libcxx/docs/VendorDocumentation.md

(vendordocumentation)=

# Vendor Documentation

```{contents}
:local: true
```

The instructions on this page are aimed at vendors who ship libc++ as part of an
operating system distribution, a toolchain or similar shipping vehicles. If you
are a user merely trying to use libc++ in your program, you most likely want to
refer to your vendor's documentation, or to the general user documentation
{ref}`here <user-documentation>`.

:::{warning}
If your operating system already provides libc++, it is important to be careful
not to replace it. Replacing your system's libc++ installation could render it
non-functional. Use the CMake option `CMAKE_INSTALL_PREFIX` to select a safe
place to install libc++.
:::

## The default build

The default way of building libc++, libc++abi and libunwind is to root the CMake
invocation at `<monorepo>/runtimes`. While those projects are under the LLVM
umbrella, they are different in nature from other build tools, so it makes sense
to treat them as a separate set of entities. The default build can be achieved
with the following CMake invocation:

```bash
$ git clone https://github.com/llvm/llvm-project.git
$ cd llvm-project
$ mkdir build
$ cmake -G Ninja -S runtimes -B build -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind" # Configure
$ ninja -C build cxx cxxabi unwind                                                        # Build
$ ninja -C build check-cxx check-cxxabi check-unwind                                      # Test
$ ninja -C build install-cxx install-cxxabi install-unwind                                # Install
```

:::{note}
See [Vendor Configuration Options](#vendor-configuration-options) below for more configuration options.
:::

After building the various `install-XXX` targets, shared libraries for libc++, libc++abi and
libunwind should now be present in `<CMAKE_INSTALL_PREFIX>/lib`, and headers in
`<CMAKE_INSTALL_PREFIX>/include/c++/v1`. See the instructions below for information on how
to use this libc++ over the default one.

In the default configuration, the runtimes will be built using the compiler available by default
on your system. Of course, you can change what compiler is being used with the usual CMake
variables. If you wish to build the runtimes from a just-built Clang, the bootstrapping build
explained below makes this task easy.

### Using the just-built libc++

Most compilers provide a way to disable the default behavior for finding the standard library and
to override it with custom paths. With Clang, this can be done with:

```bash
$ clang++ -nostdinc++ -isystem <install>/include/c++/v1 \
          -nostdlib++ -L <install>/lib -lc++            \
          -Wl,-rpath,<install>/lib                      \
          test.cpp
```

The option `-Wl,-rpath,<install>/lib` adds a runtime library search path, which causes the system's
dynamic linker to look for libc++ in `<install>/lib` whenever the program is loaded.

:::{note}
If the runtimes were built using the "per-target runtime directory" layout,
they will be in `<install>/lib/<target-triple>` instead of `<install>/lib`.
In this case, use the former path for all library paths in the command above
(the path to include files does not change).
:::

## The Bootstrapping build

It is possible to build Clang and then build the runtimes using that just-built compiler in a
single CMake invocation. This is usually the correct way to build the runtimes when putting together
a toolchain, or when the system compiler is not adequate to build them (too old, unsupported, etc.).
To do this, use the following CMake invocation, and in particular notice how we're now rooting the
CMake invocation at `<monorepo>/llvm`:

```bash
$ mkdir build
$ # Configure
$ cmake -G Ninja -S llvm -B build                                       \
        -DCMAKE_BUILD_TYPE=RelWithDebInfo                               \
        -DLLVM_ENABLE_PROJECTS="clang"                                  \
        -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind;compiler-rt" \
        -DLLVM_RUNTIME_TARGETS="<target-triple>"
$ ninja -C build runtimes          # Build
$ ninja -C build check-runtimes    # Test
$ ninja -C build install-runtimes  # Install
```

:::{note}
- This type of build is also commonly called a "Runtimes build", but we would like to move
  away from that terminology, which is too confusing.
- Adding the `--fresh` flag to the top-level cmake invocation in a bootstrapping build *will not*
  freshen the cmake cache of any of the enabled runtimes.
:::

(vendor-configuration-options)=

## Vendor Configuration Options

This section documents configuration options that can be used by vendors when building the library.
These options provide a great deal of flexibility to customize libc++, such as selecting the ABI in
use, whether some features are provided, etc.

:::{warning}
Many of these CMake options are tied to configuration macros with a corresponding name in the source
code. However, these configuration macros are not intended to be customized by users directly, since
many of them require the library to be built with a matching configuration. If you don't build libc++
yourself, you should not use the options documented here.
:::

### General purpose options

:::{option} LIBCXX_INSTALL_LIBRARY:BOOL
**Default**: `ON`

Toggle the installation of the library portion of libc++.
:::

:::{option} LIBCXX_INSTALL_HEADERS:BOOL
**Default**: `ON`

Toggle the installation of the libc++ headers.
:::

:::{option} LIBCXX_INSTALL_MODULES:BOOL
**Default**: `ON`

Toggle the installation of the experimental libc++ module sources.
:::

:::{option} LIBCXX_ENABLE_SHARED:BOOL
**Default**: `ON`

Build libc++ as a shared library. Either `LIBCXX_ENABLE_SHARED` or
`LIBCXX_ENABLE_STATIC` has to be enabled.
:::

:::{option} LIBCXX_ENABLE_STATIC:BOOL
**Default**: `ON`

Build libc++ as a static library. Either `LIBCXX_ENABLE_SHARED` or
`LIBCXX_ENABLE_STATIC` has to be enabled.
:::

:::{option} LIBCXX_LIBDIR_SUFFIX:STRING
Extra suffix to append to the directory where libraries are to be installed.
This option overrides `LLVM_LIBDIR_SUFFIX`.
:::

:::{option} LIBCXX_HERMETIC_STATIC_LIBRARY:BOOL
**Default**: `OFF`

Do not export any symbols from the static libc++ library.
This is useful when the static libc++ library is being linked into shared
libraries that may be used in with other shared libraries that use different
C++ library. We want to avoid exporting any libc++ symbols in that case.
:::

:::{option} LIBCXX_ENABLE_FILESYSTEM:BOOL
**Default**: `ON`

This option can be used to enable or disable the filesystem components on
platforms that may not support them.
:::

:::{option} LIBCXX_ENABLE_WIDE_CHARACTERS:BOOL
**Default**: `ON`

This option can be used to disable support for `wchar_t` in the library. It also
allows the library to work on top of a C Standard Library that does not provide
support for `wchar_t`. This is especially useful in embedded settings where
C Standard Libraries don't always provide all the usual bells and whistles.
:::

:::{option} LIBCXX_ENABLE_TIME_ZONE_DATABASE:BOOL
**Default**: `ON`

Whether to include support for time zones in the library. Disabling
time zone support can be useful when porting to platforms that don't
ship the IANA time zone database. When time zones are not supported,
time zone support in `<chrono>` will be disabled.
:::

:::{option} LIBCXX_INSTALL_LIBRARY_DIR:PATH
**Default**: `lib${LIBCXX_LIBDIR_SUFFIX}`

Path where built libc++ libraries should be installed. If a relative path,
relative to `CMAKE_INSTALL_PREFIX`.
:::

:::{option} LIBCXX_INSTALL_INCLUDE_DIR:PATH
**Default**: `include/c++/v1`

Path where target-agnostic libc++ headers should be installed. If a relative
path, relative to `CMAKE_INSTALL_PREFIX`.
:::

:::{option} LIBCXX_INSTALL_INCLUDE_TARGET_DIR:PATH
**Default**: `include/c++/v1` or
`include/${LLVM_DEFAULT_TARGET_TRIPLE}/c++/v1`

Path where target-specific libc++ headers should be installed. If a relative
path, relative to `CMAKE_INSTALL_PREFIX`.
:::

:::{option} LIBCXX_SHARED_OUTPUT_NAME:STRING
**Default**: `c++`

Output name for the shared libc++ runtime library.
:::

:::{option} {LIBCXX,LIBCXXABI,LIBUNWIND}_ADDITIONAL_COMPILE_FLAGS:STRING
**Default**: `""`

Additional compile flags to use when building the runtimes. This should be a CMake `;`-delimited list of individual
compiler options to use. For options that must be passed as-is to the compiler without deduplication (e.g.
`-Xclang -foo` option groups), consider using `SHELL:` as [documented here](https://cmake.org/cmake/help/latest/command/add_compile_options.html#option-de-duplication).
:::

:::{option} LIBCXX_ADDITIONAL_LIBRARIES:STRING
**Default**: `""`

Additional libraries libc++ is linked to which can be provided in cache.
:::

:::{option} LIBCXX_ENABLE_EXCEPTIONS:BOOL
**Default**: `ON`

Build libc++ with exception support.
:::

:::{option} LIBCXX_ENABLE_RTTI:BOOL
**Default**: `ON`

Build libc++ with run time type information.
This option may only be set to OFF when LIBCXX_ENABLE_EXCEPTIONS=OFF.
:::

:::{option} LIBCXX_INCLUDE_TESTS:BOOL
**Default**: `ON` (or value of `LLVM_INCLUDE_TESTS`)

Build the libc++ test suite, which includes various types of tests like conformance
tests, vendor-specific tests and benchmarks.
:::

:::{option} LIBCXX_INCLUDE_BENCHMARKS:BOOL
**Default**: `ON`

Build the libc++ benchmark tests and the Google Benchmark library needed
to support them.
:::

:::{option} LIBCXX_ASSERTION_HANDLER_FILE:PATH
**Default**:: `"${CMAKE_CURRENT_SOURCE_DIR}/vendor/llvm/default_assertion_handler.in"`

Specify the path to a header that contains a custom implementation of the
assertion handler that gets invoked when a hardening assertion fails. If
provided, this header will be included by the library, replacing the
default assertion handler. If this is specified as a relative path, it
is assumed to be relative to `<monorepo>/libcxx`.
:::

### ABI Specific Options

The following options allow building libc++ for a different ABI version.

:::{option} LIBCXX_ABI_VERSION:STRING
**Default**: `1`

Defines the target ABI version of libc++.
:::

:::{option} LIBCXX_ABI_UNSTABLE:BOOL
**Default**: `OFF`

Build the "unstable" ABI version of libc++. Includes all ABI changing features
on top of the current stable version.
:::

::::{option} LIBCXX_ABI_NAMESPACE:STRING
**Default**: `__n` where `n` is the current ABI version.

This option defines the name of the inline ABI versioning namespace. It can be used for building
custom versions of libc++ with unique symbol names in order to prevent conflicts or ODR issues
with other libc++ versions.

:::{warning}
When providing a custom namespace, it's the vendor's responsibility to ensure the name won't cause
conflicts with other names defined by libc++, both now and in the future. In particular, inline
namespaces of the form `__[0-9]+` could cause conflicts with future versions of the library,
and so should be avoided.
:::
::::

:::{option} LIBCXX_ABI_DEFINES:STRING
**Default**: `""`

A semicolon-separated list of ABI macros to persist in the site config header.
See `include/__config` for the list of ABI macros.
:::

:::{option} LIBCXX_CXX_ABI:STRING
**Values**: `none`, `libcxxabi`, `system-libcxxabi`, `libcxxrt`, `libstdc++`, `libsupc++`, `vcruntime`.

Select the ABI library to build libc++ against.
:::

:::{option} LIBCXX_CXX_ABI_INCLUDE_PATHS:PATHS
Provide additional search paths for the ABI library headers.
:::

:::{option} LIBCXX_CXX_ABI_LIBRARY_PATH:PATH
Provide the path to the ABI library that libc++ should link against. This is only
useful when linking against an out-of-tree ABI library.
:::

:::{option} LIBCXX_ENABLE_STATIC_ABI_LIBRARY:BOOL
**Default**: `OFF`

If this option is enabled, libc++ will try and link the selected ABI library
statically.
:::

:::{option} LIBCXX_ENABLE_ABI_LINKER_SCRIPT:BOOL
**Default**: `ON` by default on UNIX platforms other than Apple unless
'LIBCXX_ENABLE_STATIC_ABI_LIBRARY' is ON. Otherwise the default value is `OFF`.

This option generate and installs a linker script as `libc++.so` which
links the correct ABI library.
:::

:::{option} LIBCXX_AVAILABILITY_MINIMUM_HEADER_VERSION:STRING
**Default**: `2`

This option configures the oldest version of the libc++ headers that the built
library has to be compatible with. See the
{ref}`minimum header version documentation<MinimumHeaderVersion>` for details.
:::

:::{option} LIBCXXABI_AVAILABILITY_MINIMUM_HEADER_VERSION:STRING
**Default**: `2`

This is the same as `LIBCXX_AVAILABILITY_MINIMUM_HEADER_VERSION` documented
above, but for libc++abi. The two options should be set to the same value if
libc++abi is used.
:::

:::{option} LIBCXXABI_USE_LLVM_UNWINDER:BOOL
**Default**: `ON`

Build and use the LLVM unwinder. Note: This option can only be used when
libc++abi is the C++ ABI library used.
:::

:::{option} LIBCXXABI_ADDITIONAL_LIBRARIES:STRING
**Default**: `""`

Additional libraries libc++abi is linked to which can be provided in cache.
:::

### LLVM-specific options

:::{option} LLVM_LIBDIR_SUFFIX:STRING
Extra suffix to append to the directory where libraries are to be
installed. On a 64-bit architecture, one could use `-DLLVM_LIBDIR_SUFFIX=64`
to install libraries to `/usr/lib64`.
:::

:::{option} LLVM_BUILD_32_BITS:BOOL
Build 32-bits executables and libraries on 64-bits systems. This option is
available only on some 64-bits Unix systems. Defaults to OFF.
:::

:::{option} LLVM_LIT_ARGS:STRING
Arguments given to lit. `make check` and `make clang-test` are affected.
By default, `'-sv --no-progress-bar'` on Visual C++ and Xcode, `'-sv'` on
others.
:::

## Support for Windows

Libc++ supports being built with clang-cl, but not with MSVC's cl.exe, as
cl doesn't support the `#include_next` extension. Furthermore, VS 2017 or
newer (19.14) is required.

Libc++ also supports being built with clang targeting MinGW environments.

Libc++ supports Windows 7 or newer. However, the minimum runtime version
of the build is determined by the `_WIN32_WINNT` define, which in many
SDKs defaults to the latest version. To build a version that runs on an
older version, define e.g. `_WIN32_WINNT=0x601` while building libc++,
to target Windows 7.

### CMake + Visual Studio

Building with Visual Studio currently does not permit running tests. However,
it is the simplest way to build.

```batch
> cmake -G "Visual Studio 16 2019" -S runtimes -B build ^
        -T "ClangCL"                                    ^
        -DLLVM_ENABLE_RUNTIMES=libcxx                   ^
        -DLIBCXX_ENABLE_SHARED=YES                      ^
        -DLIBCXX_ENABLE_STATIC=NO
> cmake --build build
```

### CMake + ninja (MSVC)

Building with ninja is required for development to enable tests.
A couple of tests require Bash to be available, and a couple dozens
of tests require other posix tools (cp, grep and similar - LLVM's tests
require the same). Without those tools the vast majority of tests
can still be ran successfully.

If Git for Windows is available, that can be used to provide the bash
shell by adding the right bin directory to the path, e.g.
`set PATH=%PATH%;C:\Program Files\Git\usr\bin`.

Alternatively, one can also choose to run the whole build in a MSYS2
shell. That can be set up e.g. by starting a Visual Studio Tools Command
Prompt (for getting the environment variables pointing to the headers and
import libraries), and making sure that clang-cl is available in the
path. From there, launch an MSYS2 shell via e.g.
`C:\msys64\msys2_shell.cmd -full-path -mingw64` (preserving the earlier
environment, allowing the MSVC headers/libraries and clang-cl to be found).

In either case, then run:

```batch
> cmake -G Ninja -S runtimes -B build                                               ^
        -DCMAKE_C_COMPILER=clang-cl                                                 ^
        -DCMAKE_CXX_COMPILER=clang-cl                                               ^
        -DLLVM_ENABLE_RUNTIMES=libcxx
> ninja -C build cxx
> ninja -C build check-cxx
```

If you are running in an MSYS2 shell and you have installed the
MSYS2-provided clang package (which defaults to a non-MSVC target), you
should add e.g. `-DCMAKE_CXX_COMPILER_TARGET=x86_64-windows-msvc` (replacing
`x86_64` with the architecture you're targeting) to the `cmake` command
line above. This will instruct `check-cxx` to use the right target triple
when invoking `clang++`.

### CMake + ninja (MinGW)

libcxx can also be built in MinGW environments, e.g. with the MinGW
compilers in MSYS2. This requires clang to be available (installed with
e.g. the `mingw-w64-x86_64-clang` package), together with CMake and ninja.

```bash
> cmake -G Ninja -S runtimes -B build                                               \
        -DCMAKE_C_COMPILER=clang                                                    \
        -DCMAKE_CXX_COMPILER=clang++                                                \
        -DLLVM_ENABLE_LLD=ON                                                        \
        -DLLVM_ENABLE_RUNTIMES="libcxx;libcxxabi;libunwind"                         \
        -DLIBCXXABI_ENABLE_SHARED=OFF                                               \
        -DLIBCXX_ENABLE_STATIC_ABI_LIBRARY=ON
> ninja -C build cxx
> ninja -C build check-cxx
```

(assertion-handler)=

## Overriding the default assertion handler

When the library wants to terminate due to a hardening assertion failure, the
program is aborted by invoking a trap instruction (or in debug mode, by
a special verbose termination function that prints an error message and calls
`std::abort()`). This is done to minimize the code size impact of enabling
hardening in the library. However, vendors can also override that mechanism at
CMake configuration time.

Under the hood, a hardening assertion will invoke the
`_LIBCPP_ASSERTION_HANDLER` macro upon failure. A vendor may provide a header
that contains a custom definition of this macro and specify the path to the
header via the `LIBCXX_ASSERTION_HANDLER_FILE` CMake variable. If provided,
this header will be included by the library and replace the default
implementation. The header must not include any standard library headers
(directly or transitively) because doing so will almost always create a circular
dependency. The `_LIBCPP_ASSERTION_HANDLER(message)` macro takes a single
parameter that contains an error message explaining the hardening failure and
some details about the source location that triggered it.

When a hardening assertion fails, it means that the program is about to invoke
library undefined behavior. For this reason, the custom assertion handler is
generally expected to terminate the program. If a custom assertion handler
decides to avoid doing so (e.g. it chooses to log and continue instead), it does
so at its own risk -- this approach should only be used in non-production builds
and with an understanding of potential consequences. Furthermore, the custom
assertion handler should not throw any exceptions as it may be invoked from
standard library functions that are marked `noexcept` (so throwing will result
in `std::terminate` being called).

## Using Alternate ABI libraries

In order to implement various features like exceptions, RTTI, `dynamic_cast` and
more, libc++ requires what we refer to as an ABI library. Typically, that library
implements the [Itanium C++ ABI](https://itanium-cxx-abi.github.io/cxx-abi/abi.html).

By default, libc++ uses libc++abi as an ABI library. However, it is possible to use
other ABI libraries too.

### Using libsupc++ on Linux

You will need libstdc++ in order to provide libsupc++.

Figure out where the libsupc++ headers are on your system. On Ubuntu this
is `/usr/include/c++/<version>` and `/usr/include/c++/<version>/<target-triple>`

You can also figure this out by running

```bash
$ echo | g++ -Wp,-v -x c++ - -fsyntax-only
ignoring nonexistent directory "/usr/local/include/x86_64-linux-gnu"
ignoring nonexistent directory "/usr/lib/gcc/x86_64-linux-gnu/4.7/../../../../x86_64-linux-gnu/include"
#include "..." search starts here:
#include &lt;...&gt; search starts here:
/usr/include/c++/4.7
/usr/include/c++/4.7/x86_64-linux-gnu
/usr/include/c++/4.7/backward
/usr/lib/gcc/x86_64-linux-gnu/4.7/include
/usr/local/include
/usr/lib/gcc/x86_64-linux-gnu/4.7/include-fixed
/usr/include/x86_64-linux-gnu
/usr/include
End of search list.
```

Note that the first two entries happen to be what we are looking for. This
may not be correct on all platforms.

We can now run CMake:

```bash
$ cmake -G Ninja -S runtimes -B build       \
  -DLLVM_ENABLE_RUNTIMES="libcxx"           \
  -DLIBCXX_CXX_ABI=libstdc++                \
  -DLIBCXXABI_USE_LLVM_UNWINDER=OFF         \
  -DLIBCXX_CXX_ABI_INCLUDE_PATHS="/usr/include/c++/4.7/;/usr/include/c++/4.7/x86_64-linux-gnu/"
$ ninja -C build install-cxx
```

You can also substitute `-DLIBCXX_CXX_ABI=libsupc++`
above, which will cause the library to be linked to libsupc++ instead
of libstdc++, but this is only recommended if you know that you will
never need to link against libstdc++ in the same executable as libc++.
GCC ships libsupc++ separately but only as a static library. If a
program also needs to link against libstdc++, it will provide its
own copy of libsupc++ and this can lead to subtle problems.

### Using libcxxrt on Linux

You will need to keep the source tree of [libcxxrt][libcxxrt] available
on your build machine and your copy of the libcxxrt shared library must
be placed where your linker will find it.

We can now run CMake like:

```bash
$ cmake -G Ninja -S runtimes -B build                               \
        -DLLVM_ENABLE_RUNTIMES="libcxx"                             \
        -DLIBCXX_CXX_ABI=libcxxrt                                   \
        -DLIBCXX_ENABLE_NEW_DELETE_DEFINITIONS=ON                   \
        -DLIBCXXABI_USE_LLVM_UNWINDER=OFF                           \
        -DLIBCXX_CXX_ABI_INCLUDE_PATHS=path/to/libcxxrt-sources/src
$ ninja -C build install-cxx
```

Unfortunately you can't simply run clang with "-stdlib=libc++" at this point, as
clang is set up to link for libc++ linked to libsupc++. To get around this
you'll have to set up your linker yourself (or patch clang). For example,

```bash
$ clang++ -stdlib=libc++ helloworld.cpp \
          -nodefaultlibs -lc++ -lcxxrt -lm -lc -lgcc_s -lgcc
```

Alternately, you could just add libcxxrt to your libraries list, which in most
situations will give the same result:

```bash
$ clang++ -stdlib=libc++ helloworld.cpp -lcxxrt
```

## libc++'s ABI guarantees

Libc++ provides several ABI guarantees, which are documented {ref}`here <ABIGuarantees>`.

## Availability Markup

Libc++ is shipped by various vendors. In particular, it is used as a system library on macOS, iOS and other Apple
platforms. In order for users to be able to compile a binary that is intended to be deployed to an older version of a
platform, Clang provides [availability attributes](https://clang.llvm.org/docs/AttributeReference.html#availability).
These attributes can be placed on declarations and are used to describe the life cycle of a symbol in the library.

The main goal is to ensure a compile-time error if a symbol that hasn't been introduced in a previously released library
is used in a program that targets that previously released library. Normally, this would be a load-time error when one
tries to launch the program against the older library.

For example, the filesystem library was introduced in the dylib in LLVM 9. On Apple platforms, this corresponds to
macOS 10.15. If a user compiles on a macOS 10.15 host but targets macOS 10.13 with their program, the compiler would
normally not complain (because the required declarations are in the headers), but the dynamic loader would fail to find
the symbols when actually trying to launch the program on macOS 10.13. To turn this into a compile-time issue instead,
declarations are annotated with when they were introduced, and the compiler can produce a diagnostic if the program
references something that isn't available on the deployment target.

This mechanism is general in nature, and any vendor can add their markup to the library (see below). Whenever a new
feature is added that requires support in the shared library, two macros are added below to allow marking the feature as
unavailable:

1. A macro named `_LIBCPP_AVAILABILITY_HAS_<feature>` which must be defined to `_LIBCPP_INTRODUCED_IN_<version>` for
   the appropriate LLVM version.
2. A macro named `_LIBCPP_AVAILABILITY_<feature>`, which must be defined to `_LIBCPP_INTRODUCED_IN_<version>_MARKUP`
   for the appropriate LLVM version.

When vendors decide to ship the feature as part of their shared library, they can update the
`_LIBCPP_INTRODUCED_IN_<version>` macro (and the markup counterpart) based on the platform version they shipped that
version of LLVM in. The library will then use this markup to provide an optimal user experience on these platforms.

Furthermore, many features in the standard library have corresponding feature-test macros. The
`_LIBCPP_AVAILABILITY_HAS_<feature>` macros are checked by the corresponding feature-test macros generated by
`generate_feature_test_macro_components.py` to ensure that the library doesn't announce a feature as being implemented
if it is unavailable on the deployment target.

Note that this mechanism is disabled by default in the "upstream" libc++. Availability annotations are only meaningful
when shipping libc++ inside a platform (i.e. as a system library), and so vendors that want them should turn those
annotations on at CMake configuration time.

(minimumheaderversion)=

## Minimum Header Version

In libc++ we add new functions and remove the use of other functions in the built library on a regular basis. To avoid
breaking programs, we keep old functions in the built library as documented in our
{ref}`header support policy<HeaderSupportPolicy>`. However, there are platforms where some of these functions could
never be referenced, because that platform never provided headers which referenced these functions. To reduce the size
of the built library on these platforms, libc++ provides the notion of a minimum header version. The minimum header
version describes the earliest version of the libc++ headers that can be used in a program linking against the library
currently being built. Functions which have never been referenced in headers since the minimum header version are
removed from the library. The minimum header version can be set with the CMake variables
`LIBCXX_AVAILABILITY_MINIMUM_HEADER_VERSION` and `LIBCXXABI_AVAILABILITY_MINIMUM_HEADER_VERSION`.

[libc++abi]: http://libcxxabi.llvm.org/
[libcxxrt]: https://github.com/libcxxrt/libcxxrt