Using shared libraries¶
Python projects may build shared libraries as part of their project, or link
with shared libraries from a dependency. This tends to be a common source of
issues, hence this page aims to explain how to include shared libraries in
wheels, any limitations and gotchas, and how support is implemented in
meson-python
under the hood.
We distinguish between internal shared libraries that are built as part of
the project, and external shared libraries that are provided by project
dependencies and that are linked with the project build artifacts.
For internal shared libraries, we also distinguish whether the shared library
is being installed to its default system location (typically
/usr/local/lib
on Unix-like systems, and C:\\lib
on Windows - we call
this libdir
in this guide) or to a location in site-packages
within the
Python package install tree. All these scenarios are (or will be) supported,
with some caveats:
shared library source |
install location |
Windows |
macOS |
Linux |
---|---|---|---|---|
internal |
libdir |
no (1) |
✓ |
✓ |
internal |
site-packages |
✓ |
✓ |
✓ |
external |
n/a |
✓ (2) |
✓ |
✓ |
Internal shared libraries on Windows cannot be automatically handled correctly, and currently
meson-python
therefore raises an error for them. PR meson-python#551 may improve that situation in the near future.External shared libraries require
delvewheel
usage on Windows (or some equivalent way, like amending the DLL search path to include the directory in which the external shared library is located). Due to the lack of RPATH support on Windows, there is no good way around this.
Internal shared libraries¶
A shared library produced by library()
or shared_library()
built like this
example_lib = shared_library(
'example',
'examplelib.c',
install: true,
)
is installed to libdir
by default. If the only reason the shared library exists
is to be used inside the Python package being built, then it is best to modify
the install location to be within the Python package itself:
install_path: py.get_install_dir() / 'mypkg/subdir'
Then an extension module in the same install directory can link against the
shared library in a portable manner by using install_rpath
:
py3.extension_module('_extmodule',
'_extmodule.c',
link_with: example_lib,
install: true,
subdir: 'mypkg/subdir',
install_rpath: '$ORIGIN'
)
The above method will work as advertised on macOS and Linux; meson-python
does
nothing special for this case. Windows needs some special handling though, due to
the lack of RPATH support:
def _append_to_sharedlib_load_path():
"""
Ensure the shared libraries in this package can be loaded on Windows.
Windows lacks a concept equivalent to RPATH: Python extension modules
cannot find DLLs installed outside the DLL search path. This function
ensures that the location of the shared libraries distributed inside this
Python package is in the DLL search path of the process.
The Windows DLL search path includes the object depending on it is located:
the DLL search path needs to be augmented only when the Python extension
modules and the DLLs they require are installed in separate directories.
Cygwin does not have the same default library search path: all locations
where the shared libraries are installed need to be added to the search
path.
This function is very similar to the snippet inserted into the main
``__init__.py`` of a package by ``delvewheel`` when it vendors external
shared libraries.
.. note::
`os.add_dll_directory` is only available for Python 3.8 and later, and
in the Conda ``python`` packages it works as advertised only for
version 3.10 and later. For older Python versions, pre-loading the DLLs
with `ctypes.WinDLL` may be preferred.
"""
basedir = os.path.dirname(__file__)
subdir = os.path.join(basedir, 'sub')
if os.name == 'nt':
os.add_dll_directory(subdir)
elif sys.platform == 'cygwin':
os.environ['PATH'] = os.pathsep.join((os.environ['PATH'], basedir, subdir))
_append_to_sharedlib_load_path()
If an internal shared library is not only used as part of a Python package, but
for example also as a regular shared library in a C/C++ project or as a
standalone library, then the method shown above won’t work. The library is
then marked for installation into the system default libdir
location.
Actually installing into libdir
isn’t possible with wheels, hence
meson-python
will instead do the following on platforms other than
Windows:
Install the shared library to
<project-name>.mesonpy.libs
(i.e., a top-level directory in the wheel, which on install will end up insite-packages
).Rewrite RPATH entries for install targets that depend on the shared library to point to that new install location instead.
This will make the shared library work automatically, with no other action needed from the package author. However, currently an error is raised for this situation on Windows. This is documented also in Limitations.
External shared libraries¶
External shared libraries are installed somewhere on the build machine, and
usually detected by a dependency()
or compiler.find_library()
call in a
meson.build
file. When a Python extension module or executable uses the
dependency, the shared library will be linked against at build time.
If the shared library is located in a directory on the loader search path,
the wheel created by meson-python
will work locally when installed.
If it’s in a non-standard location however, the shared library will go
missing at runtime. The Python extension module linked against it needs an
RPATH entry - and Meson will not automatically manage RPATH entries for you.
Hence you’ll need to add the needed RPATH yourself, for example by adding
-Wl,rpath=/path/to/dir/sharedlib/is/in
to LDFLAGS
before starting
the build. In case you run into this problem after a wheel is built and
installed, adding that same path to the LD_LIBRARY_PATH
environment variable is a quick way of
checking if that is indeed the problem.
On Windows, the solution is similar - the shared library can either be
preloaded, or the directory that the library is located in added to the DLL
search path with os.add_dll_directory
, or vendored into the wheel with
delvewheel
in order to make the built Python package usable locally.
Publishing wheels which depend on external shared libraries¶
On all platforms, wheels which depend on external shared libraries usually need
post-processing to make them usable on machines other than the one on which
they were built. This is because the RPATH entry for an external shared library
contains a path specific to the build machine. This post-processing is done by
tools like auditwheel
(Linux), delvewheel
(Windows), delocate
(macOS) or repair-wheel
(any platform, wraps the other tools).
Running any of those tools on a wheel produced by meson-python
will vendor
the external shared library into the wheel and rewrite the RPATH entries (it
may also do some other things, like symbol mangling).
On Windows, the package author may also have to add the preloading like shown
above with _append_to_sharedlib_load_path()
to the main __init__.py
of
the package, delvewheel
may or may not take care of this (please check its
documentation if your shared library goes missing at runtime).
Note that we don’t cover using shared libraries contained in another wheel and depending on such a wheel at runtime in this guide. This is inherently complex and not recommended (you need to be in control of both packages, or upgrades may be impossible/breaking).
Using libraries from a Meson subproject¶
It can often be useful to build a shared library in a
Meson subproject, for example as
a fallback in case an external dependency isn’t detected. There are two main
strategies for folding a library built in a subproject into a wheel built with
meson-python
:
Build the library as a static library instead of a shared library, and link it into a Python extension module that needs it.
Build the library as a shared library, and either change its install path to be within the Python package’s tree, or rely on
meson-python
to fold it into the wheel when it’d otherwise be installed tolibdir
.
Option (1) tends to be easier, so unless the library of interest cannot be built as a static library or it would inflate the wheel size too much because it’s needed by multiple Python extension modules, we recommend trying option (1) first.
A typical C or C++ project providing a library to link against tends to provide
(a) one or more library()
targets, which can be built as shared, static, or both,
and (b) headers, pkg-config files, tests and perhaps other development targets
that are needed to use the library()
target(s). One of the challenges to use
such projects as a subproject is that the headers and other installable targets
are targeting system locations (e.g., <prefix>/include/
) which isn’t supported
by wheels and hence meson-python
errors out when it encounters such an install
target. This is perhaps the main issue one encounters with subproject usage,
and the following two sections discuss how options (1) and (2) can work around
that.
Static library from subproject¶
The major advantage of building a library target as static and folding it directly
into an extension module is that no targets from the subproject need to be installed.
To configure the subproject for this use case, add the following to the
pyproject.toml
file of your package:
[tool.meson-python.args]
setup = ['--default-library=static']
install = ['--skip-subprojects']
This ensures that library
targets are built as static, and nothing gets installed.
To then link against the static library in the subproject, say for a subproject
named bar
with the main library target contained in a bar_dep
dependency,
add this to your meson.build
file:
bar_proj = subproject('bar')
bar_dep = bar_proj.get_variable('bar_dep')
py.extension_module(
'_example',
'_examplemod.c',
dependencies: bar_dep,
install: true,
)
That is all!
Shared library from subproject¶
If we can’t use the static library approach from the section above and we need
a shared library, then we must have install: true
for that shared library
target. This can only work if we can pass some build option to the subproject
that tells it to only install the shared library and not headers or other
targets that we don’t need. Install tags don’t work per subproject, so
this will look something like:
foo_subproj = subproject('foo',
default_options: {
# This is a custom option - if it doesn't exist, can you add it
# upstream or in WrapDB?
'only_install_main_lib': true,
})
foo_dep = foo_subproj.get_variable('foo_dep')
Now we can use foo_dep
like a normal dependency, meson-python
will
include it into the wheel in <project-name>.mesonpy.libs
just like an
internal shared library that targets libdir
(see
Internal shared libraries).
Remember: this method doesn’t support Windows (yet)!