NumPy Distutils - Users Guide

SciPy structure

Currently SciPy project consists of two packages:

  • NumPy — it provides packages like:

    • numpy.distutils - extension to Python distutils

    • numpy.f2py - a tool to bind Fortran/C codes to Python

    • numpy.core - future replacement of Numeric and numarray packages

    • numpy.lib - extra utility functions

    • numpy.testing - numpy-style tools for unit testing

    • etc

  • SciPy — a collection of scientific tools for Python.

The aim of this document is to describe how to add new tools to SciPy.

Requirements for SciPy packages

SciPy consists of Python packages, called SciPy packages, that are available to Python users via the scipy namespace. Each SciPy package may contain other SciPy packages. And so on. Therefore, the SciPy directory tree is a tree of packages with arbitrary depth and width. Any SciPy package may depend on NumPy packages but the dependence on other SciPy packages should be kept minimal or zero.

A SciPy package contains, in addition to its sources, the following files and directories:

  • setup.py — building script

  • __init__.py — package initializer

  • tests/ — directory of unittests

Their contents are described below.

The setup.py file

In order to add a Python package to SciPy, its build script (setup.py) must meet certain requirements. The most important requirement is that the package define a configuration(parent_package='',top_path=None) function which returns a dictionary suitable for passing to numpy.distutils.core.setup(..). To simplify the construction of this dictionary, numpy.distutils.misc_util provides the Configuration class, described below.

SciPy pure Python package example

Below is an example of a minimal setup.py file for a pure SciPy package:

#!/usr/bin/env python3
def configuration(parent_package='',top_path=None):
    from numpy.distutils.misc_util import Configuration
    config = Configuration('mypackage',parent_package,top_path)
    return config

if __name__ == "__main__":
    from numpy.distutils.core import setup
    #setup(**configuration(top_path='').todict())
    setup(configuration=configuration)

The arguments of the configuration function specify the name of parent SciPy package (parent_package) and the directory location of the main setup.py script (top_path). These arguments, along with the name of the current package, should be passed to the Configuration constructor.

The Configuration constructor has a fourth optional argument, package_path, that can be used when package files are located in a different location than the directory of the setup.py file.

Remaining Configuration arguments are all keyword arguments that will be used to initialize attributes of Configuration instance. Usually, these keywords are the same as the ones that setup(..) function would expect, for example, packages, ext_modules, data_files, include_dirs, libraries, headers, scripts, package_dir, etc. However, the direct specification of these keywords is not recommended as the content of these keyword arguments will not be processed or checked for the consistency of SciPy building system.

Finally, Configuration has .todict() method that returns all the configuration data as a dictionary suitable for passing on to the setup(..) function.

Configuration instance attributes

In addition to attributes that can be specified via keyword arguments to Configuration constructor, Configuration instance (let us denote as config) has the following attributes that can be useful in writing setup scripts:

  • config.name - full name of the current package. The names of parent packages can be extracted as config.name.split('.').

  • config.local_path - path to the location of current setup.py file.

  • config.top_path - path to the location of main setup.py file.

Configuration instance methods

  • config.todict() — returns configuration dictionary suitable for passing to numpy.distutils.core.setup(..) function.

  • config.paths(*paths) --- applies ``glob.glob(..) to items of paths if necessary. Fixes paths item that is relative to config.local_path.

  • config.get_subpackage(subpackage_name,subpackage_path=None) — returns a list of subpackage configurations. Subpackage is looked in the current directory under the name subpackage_name but the path can be specified also via optional subpackage_path argument. If subpackage_name is specified as None then the subpackage name will be taken the basename of subpackage_path. Any * used for subpackage names are expanded as wildcards.

  • config.add_subpackage(subpackage_name,subpackage_path=None) — add SciPy subpackage configuration to the current one. The meaning and usage of arguments is explained above, see config.get_subpackage() method.

  • config.add_data_files(*files) — prepend files to data_files list. If files item is a tuple then its first element defines the suffix of where data files are copied relative to package installation directory and the second element specifies the path to data files. By default data files are copied under package installation directory. For example,

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

    will install data files to the following locations

    <installation path of config.name package>/
      foo.dat
      fun/
        gun.dat
        pun.dat
        sun.dat
      bar/
        car.dat
      can.dat
    

    Path to data files can be a function taking no arguments and returning path(s) to data files – this is a useful when data files are generated while building the package. (XXX: explain the step when this function are called exactly)

  • config.add_data_dir(data_path) — add directory data_path recursively to data_files. The whole directory tree starting at data_path will be copied under package installation directory. If data_path is a tuple then its first element defines the suffix of where data files are copied relative to package installation directory and the second element specifies the path to data directory. By default, data directory are copied under package installation directory under the basename of data_path. For example,

    config.add_data_dir('fun')  # fun/ contains foo.dat bar/car.dat
    config.add_data_dir(('sun','fun'))
    config.add_data_dir(('gun','/full/path/to/fun'))
    

    will install data files to the following locations

    <installation path of config.name package>/
      fun/
         foo.dat
         bar/
            car.dat
      sun/
         foo.dat
         bar/
            car.dat
      gun/
         foo.dat
         bar/
            car.dat
    
  • config.add_include_dirs(*paths) — prepend paths to include_dirs list. This list will be visible to all extension modules of the current package.

  • config.add_headers(*files) — prepend files to headers list. By default, headers will be installed under <prefix>/include/pythonX.X/<config.name.replace('.','/')>/ directory. If files item is a tuple then it’s first argument specifies the installation suffix relative to <prefix>/include/pythonX.X/ path. This is a Python distutils method; its use is discouraged for NumPy and SciPy in favour of config.add_data_files(*files).

  • config.add_scripts(*files) — prepend files to scripts list. Scripts will be installed under <prefix>/bin/ directory.

  • config.add_extension(name,sources,**kw) — create and add an Extension instance to ext_modules list. The first argument name defines the name of the extension module that will be installed under config.name package. The second argument is a list of sources. add_extension method takes also keyword arguments that are passed on to the Extension constructor. The list of allowed keywords is the following: include_dirs, define_macros, undef_macros, library_dirs, libraries, runtime_library_dirs, extra_objects, extra_compile_args, extra_link_args, export_symbols, swig_opts, depends, language, f2py_options, module_dirs, extra_info, extra_f77_compile_args, extra_f90_compile_args.

    Note that config.paths method is applied to all lists that may contain paths. extra_info is a dictionary or a list of dictionaries that content will be appended to keyword arguments. The list depends 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.

    The list of sources may contain functions (‘source generators’) with a pattern def <funcname>(ext, build_dir): return <source(s) or None>. If funcname returns None, no sources are generated. And if the Extension instance has no sources after processing all source generators, no extension module will be built. This is the recommended way to conditionally define extension modules. Source generator functions are called by the build_src sub-command of numpy.distutils.

    For example, here is a typical source generator function:

    def generate_source(ext,build_dir):
        import os
        from distutils.dep_util import newer
        target = os.path.join(build_dir,'somesource.c')
        if newer(target,__file__):
            # create target file
        return target
    

    The first argument contains the Extension instance that can be useful to access its attributes like depends, sources, etc. lists and modify them during the building process. The second argument gives a path to a build directory that must be used when creating files to a disk.

  • config.add_library(name, sources, **build_info) — add a library to libraries list. Allowed keywords arguments are depends, macros, include_dirs, extra_compiler_args, f2py_options, extra_f77_compile_args, extra_f90_compile_args. See .add_extension() method for more information on arguments.

  • config.have_f77c() — return True if Fortran 77 compiler is available (read: a simple Fortran 77 code compiled successfully).

  • config.have_f90c() — return True if Fortran 90 compiler is available (read: a simple Fortran 90 code compiled successfully).

  • config.get_version() — return version string of the current package, None if version information could not be detected. This methods scans files __version__.py, <packagename>_version.py, version.py, __svn_version__.py for string variables version, __version__, <packagename>_version.

  • config.make_svn_version_py() — appends a data function to data_files list that will generate __svn_version__.py file to the current package directory. The file will be removed from the source directory when Python exits.

  • config.get_build_temp_dir() — return a path to a temporary directory. This is the place where one should build temporary files.

  • config.get_distribution() — return distutils Distribution instance.

  • config.get_config_cmd() — returns numpy.distutils config command instance.

  • config.get_info(*names)

Conversion of .src files using Templates

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.

Fortran files

This template converter will replicate all function and subroutine blocks in the file with names that contain ‘<…>’ according to the rules in ‘<…>’. The number of comma-separated words in ‘<…>’ determines the number of times the block is repeated. What these words are indicates what that repeat rule, ‘<…>’, should be replaced with in each block. All of the repeat rules in a block must contain the same number of comma-separated words indicating the number of times that block should be repeated. If the word in the repeat rule needs a comma, leftarrow, or rightarrow, then prepend it with a backslash ‘ '. If a word in the repeat rule matches ‘ \<index>’ then it will be replaced with the <index>-th word in the same repeat specification. There are two forms for the repeat rule: named and short.

Named repeat rule

A named repeat rule is useful when the same set of repeats must be used several times in a block. It is specified using <rule1=item1, item2, item3,…, itemN>, where N is the number of times the block should be repeated. On each repeat of the block, the entire expression, ‘<…>’ will be replaced first with item1, and then with item2, and so forth until N repeats are accomplished. Once a named repeat specification has been introduced, the same repeat rule may be used in the current block by referring only to the name (i.e. <rule1>).

Short repeat rule

A short repeat rule looks like <item1, item2, item3, …, itemN>. The rule specifies that the entire expression, ‘<…>’ should be replaced first with item1, and then with item2, and so forth until N repeats are accomplished.

Pre-defined names

The following predefined named repeat rules are available:

  • <prefix=s,d,c,z>

  • <_c=s,d,c,z>

  • <_t=real, double precision, complex, double complex>

  • <ftype=real, double precision, complex, double complex>

  • <ctype=float, double, complex_float, complex_double>

  • <ftypereal=float, double precision, \0, \1>

  • <ctypereal=float, double, \0, \1>

Other files

Non-Fortran files use a separate syntax for defining template blocks that should be repeated using a variable expansion similar to the named repeat rules of the Fortran-specific repeats.

NumPy Distutils preprocesses C source files (extension: .c.src) written in a custom templating language to generate C code. The @ symbol is used to wrap macro-style variables to empower a string substitution mechanism that might describe (for instance) a set of data types.

The template language blocks are delimited by /**begin repeat and /**end repeat**/ lines, which may also be nested using consecutively numbered delimiting lines such as /**begin repeat1 and /**end repeat1**/:

  1. /**begin repeat on a line by itself marks the beginning of a segment that should be repeated.

  2. Named variable expansions are defined using #name=item1, item2, item3, ..., itemN# and placed on successive lines. These variables are replaced in each repeat block with corresponding word. All named variables in the same repeat block must define the same number of words.

  3. In specifying the repeat rule for a named variable, item*N is short- hand for item, item, ..., item repeated N times. In addition, parenthesis in combination with *N can be used for grouping several items that should be repeated. Thus, #name=(item1, item2)*4# is equivalent to #name=item1, item2, item1, item2, item1, item2, item1, item2#.

  4. */ on a line by itself marks the end of the variable expansion naming. The next line is the first line that will be repeated using the named rules.

  5. Inside the block to be repeated, the variables that should be expanded are specified as @name@.

  6. /**end repeat**/ on a line by itself marks the previous line as the last line of the block to be repeated.

  7. A loop in the NumPy C source code may have a @TYPE@ variable, targeted for string substitution, which is preprocessed to a number of otherwise identical loops with several strings such as INT, LONG, UINT, ULONG. The @TYPE@ style syntax thus reduces code duplication and maintenance burden by mimicking languages that have generic type support.

The above rules may be clearer in the following template source example:

 1 /* TIMEDELTA to non-float types */
 2
 3 /**begin repeat
 4  *
 5  * #TOTYPE = BYTE, UBYTE, SHORT, USHORT, INT, UINT, LONG, ULONG,
 6  *           LONGLONG, ULONGLONG, DATETIME,
 7  *           TIMEDELTA#
 8  * #totype = npy_byte, npy_ubyte, npy_short, npy_ushort, npy_int, npy_uint,
 9  *           npy_long, npy_ulong, npy_longlong, npy_ulonglong,
10  *           npy_datetime, npy_timedelta#
11  */
12
13 /**begin repeat1
14  *
15  * #FROMTYPE = TIMEDELTA#
16  * #fromtype = npy_timedelta#
17  */
18 static void
19 @FROMTYPE@_to_@TOTYPE@(void *input, void *output, npy_intp n,
20         void *NPY_UNUSED(aip), void *NPY_UNUSED(aop))
21 {
22     const @fromtype@ *ip = input;
23     @totype@ *op = output;
24
25     while (n--) {
26         *op++ = (@totype@)*ip++;
27     }
28 }
29 /**end repeat1**/
30
31 /**end repeat**/

The preprocessing of generically-typed C source files (whether in NumPy proper or in any third party package using NumPy Distutils) is performed by conv_template.py. The type-specific C files generated (extension: .c) by these modules during the build process are ready to be compiled. This form of generic typing is also supported for C header files (preprocessed to produce .h files).

Useful functions in numpy.distutils.misc_util

  • get_numpy_include_dirs() — return a list of NumPy base include directories. NumPy base include directories contain header files such as numpy/arrayobject.h, numpy/funcobject.h etc. For installed NumPy the returned list has length 1 but when building NumPy the list may contain more directories, for example, a path to config.h file that numpy/base/setup.py file generates and is used by numpy header files.

  • append_path(prefix,path) — smart append path to prefix.

  • gpaths(paths, local_path='') — apply glob to paths and prepend local_path if needed.

  • njoin(*path) — join pathname components + convert /-separated path to os.sep-separated path and resolve .., . from paths. Ex. njoin('a',['b','./c'],'..','g') -> os.path.join('a','b','g').

  • minrelpath(path) — resolves dots in path.

  • rel_path(path, parent_path) — return path relative to parent_path.

  • def get_cmd(cmdname,_cache={}) — returns numpy.distutils command instance.

  • all_strings(lst)

  • has_f_sources(sources)

  • has_cxx_sources(sources)

  • filter_sources(sources) — return c_sources, cxx_sources, f_sources, fmodule_sources

  • get_dependencies(sources)

  • is_local_src_dir(directory)

  • get_ext_source_files(ext)

  • get_script_files(scripts)

  • get_lib_source_files(lib)

  • get_data_files(data)

  • dot_join(*args) — join non-zero arguments with a dot.

  • get_frame(level=0) — return frame object from call stack with given level.

  • cyg2win32(path)

  • mingw32() — return True when using mingw32 environment.

  • terminal_has_colors(), red_text(s), green_text(s), yellow_text(s), blue_text(s), cyan_text(s)

  • get_path(mod_name,parent_path=None) — return path of a module relative to parent_path when given. Handles also __main__ and __builtin__ modules.

  • allpath(name) — replaces / with os.sep in name.

  • cxx_ext_match, fortran_ext_match, f90_ext_match, f90_module_name_match

numpy.distutils.system_info module

  • get_info(name,notfound_action=0)

  • combine_paths(*args,**kws)

  • show_all()

numpy.distutils.cpuinfo module

  • cpuinfo

numpy.distutils.log module

  • set_verbosity(v)

numpy.distutils.exec_command module

  • get_pythonexe()

  • find_executable(exe, path=None)

  • exec_command( command, execute_in='', use_shell=None, use_tee=None, **env )

The __init__.py file

The header of a typical SciPy __init__.py is:

"""
Package docstring, typically with a brief description and function listing.
"""

# import functions into module namespace
from .subpackage import *
...

__all__ = [s for s in dir() if not s.startswith('_')]

from numpy.testing import Tester
test = Tester().test
bench = Tester().bench

Extra features in NumPy Distutils

Specifying config_fc options for libraries in setup.py script

It is possible to specify config_fc options in setup.py scripts. For example, using

config.add_library(‘library’,

sources=[…], config_fc={‘noopt’:(__file__,1)})

will compile the library sources without optimization flags.

It’s recommended to specify only those config_fc options in such a way that are compiler independent.

Getting extra Fortran 77 compiler options from source

Some old Fortran codes need special compiler options in order to work correctly. In order to specify compiler options per source file, numpy.distutils Fortran compiler looks for the following pattern:

CF77FLAGS(<fcompiler type>) = <fcompiler f77flags>

in the first 20 lines of the source and use the f77flags for specified type of the fcompiler (the first character C is optional).

TODO: This feature can be easily extended for Fortran 90 codes as well. Let us know if you would need such a feature.