How can I import a python package or __init__.py file using the full path?

Question

I have a package: pyfoo in a directory /home/user/somedir

The package is a normal one with '/home/user/somedir/pyfoo/__init__.py'

I would like to be able to import this module using its full-path '/home/user/somedir/pyfoo/'

How to do this for non-package modules is shown in: How to import a module given the full path?

But I seem to be unable to get this to work when the module is a package.

---

The very odd use case I found for this is I was embed deep into a script execution before writing to a file with h5py.

I had to uninstall h5py and reinstall a parallel version with openmpi, but even though it was uninstalled h5py(serial) was still in memory. I did not want to restart, because the script took a long time. I attempted to reload the module, but it did not work. I also tried to import it from its filename using the directory and with __init__.py, but I got relative import errors. I even tried adding the new install location to sys.path and executing a normal import, but that too failed.

I already have a import_from_filepath function in my personal utility library, and I'd like to add the import_from_dirpath as well, but I'm having a bit of trouble figuring out how its done.

---

Here is a script illustrating the issue:

    # Define two temporary modules that are not in sys.path
    # and have the same name but different values.
    import sys, os, os.path
    def ensuredir(path):
        if not os.path.exists(path):
            os.mkdir(path)
    ensuredir('tmp')
    ensuredir('tmp/tmp1')
    ensuredir('tmp/tmp2')
    ensuredir('tmp/tmp1/testmod')
    ensuredir('tmp/tmp2/testmod')
    with open('tmp/tmp1/testmod/__init__.py', 'w') as file_:
        file_.write('foo = \"spam\"\nfrom . import sibling')
    with open('tmp/tmp1/testmod/sibling.py', 'w') as file_:
        file_.write('bar = \"ham\"')
    with open('tmp/tmp2/testmod/__init__.py', 'w') as file_:
        file_.write('foo = \"eggs\"\nfrom . import sibling')
    with open('tmp/tmp1/testmod/sibling.py', 'w') as file_:
        file_.write('bar = \"jam\"')

    # Neither module should be importable through the normal mechanism
    try:
        import testmod
        assert False, 'should fail'
    except ImportError as ex:
        pass

    # Try temporarilly adding the directory of a module to the path
    sys.path.insert(0, 'tmp/tmp1')
    testmod1 = __import__('testmod', globals(), locals(), 0)
    sys.path.remove('tmp/tmp1')
    print(testmod1.foo)
    print(testmod1.sibling.bar)

    sys.path.insert(0, 'tmp/tmp2')
    testmod2 = __import__('testmod', globals(), locals(), 0)
    sys.path.remove('tmp/tmp2')
    print(testmod2.foo)
    print(testmod2.sibling.bar)

    assert testmod1.foo == "spam"
    assert testmod1.sibling.bar == "ham"

    # Fails, returns spam
    assert testmod2.foo == "eggs"
    assert testmod2.sibling.bar == "jam"

The sys.path method does not import via file path. It defaults to importing the previously loaded module.

Answer 1

In general you don't do imports from absolute paths in Python. It is possible but it break things like the imports in that module.

What you do is to adapt the PYTHONPATH. It tells python where to look for the stuff you want to import. You can set the environmental variable PYTHONPATH before calling your script or you can see and manipulate it at runtime. To do so, change sys.path. It's a list, you can append() other paths.

An other option is to use virtual environments they offer a special python environments for different projects and project setups. Read http://docs.python-guide.org/en/latest/dev/virtualenvs/ to get an idea of it.

If you still want to import by path, read How to import a module given the full path? .