Is it a good practice for a module to recursively import its submodules?

Question

I see a common pattern with Python modules. Sometimes __init__.py imports its submodules:

E.g.

mymod/
    __init__.py:
        from . import subm1, subm2
        __all__ = ['subm1', 'subm2']

    subm1.py:
        def subf1():
            ...
        __all__ = ['subf1']

    subm2.py:
        def subf2():
            ...
        __all__ = ['subf2']

So the client code only has to do this:

import mymod
mymod.subm1.subf1()
mymod.subm2.subf2()

Instead of:

import mymod.subm1
import mymod.subm2
mymod.subm1.subf1()
mymod.subm2.subf2()

Which one is The Best Practice^TM? Is there anything wrong with the recursive import or should I stick to the default behaviour? What's the rationale for the default behaviour?

Answer 1

This would depend on the usage of your module. If the submodules are just singleton objects that happen to be written as modules instead of having, eg, Java-like classes with all-static methods, recursive import is probably fine. On the other hand, if your module is grouping of several related sub modules (see, eg, the stdlib html module), you would want explicit submodule imports.

In general, a recursive import setup says "any client of this module will always want all of these". Consider whether that is true for your module (and other modules, case-by-case), and go from there.