Packaging (numpy.distutils)#

Warning

numpy.distutils is deprecated, and will be removed for Python >= 3.12. For more details, see Status of numpy.distutils and migration advice

Warning

Note that setuptools does major releases often and those may contain changes that break numpy.distutils, which will not be updated anymore for new setuptools versions. It is therefore recommended to set an upper version bound in your build configuration for the last known version of setuptools that works with your build.

NumPy provides enhanced distutils functionality to make it easier to build and install sub-packages, auto-generate code, and extension modules that use Fortran-compiled libraries. To use features of NumPy distutils, use the setup command from numpy.distutils.core. A useful Configuration class is also provided in numpy.distutils.misc_util that can make it easier to construct keyword arguments to pass to the setup function (by passing the dictionary obtained from the todict() method of the class). More information is available in the NumPy distutils - users guide.

The choice and location of linked libraries such as BLAS and LAPACK as well as include paths and other such build options can be specified in a site.cfg file located in the NumPy root repository or a .numpy-site.cfg file in your home directory. See the site.cfg.example example file included in the NumPy repository or sdist for documentation.

Modules in numpy.distutils#

ccompiler

ccompiler_opt

Provides the CCompilerOpt class, used for handling the CPU/hardware optimization, starting from parsing the command arguments, to managing the relation between the CPU baseline and dispatch-able features, also generating the required C headers and ending with compiling the sources with proper compiler's flags.

cpuinfo.cpu

core.Extension(name, sources[, ...])

Parameters:

exec_command

exec_command

log.set_verbosity(v[, force])

system_info.get_info(name[, notfound_action])

notfound_action:

system_info.get_standard_file(fname)

Returns a list of files named 'fname' from 1) System-wide directory (directory-location of this module) 2) Users HOME directory (os.environ['HOME']) 3) Local directory

Configuration class#

class numpy.distutils.misc_util.Configuration(package_name=None, parent_name=None, top_path=None, package_path=None, **attrs)[source]#

Construct a configuration instance for the given package name. If parent_name is not None, then construct the package as a sub-package of the parent_name package. If top_path and package_path are None then they are assumed equal to the path of the file this instance was created in. The setup.py files in the numpy distribution are good examples of how to use the Configuration instance.

todict()[source]#

Return a dictionary compatible with the keyword arguments of distutils setup function.

Examples

>>> setup(**config.todict())                           
get_distribution()[source]#

Return the distutils distribution object for self.

get_subpackage(subpackage_name, subpackage_path=None, parent_name=None, caller_level=1)[source]#

Return list of subpackage configurations.

Parameters:
subpackage_namestr or None

Name of the subpackage to get the configuration. ‘*’ in subpackage_name is handled as a wildcard.

subpackage_pathstr

If None, then the path is assumed to be the local path plus the subpackage_name. If a setup.py file is not found in the subpackage_path, then a default configuration is used.

parent_namestr

Parent name.

add_subpackage(subpackage_name, subpackage_path=None, standalone=False)[source]#

Add a sub-package to the current Configuration instance.

This is useful in a setup.py script for adding sub-packages to a package.

Parameters:
subpackage_namestr

name of the subpackage

subpackage_pathstr

if given, the subpackage path such as the subpackage is in subpackage_path / subpackage_name. If None,the subpackage is assumed to be located in the local path / subpackage_name.

standalonebool
add_data_files(*files)[source]#

Add data files to configuration data_files.

Parameters:
filessequence

Argument(s) can be either

  • 2-sequence (<datadir prefix>,<path to data file(s)>)

  • paths to data files where python datadir prefix defaults to package dir.

Notes

The form of each element of the files sequence is very flexible allowing many combinations of where to get the files from the package and where they should ultimately be installed on the system. The most basic usage is for an element of the files argument sequence to be a simple filename. This will cause that file from the local path to be installed to the installation path of the self.name package (package path). The file argument can also be a relative path in which case the entire relative path will be installed into the package directory. Finally, the file can be an absolute path name in which case the file will be found at the absolute path name but installed to the package path.

This basic behavior can be augmented by passing a 2-tuple in as the file argument. The first element of the tuple should specify the relative path (under the package install directory) where the remaining sequence of files should be installed to (it has nothing to do with the file-names in the source distribution). The second element of the tuple is the sequence of files that should be installed. The files in this sequence can be filenames, relative paths, or absolute paths. For absolute paths the file will be installed in the top-level package installation directory (regardless of the first argument). Filenames and relative path names will be installed in the package install directory under the path name given as the first element of the tuple.

Rules for installation paths:

  1. file.txt -> (., file.txt)-> parent/file.txt

  2. foo/file.txt -> (foo, foo/file.txt) -> parent/foo/file.txt

  3. /foo/bar/file.txt -> (., /foo/bar/file.txt) -> parent/file.txt

  4. *.txt -> parent/a.txt, parent/b.txt

  5. foo/*.txt`` -> parent/foo/a.txt, parent/foo/b.txt

  6. */*.txt -> (*, */*.txt) -> parent/c/a.txt, parent/d/b.txt

  7. (sun, file.txt) -> parent/sun/file.txt

  8. (sun, bar/file.txt) -> parent/sun/file.txt

  9. (sun, /foo/bar/file.txt) -> parent/sun/file.txt

  10. (sun, *.txt) -> parent/sun/a.txt, parent/sun/b.txt

  11. (sun, bar/*.txt) -> parent/sun/a.txt, parent/sun/b.txt

  12. (sun/*, */*.txt) -> parent/sun/c/a.txt, parent/d/b.txt

An additional feature is that the path to a data-file can actually be a function that takes no arguments and returns the actual path(s) to the data-files. This is useful when the data files are generated while building the package.

Examples

Add files to the list of data_files to be included with the package.

>>> self.add_data_files('foo.dat',
...     ('fun', ['gun.dat', 'nun/pun.dat', '/tmp/sun.dat']),
...     'bar/cat.dat',
...     '/full/path/to/can.dat')                   

will install these data files to:

<package install directory>/
 foo.dat
 fun/
   gun.dat
   nun/
     pun.dat
 sun.dat
 bar/
   car.dat
 can.dat

where <package install directory> is the package (or sub-package) directory such as ‘/usr/lib/python2.4/site-packages/mypackage’ (‘C: Python2.4 Lib site-packages mypackage’) or ‘/usr/lib/python2.4/site- packages/mypackage/mysubpackage’ (‘C: Python2.4 Lib site-packages mypackage mysubpackage’).

add_data_dir(data_path)[source]#

Recursively add files under data_path to data_files list.

Recursively add files under data_path to the list of data_files to be installed (and distributed). The data_path can be either a relative path-name, or an absolute path-name, or a 2-tuple where the first argument shows where in the install directory the data directory should be installed to.

Parameters:
data_pathseq or str

Argument can be either

  • 2-sequence (<datadir suffix>, <path to data directory>)

  • path to data directory where python datadir suffix defaults to package dir.

Notes

Rules for installation paths:

foo/bar -> (foo/bar, foo/bar) -> parent/foo/bar
(gun, foo/bar) -> parent/gun
foo/* -> (foo/a, foo/a), (foo/b, foo/b) -> parent/foo/a, parent/foo/b
(gun, foo/*) -> (gun, foo/a), (gun, foo/b) -> gun
(gun/*, foo/*) -> parent/gun/a, parent/gun/b
/foo/bar -> (bar, /foo/bar) -> parent/bar
(gun, /foo/bar) -> parent/gun
(fun/*/gun/*, sun/foo/bar) -> parent/fun/foo/gun/bar

Examples

For example suppose the source directory contains fun/foo.dat and fun/bar/car.dat:

>>> self.add_data_dir('fun')                       
>>> self.add_data_dir(('sun', 'fun'))              
>>> self.add_data_dir(('gun', '/full/path/to/fun'))

Will install data-files to the locations:

<package install directory>/
  fun/
    foo.dat
    bar/
      car.dat
  sun/
    foo.dat
    bar/
      car.dat
  gun/
    foo.dat
    car.dat
add_include_dirs(*paths)[source]#

Add paths to configuration include directories.

Add the given sequence of paths to the beginning of the include_dirs list. This list will be visible to all extension modules of the current package.

add_headers(*files)[source]#

Add installable headers to configuration.

Add the given sequence of files to the beginning of the headers list. By default, headers will be installed under <python- include>/<self.name.replace(‘.’,’/’)>/ directory. If an item of files is a tuple, then its first argument specifies the actual installation location relative to the <python-include> path.

Parameters:
filesstr or seq

Argument(s) can be either:

  • 2-sequence (<includedir suffix>,<path to header file(s)>)

  • path(s) to header file(s) where python includedir suffix will default to package name.

add_extension(name, sources, **kw)[source]#

Add extension to configuration.

Create and add an Extension instance to the ext_modules list. This method also takes the following optional keyword arguments that are passed on to the Extension constructor.

Parameters:
namestr

name of the extension

sourcesseq

list of the sources. The list of sources may contain functions (called source generators) which must take an extension instance and a build directory as inputs and return a source file or list of source files or None. If None is returned then no sources are generated. If the Extension instance has no sources after processing all source generators, then no extension module is built.

include_dirs
define_macros
undef_macros
library_dirs
libraries
runtime_library_dirs
extra_objects
extra_compile_args
extra_link_args
extra_f77_compile_args
extra_f90_compile_args
export_symbols
swig_opts
depends

The depends list contains paths to files or directories that the sources of the extension module depend on. If any path in the depends list is newer than the extension module, then the module will be rebuilt.

language
f2py_options
module_dirs
extra_infodict or list

dict or list of dict of keywords to be appended to keywords.

Notes

The self.paths(…) method is applied to all lists that may contain paths.

add_library(name, sources, **build_info)[source]#

Add library to configuration.

Parameters:
namestr

Name of the extension.

sourcessequence

List of the sources. The list of sources may contain functions (called source generators) which must take an extension instance and a build directory as inputs and return a source file or list of source files or None. If None is returned then no sources are generated. If the Extension instance has no sources after processing all source generators, then no extension module is built.

build_infodict, optional

The following keys are allowed:

  • depends

  • macros

  • include_dirs

  • extra_compiler_args

  • extra_f77_compile_args

  • extra_f90_compile_args

  • f2py_options

  • language

add_scripts(*files)[source]#

Add scripts to configuration.

Add the sequence of files to the beginning of the scripts list. Scripts will be installed under the <prefix>/bin/ directory.

add_installed_library(name, sources, install_dir, build_info=None)[source]#

Similar to add_library, but the specified library is installed.

Most C libraries used with distutils are only used to build python extensions, but libraries built through this method will be installed so that they can be reused by third-party packages.

Parameters:
namestr

Name of the installed library.

sourcessequence

List of the library’s source files. See add_library for details.

install_dirstr

Path to install the library, relative to the current sub-package.

build_infodict, optional

The following keys are allowed:

  • depends

  • macros

  • include_dirs

  • extra_compiler_args

  • extra_f77_compile_args

  • extra_f90_compile_args

  • f2py_options

  • language

Returns:
None

Notes

The best way to encode the options required to link against the specified C libraries is to use a “libname.ini” file, and use get_info to retrieve the required options (see add_npy_pkg_config for more information).

add_npy_pkg_config(template, install_dir, subst_dict=None)[source]#

Generate and install a npy-pkg config file from a template.

The config file generated from template is installed in the given install directory, using subst_dict for variable substitution.

Parameters:
templatestr

The path of the template, relatively to the current package path.

install_dirstr

Where to install the npy-pkg config file, relatively to the current package path.

subst_dictdict, optional

If given, any string of the form @key@ will be replaced by subst_dict[key] in the template file when installed. The install prefix is always available through the variable @prefix@, since the install prefix is not easy to get reliably from setup.py.

Notes

This works for both standard installs and in-place builds, i.e. the @prefix@ refer to the source directory for in-place builds.

Examples

config.add_npy_pkg_config('foo.ini.in', 'lib', {'foo': bar})

Assuming the foo.ini.in file has the following content:

[meta]
Name=@foo@
Version=1.0
Description=dummy description

[default]
Cflags=-I@prefix@/include
Libs=

The generated file will have the following content:

[meta]
Name=bar
Version=1.0
Description=dummy description

[default]
Cflags=-Iprefix_dir/include
Libs=

and will be installed as foo.ini in the ‘lib’ subpath.

When cross-compiling with numpy distutils, it might be necessary to use modified npy-pkg-config files. Using the default/generated files will link with the host libraries (i.e. libnpymath.a). For cross-compilation you of-course need to link with target libraries, while using the host Python installation.

You can copy out the numpy/core/lib/npy-pkg-config directory, add a pkgdir value to the .ini files and set NPY_PKG_CONFIG_PATH environment variable to point to the directory with the modified npy-pkg-config files.

Example npymath.ini modified for cross-compilation:

[meta]
Name=npymath
Description=Portable, core math library implementing C99 standard
Version=0.1

[variables]
pkgname=numpy.core
pkgdir=/build/arm-linux-gnueabi/sysroot/usr/lib/python3.7/site-packages/numpy/core
prefix=${pkgdir}
libdir=${prefix}/lib
includedir=${prefix}/include

[default]
Libs=-L${libdir} -lnpymath
Cflags=-I${includedir}
Requires=mlib

[msvc]
Libs=/LIBPATH:${libdir} npymath.lib
Cflags=/INCLUDE:${includedir}
Requires=mlib
paths(*paths, **kws)[source]#

Apply glob to paths and prepend local_path if needed.

Applies glob.glob(…) to each path in the sequence (if needed) and pre-pends the local_path if needed. Because this is called on all source lists, this allows wildcard characters to be specified in lists of sources for extension modules and libraries and scripts and allows path-names be relative to the source directory.

get_config_cmd()[source]#

Returns the numpy.distutils config command instance.

get_build_temp_dir()[source]#

Return a path to a temporary directory where temporary files should be placed.

have_f77c()[source]#

Check for availability of Fortran 77 compiler.

Use it inside source generating function to ensure that setup distribution instance has been initialized.

Notes

True if a Fortran 77 compiler is available (because a simple Fortran 77 code was able to be compiled successfully).

have_f90c()[source]#

Check for availability of Fortran 90 compiler.

Use it inside source generating function to ensure that setup distribution instance has been initialized.

Notes

True if a Fortran 90 compiler is available (because a simple Fortran 90 code was able to be compiled successfully)

get_version(version_file=None, version_variable=None)[source]#

Try to get version string of a package.

Return a version string of the current package or None if the version information could not be detected.

Notes

This method scans files named __version__.py, <packagename>_version.py, version.py, and __svn_version__.py for string variables version, __version__, and <packagename>_version, until a version number is found.

make_svn_version_py(delete=True)[source]#

Appends a data function to the data_files list that will generate __svn_version__.py file to the current package directory.

Generate package __svn_version__.py file from SVN revision number, it will be removed after python exits but will be available when sdist, etc commands are executed.

Notes

If __svn_version__.py existed before, nothing is done.

This is intended for working with source directories that are in an SVN repository.

make_config_py(name='__config__')[source]#

Generate package __config__.py file containing system_info information used during building the package.

This file is installed to the package installation directory.

get_info(*names)[source]#

Get resources information.

Return information (from system_info.get_info) for all of the names in the argument list in a single dictionary.

Building Installable C libraries#

Conventional C libraries (installed through add_library) are not installed, and are just used during the build (they are statically linked). An installable C library is a pure C library, which does not depend on the python C runtime, and is installed such that it may be used by third-party packages. To build and install the C library, you just use the method add_installed_library instead of add_library, which takes the same arguments except for an additional install_dir argument:

.. hidden in a comment so as to be included in refguide but not rendered documentation
  >>> import numpy.distutils.misc_util
  >>> config = np.distutils.misc_util.Configuration(None, '', '.')
  >>> with open('foo.c', 'w') as f: pass

>>> config.add_installed_library('foo', sources=['foo.c'], install_dir='lib')

npy-pkg-config files#

To make the necessary build options available to third parties, you could use the npy-pkg-config mechanism implemented in numpy.distutils. This mechanism is based on a .ini file which contains all the options. A .ini file is very similar to .pc files as used by the pkg-config unix utility:

[meta]
Name: foo
Version: 1.0
Description: foo library

[variables]
prefix = /home/user/local
libdir = ${prefix}/lib
includedir = ${prefix}/include

[default]
cflags = -I${includedir}
libs = -L${libdir} -lfoo

Generally, the file needs to be generated during the build, since it needs some information known at build time only (e.g. prefix). This is mostly automatic if one uses the Configuration method add_npy_pkg_config. Assuming we have a template file foo.ini.in as follows:

[meta]
Name: foo
Version: @version@
Description: foo library

[variables]
prefix = @prefix@
libdir = ${prefix}/lib
includedir = ${prefix}/include

[default]
cflags = -I${includedir}
libs = -L${libdir} -lfoo

and the following code in setup.py:

>>> config.add_installed_library('foo', sources=['foo.c'], install_dir='lib')
>>> subst = {'version': '1.0'}
>>> config.add_npy_pkg_config('foo.ini.in', 'lib', subst_dict=subst)

This will install the file foo.ini into the directory package_dir/lib, and the foo.ini file will be generated from foo.ini.in, where each @version@ will be replaced by subst_dict['version']. The dictionary has an additional prefix substitution rule automatically added, which contains the install prefix (since this is not easy to get from setup.py). npy-pkg-config files can also be installed at the same location as used for numpy, using the path returned from get_npy_pkg_dir function.

Reusing a C library from another package#

Info are easily retrieved from the get_info function in numpy.distutils.misc_util:

>>> info = np.distutils.misc_util.get_info('npymath')
>>> config.add_extension('foo', sources=['foo.c'], extra_info=info)
<numpy.distutils.extension.Extension('foo') at 0x...>

An additional list of paths to look for .ini files can be given to get_info.

Conversion of .src files#

NumPy distutils supports automatic conversion of source files named <somefile>.src. This facility can be used to maintain very similar code blocks requiring only simple changes between blocks. During the build phase of setup, if a template file named <somefile>.src is encountered, a new file named <somefile> is constructed from the template and placed in the build directory to be used instead. Two forms of template conversion are supported. The first form occurs for files named <file>.ext.src where ext is a recognized Fortran extension (f, f90, f95, f77, for, ftn, pyf). The second form is used for all other cases. See Conversion of .src files using Templates.