented with _aligned_malloc
on Windows) can't be freed using std::free() on Windows (but they need
to be freed with the corresponding function _aligned_free instead).
Instead override the aligned operator new to return a dummy suitably
aligned pointer instead, like other tests that override aligned operator
new.
Also override `operator delete[]` instead of plain `operator delete`
in the array testcase; the fallback from `operator delete[]` to
user defined `operator delete` doesn't work in all DLL build
configurations on Windows.
Also expand the TEST_NOEXCEPT macros, as these tests only are built
in C++17 mode.
By providing the aligned operator new within the tests, this also makes
these test cases pass when testing back deployment on macOS 10.9.
Differential Revision: https://reviews.llvm.org/D105962
The feature was always defined, which means that the two test cases
guarded by it were never run.
Differential Revision: https://reviews.llvm.org/D106062
As we automate more and more things in the library, it becomes useful for
contributors to have a single target for running all the automation as
part of their workflow. This commit adds a new `libcxx-generate-files`
target that should re-generate all the auto-generated files in the library.
As a fly-by, I also revamped the documentation on Contributing to account
for this new target and present it as a bullet list of things to check
before committing. I also added a few things that are often overlooked
to that list, such as updating the synopsis and the status files.
Differential Revision: https://reviews.llvm.org/D106067
add_lit_testsuite() takes Lit parameters passed to it and adds them
to the parameters used globally when running all test suites. That
means that a target like `check-all`, which ends up calling Lit on
the whole monorepo, will see the test parameters for all the individual
project's test suites.
So, for example, it would see `--param std=c++03` (from libc++abi), and
`--param std=c++03` (from libc++), and `--param whatever` (from another
project being tested at the same time). While always unclean, that works
when the parameters all agree. However, if the parameters share the same
name but have different values, only one of those two values will be used
and it will be incredibly confusing to understand why one of the test
suites is being run with the incorrect parameter value.
For that reason, this commit moves away from using add_lit_testsuite()'s
PARAM functionality, and serializes the parameter values for the runtimes
in the generated config.py file instead, which is local to the specific
test suite.
Differential Revision: https://reviews.llvm.org/D105991
It turns out that D105040 broke `std::rel_ops`; we actually do need
both a one-template-parameter and a two-template-parameter version of
all the comparison operators, because if we have only the heterogeneous
two-parameter version, then `x > x` is ambiguous:
template<class T, class U> int f(S<T>, S<U>) { return 1; }
template<class T> int f(T, T) { return 2; } // rel_ops
S<int> s; f(s,s); // ambiguous between #1 and #2
Adding the one-template-parameter version fixes the ambiguity:
template<class T, class U> int f(S<T>, S<U>) { return 1; }
template<class T> int f(T, T) { return 2; } // rel_ops
template<class T> int f(S<T>, S<T>) { return 3; }
S<int> s; f(s,s); // #3 beats both #1 and #2
We have the same problem with `reverse_iterator` as with `__wrap_iter`.
But so do libstdc++ and Microsoft, so we're not going to worry about it.
Differential Revision: https://reviews.llvm.org/D105894
This commit reverts 5099e01568 and 77396bbc98, which broke the build
in various ways. I'm reverting until I can investigate, since that
change appears to be way more subtle than it seemed.
On Windows, structs with a destructor are always returned indirectly;
add this to the list of known exceptions in the test where the class
isn't returned in registers as expected.
Differential Revision: https://reviews.llvm.org/D105906
This is a second attempt at D101497, which landed as
9a9bc76c0e but had to be reverted in
8cf7ddbdd4.
This issue was that in the case that `COMPILER_RT_INSTALL_PATH` is
empty, expressions like "${COMPILER_RT_INSTALL_PATH}/bin" evaluated to
"/bin" not "bin" as intended and as was originally.
One solution is to make `COMPILER_RT_INSTALL_PATH` always non-empty,
defaulting it to `CMAKE_INSTALL_PREFIX`. D99636 adopted that approach.
But, I think it is more ergonomic to allow those project-specific paths
to be relative the global ones. Also, making install paths absolute by
default inhibits the proper behavior of functions like
`GNUInstallDirs_get_absolute_install_dir` which make relative install
paths absolute in a more complicated way.
Given all this, I will define a function like the one asked for in
https://gitlab.kitware.com/cmake/cmake/-/issues/19568 (and needed for a
similar use-case).
---
Original message:
Instead of using `COMPILER_RT_INSTALL_PATH` through the CMake for
complier-rt, just use it to define variables for the subdirs which
themselves are used.
This preserves compatibility, but later on we might consider getting rid
of `COMPILER_RT_INSTALL_PATH` and just changing the defaults for the
subdir variables directly.
---
There was a seaming bug where the (non-Apple) per-target libdir was
`${target}` not `lib/${target}`. I suspect that has to do with the docs
on `COMPILER_RT_INSTALL_PATH` saying was the library dir when that's no
longer true, so I just went ahead and fixed it, allowing me to define
fewer and more sensible variables.
That last part should be the only behavior changes; everything else
should be a pure refactoring.
---
I added some documentation of these variables too. In particular, I
wanted to highlight the gotcha where `-DSomeCachePath=...` without the
`:PATH` will lead CMake to make the path absolute. See [1] for
discussion of the problem, and [2] for the brief official documentation
they added as a result.
[1]: https://cmake.org/pipermail/cmake/2015-March/060204.html
[2]: https://cmake.org/cmake/help/latest/manual/cmake.1.html#options
In 38b2dec37e the problem was somewhat
misidentified and so `:STRING` was used, but `:PATH` is better as it
sets the correct type from the get-go.
---
D99484 is the main thrust of the `GnuInstallDirs` work. Once this lands,
it should be feasible to follow both of these up with a simple patch for
compiler-rt analogous to the one for libcxx.
Reviewed By: phosek, #libc_abi, #libunwind
Differential Revision: https://reviews.llvm.org/D105765
std::exchange is only constexpr in C++20 and later. We were using it
in a constructor marked unconditionally constexpr, which caused issues
when building with -std=c++17.
The weird part is that the issue only showed up when building on the
arm64 macs, but that must be caused by the specific version of Clang
used on those. Since the code is clearly wrong and the fix is obvious,
I'm not going to investigate this further.
Louis Dionne