Otherwise, files don't link when using a GNU linker, which is more
sensitive on the order of the source file relative to the various
linked libraries. See http://c-faq.com/lib/libsearch.html for an
explanation of the problem.
Both test formats are equivalent, so this *should* not be a problem.
We've fixed a couple of failures uncovered by the first time we tried
making the switch, so this new attempt should go even farther.
If failures are noticed, it should be fine to revert this commit, but
please give a heads up afterwards so we know to address the issues!
Also note that it is still possible to use the old format by passing
`--param=use_old_format=True` when running Lit for the time being.
Summary:
This is a patch that Android has been carrying in its tree for several
years. This patch upstreams the existing ABI.
There's some historical cruft here. __regex_word used to be a part of
regex_traits rather than ctype_base. Bionic also used to use its own
ctype implementation because the libc++ builtin one wasn't available
yet. Bionic's ctype masks were 8 bits wide and already saturated, so a
wider type needed to be used for the regex mask, and the existing
value was already used so Android needed to specify its own.
Since then Android has migrated to the builtin ctype implementation
and this patch probably should have been dropped then. Unfortunately
that was not noticed at the time, so now we need to keep this to
maintain the current ABI.
Reviewers: EricWF, #libc, ldionne
Reviewed By: #libc, ldionne
Subscribers: dexonsmith, ldionne, libcxx-commits
Tags: #libc
Differential Revision: https://reviews.llvm.org/D76171
The LitConfig is shared across the whole test suite. However, since
enabling recursive expansion can be a breaking change for some test
suites, it's important to confine the setting to test suites that
enable it explicitly.
Note that other issues were raised with the way recursiveExpansionLimit
operates. However, this commit simply moves the setting to the right
place -- the mechanism by which it works can be improved independently.
Differential Revision: https://reviews.llvm.org/D77415
We had a workaround because GCC 5 does not evaluate static assertions
that are dependent on template parameters. This commit removes the
workaround and marks the corresponding tests as unsupported with GCC 5.
This has the benefit of bringing the new and the old test formats closer
without having to carry a workaround for an old compiler in the new
test format.
Otherwise, we're missing some flags like the flags that are used by
sanitizer builds and the 32-bit builds. In the long term, I think it
would be better to have only %{compile_flags} and %{link_flags}, but
for the benefit of adopting the new format by default, I think it's OK
to add %{flags} to it.
On Windows, we must make sure to close the temporary tar file before we
try to scp it.
This is an alternative approach to https://reviews.llvm.org/D77500.
This reverts commit 1580c76c4a.
This causes libcxx/selftest/newformat/sh.cpp/substitutions.sh.cpp to
fail with a linker error on Fedora 31 (x86-64) release (no assert)
builds.
Both test formats are equivalent, so this *should* not be a problem.
However, I'm taking advantage of the week-end to test this and see if
there are any failures. If so, it should be fine to revert this until
the failures have been addressed.
For the time being, it is still possible to use the old format by passing
`--param=use_old_format=True` when running Lit.
Summary:
Allow users to simultaneously enable address and undefined behavior
sanitizers, in the same manner that LLVM's 'HandleLLVMOptions.cmake'
allows.
Prior to this patch, `cmake -DLLVM_USE_SANITIZER="Address;Undefined"`
would succeed and the build would build most of the LLVM project with
`-fsanitize=address,undefined`, but a warning would be printed by
libcxx's CMake, and the build would use neither sanitizer. This
patch results in no warning being printed, and both sanitizers are used
in building libcxx.
Reviewers: jroelofs, EricWF, ldionne, #libc!
Subscribers: mgorny, dexonsmith, llvm-commits, libcxx-commits
Tags: #libc
Differential Revision: https://reviews.llvm.org/D77466
Always depend on the compiler to have a correct implementation of
max_align_t in stddef.h and don't provide a fallback. For pre-C++11,
require __STDCPP_NEW_ALIGNMENT__ in <new> as provided by clang in all
standard modes. Adjust test cases to avoid testing or using max_align_t
in pre-C++11 mode and also to better deal with alignof(max_align_t)>16.
Document requirements of the alignment tests around natural alignment of
power-of-two-sized types.
Differential revision: https://reviews.llvm.org/D73245
The testing script used to test libc++ historically did not like directories
without any testing files, so these tests had been added. Since this is
not necessary anymore, we can now remove these files. This has the benefit
that the total number of tests reflects the real number of tests more
closely, and we also skip some unnecessary work (especially relevant when
running tests over SSH).
However, some nothing_to_do.pass.cpp tests actually serve the purpose of
documenting that an area of the Standard doesn't need to be tested, or is
tested elsewhere. These files are not removed by this commit.
Removal done with:
import os
import itertools
for (dirpath, dirnames, filenames) in itertools.chain(os.walk('./libcxx/test'),
os.walk('./libcxxabi/test')):
if len(filenames + dirnames) > 1 and \
any(p == 'nothing_to_do.pass.cpp' for p in filenames):
os.remove(os.path.join(dirpath, 'nothing_to_do.pass.cpp'))
This new test format is simpler and more flexible. It creates Lit ShTests
on the fly that reuse existing substitutions (like %{cxx}) instead of
having complex logic in Python to run the tests. This has the benefit
that virtually no coding is required to customize how the test suite is
run -- one can achieve pretty much anything by defining the appropriate
substitutions in a simple lit.cfg file.
For example, in order to run the tests on an embedded device after
building with a specific SDK, one can set the %{cxx} and %{compile_flags}
substitutions to use that SDK, and the %{exec} substitution to the ssh.py
script currently used for .sh.cpp tests with a remote executor. Dealing with
the SSHExecutor becomes unnecessary, since all tests are treated like ShTests.
As a side effect of this design, configuration files for the test
suite can be as simple as:
config.substitutions.append(('%{cxx}', '<path-to-compiler>'))
config.substitutions.append(('%{compile_flags}', '<flags>'))
config.substitutions.append(('%{link_flags}', '<flags>'))
config.substitutions.append(('%{exec}', '<script-to-execute>'))
This should allow storing lit.cfg files for various configurations
directly in the repository instead of relying on complicated logic
in config.py to set up the right flags. I've found numerous problems
in that logic in the past years, and it seems like having simple and
explicit configuration files for the configurations we support is
going to solve most of these problems. Specifically, I am hoping to
store configuration files for testing other Standard Libraries in
the repository.
Improving the interaction with the test suite configuration is still a
work in progress, so for now this test format reuses the substitutions and
available features that are set up by the current config.py.
This new test format should support pretty much everything that the current
test format supports, however it will not be enabled by default at first to
make sure we're satisfied with it. For a short period of time, the new format
will require `--param=use_new_format=True` to be enabled, however it is a very
short term goal to replace the current testing format entirely and to simplify
the configuration accordingly.
Differential Revision: https://reviews.llvm.org/D77338
This test regressed with 5ade17e0ca, but we never noticed it because
.pass.mm tests were skipped due to a bug in our Lit config. This commit
fixes is_pointer (by essentially reverting tha part of 5ade17e0ca) and
also adds .pass.mm tests to the list of supported test suffixes.
We can explore how to support __is_pointer with Objective-C++ qualifiers
as a follow-up -- the main goal of this commit is to fix the regression
quickly and make sure all tests of the suite are run.
Instead of hardcoding absolute paths on the build-host in the executables,
use relative paths from the current working directory. Also, use
FILE_DEPENDENCIES to mark the static test env as being required by
the relevant tests.
Given a SSH executor that copies the files to the remote host properly,
the tests can be run on that remote host.
Chromium's build sets LIBCXX_CXX_ABI_SYSTEM explicitly when building
libc++, which was broken by 61e89737c5 (which stopped listening to
that option). As a workaround, this commit uses the system libc++abi
when LIBCXX_CXX_ABI_SYSTEM is used.
However, we will need to work with Chromium to standardize their build
of libc++, because LIBCXX_CXX_ABI_SYSTEM is not a public facing build
configuration for libc++, and has never been AFAICT.
The 'runtimes' build started failing because libc++ stopped using the
in-tree libc++abi when HAVE_CXXABI is set after 61e89737c. This commit
tries to bring back the old behavior when HAVE_CXXABI is set in order
to fix CIs.
However, we really need to sit down and discuss what ways of building
libc++ are supported and formalize them, because having the libc++ build
system branch on basically random variables in some CMake cache somewhere
is not a viable path forward.
... in `msvc_stdlib_force_include.h` to also ignore new MSVC warning C5215 "'%s' a function parameter with volatile qualified type is deprecated in C++20". Since we're touching it, also update from non-standard `__pragma(meow)` to standard `_Pragma("meow")`.
This patch reimplements the dynamic filesystem helper using Posix
functionality instead of relying on Python. The primary reason for
doing this is that it allows running the libc++ test suite on devices
that do not have Python.
Differential Revision: https://reviews.llvm.org/D77140
In standalone builds the cxxabi_shared and cxxabi_static targets don't exist.
We need to link against the library itself.
Differential Revision: https://reviews.llvm.org/D77294
This commit removes support for building against the system libc++abi,
which was supported on Apple platforms. This is basically never what we
want to do, since libc++ and libc++abi are coupled and building a trunk
libc++ against an older libc++abi can lead to incompatibilities (and
good luck debugging them!). It might have made some sense to support
that when the monorepo did not exist, however I don't think this is
anything but a footgun nowadays.
Furthermore, based on the newly-made assumption that we're building
against the monorepo libc++abi, we can simplify the search path logic
for finding libc++abi.
This area of our build system has a lot of technical debt accumulated,
and it's surprisingly difficult to change. We've tried different things
and failed several times in the past. I did test this change on our
Docker image for the build bots and on Apple platforms, however it is
possible that this breaks some unknown configuration, in which case it
should be fine to revert this (so we can try again!).
Instead of executing tests from within the libc++ test suite, we execute
them from the Lit execution directory. However, since some tests have
file dependencies, we must copy those dependencies to the execution
directory where they are executed.
This has the major benefit that if a test modifies a file (whether it
is wanted or not), other tests will not see those modifications. This
is good because current tests assume that input data is never modified,
however this could be an incorrect assumption if some test does not
behave properly.
Based on an issue brought up in https://reviews.llvm.org/D67900, this commit reverts the changes to is_floating_point and is_arithmetic made in D67900.
After D67900 landed, __float128 behaved differently in those two type traits, causing compiler errors in numeric limits (and possibly others).
The benefit of doing this is that we can now handle directories that
contain symlinks and other arbitrary things, such as the static_test_env
required by filesystem tests.
As a fly-by fix, we also accumulate several commands to perform over SSH
and execute them at once instead of SSHing several times. This should be
faster on average.
If a ShTest has for example another command in front of the test
executable it wants to execute, ssh.py needs to properly translate
the path of that test executable to the executable on the remote host.
For example, running '%{exec} ! %t.exe', we can't assume that the
test-executable is the first argument after '%{exec}'.
Using the ADDITIONAL_COMPILE_FLAGS annotation, it is possible to move
these tests from .sh.cpp to .pass.cpp, making them suitable for running
on remote hosts more easily.
This makes it closer to how one would run the tests by hand, and it is
also closer to how the SSHExecutor runs the tests remotely. It also
allows using shell builtins in .sh.cpp tests when using %{exec}.
Based on the current discussion in https://llvm.org/PR45307, it seems
that it's legitimate for `temp_directory_path()` to return a path with
a trailing slash. Since `p.parent_path()` will never contain a trailing
slash, comparing it to the result of `temp_directory_path()` will fail
depending on whether `temp_directory_path()` returns a trailing slash
or not.
Otherwise, trying to reproduce a failing filesystem test by copy-pasting
the command-line used and running that in the shell won't work, because
the shell will eat quoting around the define and we'll end up with a
non-stringized path in the .cpp file.
That way, local lit configuration files don't have to worry about
deep-copying the compiler instance of the test format, which is
arguably an implementation detail.
We pass the config to this method even though it is not used by the
current test format because this allows replacing the current test
format by other test formats that would require the config to add
new compile flags.
This reduces the complexity of our already complex global lit configuration,
and also avoids cluttering the compilation commands for all tests with
things that are only relevant to the filesystem tests.
Differential Revision: https://reviews.llvm.org/D76785
Previously, filesystem tests would require LIBCXX_FILESYSTEM_DYNAMIC_TEST_ROOT
to be present in the environment and to match the value provided when
compiling, as a macro. This has the problem that it only allows for the
filesystem tests to be run on the same machine they are created.
Instead, we create a temporary directory for each test. Technically,
this is tricky to do because we're relying on some of the code that
we're testing to do this. However, there's no other portable way of
creating temporary direcories in C++, so this is difficult to avoid.
Differential Revision: https://reviews.llvm.org/D76731
This re-commits cd7f9751c3, which was reverted in 12f6b024f9 because
it broke the LLVM `check-all` target. This commit addresses the underlying
issue by not setting the lit_config.recursiveExpansionLimit parameter of
the libc++ test suite, which is otherwise picked up by other test suites
in LLVM.
Once we've settled on a fix for the underlying issue with
lit_config.recursiveExpansionLimit, we can start using it
again in libc++, but for now we can just work around it.
We will soon start removing technical debt and sharing code between the
two directories, so this first step is meant to discover potential places
where the libraries are built outside of a monorepo layout. I imagine
this could happen as a remnant of the pre-monorepo setup.
This was discussed on the libcxx-dev mailing list and we got overall
consensus on the direction. All consumers of libc++ and libc++abi
should already be doing so through the monorepo, however it is
possible that we catch some stragglers with this patch, in which
case it may need to be reverted temporarily.
Differential Revision: https://reviews.llvm.org/D76102
This allows adding compilation flags for a single test, which can help
eliminate some .sh.cpp tests and some custom handling in the libc++
test format.
It also works around the issue that .sh.cpp substitutions are _not_
equivalent to the actual compiler command lines used to compile tests,
since the compiler flags can be modified in local lit configurations,
and substitutions are frozen at that point. For example using %{compile}
in a .sh.cpp test in the coroutines subdirectory will not include the
-fcoroutines-ts flag, which is added in the local lit config, because
the %{compile} substitution is created long before we add -fcoroutines-ts
to the compiler flags (in the lit.local.cfg for coroutines).
We've been meaning to remove those targets for a while, and the fix is
simple enough cause they're all just aliases to other targets.
This is a re-application of f383fb40b1, wich was reverted in 04d48111b
because the build bots had not been updated yet. The build bot configurations
have now been updated not to use the deprecated targets, and I verified
that they were using the non-deprecated targets, so we should be good
unless I missed a bot.
Differential Revision: https://reviews.llvm.org/D76104
This reverts commit cd7f9751c3 which has
unintended breakage to non-libcxx projects when using the documented way
of building LLVM. (See the Getting Started guide. I.e. one big CMake setup.)
Since lit supports expanding substitutions recursively, we can define
substitutions in terms of other substitutions. This allows us to simplify
how libc++ substitutions are defined.
This doesn't change the substitutions at all, it only makes them simpler
to define.
lit is not very clever when it performs substitution on RUN lines. It
simply looks for a match anywhere in the line (without tokenization)
and replaces it by the expansion. This means that a RUN line containing
e.g. `-verify-ignore-unexpected=note` wouod be expanded to
`-verify-ignore-unexpected=<substitution for not>e`, which is
surprising and nonsensical.
It also means that something like `%compile_module` could be expanded
to `<substitution-for-%compile>_module` or to the correct substitution,
depending on the order in which substitutions are evaluated by lit.
To avoid such problems, it is a good habit to delimit custom substitutions
with some token. This commit does that for all substitutions used in the
libc++ and libc++abi test suites.
Summary:
The gdb pretty printer misprints variables declared via
using declarations of the form:
namespace foo {
using string_view = std::string_view;
string_view bar;
}
This change fixes that, by deferring the decision to ignore
types not inside std until after desugaring.
Reviewers: #libc!
Subscribers: broadwaylamb, libcxx-commits
Tags: #libc
Differential Revision: https://reviews.llvm.org/D76816
This reverts commit a32b94c6c3.
The buildbot startup scripts need to run as root. The buildbot
worker should have already been running as a different account.
More investigation needed.
Forcing -Werror and other warnings means that the test suite isn't
actually testing what most people are seeing in their code -- it seems
better and less arbitrary to compile these tests as close as possible
to the compiler default instead.
Removing -Werror also means that we get to differentiate between
diagnostics that are errors and those that are warnings, which makes
the test suite more precise.
Differential Revision: https://reviews.llvm.org/D76311
We have several test directories whose names end in .spec. Ensure we
don't accidentally ignore them by removing a section of the .gitignore
that's irrelevant for libc++.
Differential Revision: https://reviews.llvm.org/D69195
Introduced by https://reviews.llvm.org/D72687. This condition can happen
when the tests are not being run at all, and we're only trying to generate
the libc++ headers.
Some tests do not fail at all when -verify is not supported, unless some
arbitrary warning flag is added to make them fail. We currently used
-Werror=unused-result to make them fail, but doing so makes the test
suite a lot more inscrutable. It seems better to just disable those
tests when -verify is not supported.
Differential Revision: https://reviews.llvm.org/D76256
It's hard to imagine someone using a recent version of libc++ with a
roughly 3 years old Clang. Since we're not testing libc++ with Clang 3.5
anyway, claiming support for it is somewhat of a lie.
Note that we don't test Clang 4 either, however I have no reason to bump
the requirement beyond Clang 4 at the moment, whereas removing Clang 3.5
allows simplifying the test suite.
Differential Revision: https://reviews.llvm.org/D76618
This reverts commit f383fb40b. It looks like several of our build bots
are still using the legacy target names, so we'll change those before
we commit this change again.
We've been meaning to remove those targets for a while, and the fix is
simple enough cause they're all just aliases to other targets.
There's no doubt this commit will break some CI systems, however the
fix is trivial.
Differential Revision: https://reviews.llvm.org/D76104
This commit rewrites/removes the docker files used to create
the libc++ buildbots.
The major changes in this patch are:
1. Delete Dockerfiles used to build compilers. These have moved to
github.com/efcs/compiler-images
2. Minimize the llvm-buildbot docker image. Instead of running the
buildbots from a committed docker image, the builders now build the
image on startup. This means changes to the docker file automatically
propogate to the builders (within ~24 hours without restart).
3. Version the compilers used by the builders. This means the bots
won't start failing because the apt.llvm.org clang package updated.
The current lit test suite doesn't really allow us to express that the
test should be disabled when testing the trunk variant of libc++, even
if we're running it on a supported macOS. Because of that, the test
is enabled when _LIBCPP_DISABLE_AVAILABILITY is defined, and the test
XPASSes.
Before this patch, the %run substitution did not contain the same
environment variables as normal `pass.cpp` tests. It also didn't
have the right working directory and the script wasn't aware of
potential file dependencies.
With this change, the combination of %build and %run in a .sh.cpp script
should match how pass.cpp tests are actually executed much more closely.
In C++03 mode, nullptr is defined by libc++, not the compiler so, we can't use __is_fundamental (because it will return false for nullptr).
Fixes: 5ade17e0ca
This patch updates <type_traits> to use builtin type traits whenever
possible to improve compile times.
Differential Revision: https://reviews.llvm.org/D67900
The current implementation of binomial_distribution is not guaranteed to
converge for certain extreme configurations of the engine and distribution.
This is due to a mistake in the implementation of the algorithm from the
given reference paper. The algorithm in the paper is guaranteed to
terminate but has redundant statements. The current implementation
simplified away the redundancy into a while loop, but it excludes the
return condition of the case where a good sample cannot be returned for
the particular sample being used from the uniform distribution, which is
what causes the infinite loop. This change guarantees termination by
recognizing that a good sample cannot be returned and returning 0 after
breaking the loop. This is also in contrast to the paper because the
return value as specified in the paper violates basic checks in at least
a subset of the extreme cases where the current implementation fails to
terminate. This default return value of 0 is satisfactory for the
extreme case known so far.
Since this is only meant to affect extreme cases where the algorithm
does not terminate anyways, the behavior is expected to remain exactly
the same for all non-extreme cases that have been terminating so far.
Fixes https://llvm.org/PR44847
Differential Revision: https://reviews.llvm.org/D74997
Louis Dionne