complicated_build

(released under the MIT license, see the LICENSE file.)

a python extension build system (with Cython support) which allows automated compilation a mixture of C, C++, FORTRAN and Cython sources without F2PY interference.

NEW IN v1.4: support for specifying library locations (only globally.)

NEW IN v1.3: support for linking against static libs on a per-extension basis. see new 'link_to' key on extension dicts.

inspired by the problem in this stackoverflow question.

Note, this library also provides a significant improvement on the default python build system for native extensions with many source files: by default it caches all temporary build objects, and only recompiles the particular source files that have changed. The distutils default (designed for single file extensions, no doubt) recompiles all sources if any have changed. When debugging extensions, e.g. making small changes to one or two source files in a long list, this can present a significant time saving in each build/run cycle.

SUBNOTE: There is also a minor 'bug' in that, since you don't specify header files like in a Makefile, it doesn't detect changes in header files. If this is annoying enough please post a bug, and I'll work out how to fix it without breaking the existing interface (which will be a minor pain).

This module uses the default python flags for building sources, which includes all kinds of cruft that were generated in the makefile that compiled python itself. Also, the arch argument is no longer passed to the compiler, as this is not compatible with some compilers, and is now only used in temp directory names. (if such a flag is required, it will be present in the distutils.sysconfig vars without further intervention.)

note, finally, that this has been modified to act as a decorator for distutils.core.setup, see the examples of the new invocation style below.

example usage in a setup.py:

import cb

global_includes = ['/opt/local/include']
global_lib_dirs = ['/opt/local/lib']
global_macros = [("__SOME_MACRO__",), ("__MACRO_WITH_VALUE__", "5")]

extensions = [{'name' : 'mylib.impl',
    'sources' : ['mylib/impl.pyx', 'clibs/some_other_code.cpp'],
    'link_to' : ["pthread"] # will include the -l when added to the linker line.
}]

import datetime
cb.setup(extensions, global_macros = global_macros, global_includes = global_includes, 
  global_lib_dirs = global_lib_dirs)(
    name="mylib",
    version=datetime.date.today().strftime("%d%b%Y"),
    author="A N Other",
    author_email="[email protected]",
    url="http://",
    packages=["mylib"]
)

some default values to watch for:

compiler = {
  'cpp' : " ".join(distutils.sysconfig.get_config_vars('CXX', 'CPPFLAGS')), # normally something like g++
  'c' : " ".join(distutils.sysconfig.get_config_vars('CC', 'CFLAGS')), # normally something like gcc
  'f90' : 'gfortran'
}

(sysconfig functions return what was in the Makefile that built python.)

if you want to support, for example, F77 files you can do:

cb.compiler['f'] = 'gfortran'

or, if you want to use the NAG fortran compilers:

cb.compiler['f90'] = 'nagfor'

or if you're strange (cough cough) and use .cxx or .cc instead of .cpp you can do:

cb.compiler['cxx'] = cb.compiler['cpp']

however, you may need to modify the function cb._linker_vars to better reflect what runtimes you need to link against (especially if you are mixing fortran and C++ sources.)

additionally, if you don't have Cython installed on your system, the program shouldn't crash (unless you're trying to compile .pyx files, of course) -- it only gets around to importing Cython when it needs to cythonize the sources.

Finally, a more involved example:

import cb
import numpy as np

global_includes = [np.get_include()]
global_macros = [("__FORCE_CPP__",)]

extensions = [
  {'name' : 'pywat.watershed',
    'sources' : [
      'pywat/watershed.pyx',
      'clibs/watershed.cpp',
      'clibs/stripack.f90',
      'clibs/tensors/D3Vector.cpp'
  ]},
  {'name' : 'pywat.shapes',
    'sources' : [
      'pywat/shapes.pyx',
      'clibs/custom_types/d3shape.cpp',
      'clibs/custom_types/sphere.cpp',
      'clibs/custom_types/polyhedron.cpp',
      'clibs/custom_types/cylinder.cpp',
      'clibs/tensors/D3Vector.cpp'
  ]}
]

import datetime
cb.setup(extensions, global_macros = global_macros, global_includes = global_includes)(
  name="pywat",
  version=datetime.date.today().strftime("%d%b%Y"),
  author="Joe Jordan",
  author_email="[email protected]",
  url="TBA",
  packages=["pywat"]
)

INDEPENDENCE

If you don't want your project to require the user to come and find my library and install it, you can bundle it with your software as follows:

try:
  import cb
except ImportError:
  print "downloading complicated build..."
  import urllib2
  response = urllib2.urlopen('https://raw.github.com/joe-jordan/complicated_build/master/cb/__init__.py')
  content = response.read()
  f = open('cb.py', 'w')
  f.write(content)
  f.close()
  import cb
  print "done!"

You can see this setup in action in one of my other projects, pyvoro, the snippet is simply included in the top of setup.py