Problem

Distutils requires that you manually specify each directory and data-file to be included in the distribution. For packages with large and deep sub-package hierarchies it can be a pain to keep this list in sync with the code, particularly as forgetting an entry is not noticable until a user happens to report that a resource is missing.

Solution

Use an automatic scanning mechanism to generate the data_files parameter for setup:

import os
import imp

def non_python_files(path):
    """ Return all non-python-file filenames in path """
    result = []
    all_results = []
    module_suffixes = [info[0] for info in imp.get_suffixes()]
    ignore_dirs = ['cvs']
    for item in os.listdir(path):
        name = os.path.join(path, item)
        if (
            os.path.isfile(name) and
            os.path.splitext(item)[1] not in module_suffixes
            ):
            result.append(name)
        elif os.path.isdir(name) and item.lower() not in ignore_dirs:
            all_results.extend(non_python_files(name))
    if result:
        all_results.append((path, result))
    return all_results

Then call it for the directories which contain data-files you want to include.

data_files = (
    non_python_files('pytable') +
    non_python_files(os.path.join('pytable', 'doc'))
    )

and pass the result to setup:

setup (
    name = "pytable",
    version = "0.7.7a",
    data_files = data_files,
    cmdclass = {'install_data':smart_install_data},
    **extra_arguments
    )

(See DistutilsInstallDataScattered for the smart_install_data command.)

You can see a real-world usage example in the PyTable setup script

Discussion

Again, there should be some way to do this with a distutils template processing call or something, but this direct processing approach works well enough. This approach is very similar to DistutilsAutoPackageDiscovery.

CategoryDistutilsCookbook