Python import module sharing name with a function in __init__.py

In an import statement, the module reference never uses attribute lookups. The statements

import parent.a  # as ...

and

from parent.a import ...  # as ...

will always look for parent.a in the sys.modules namespace before trying to further initiate module loading from disk.

However, for from ... import name statements, Python does look at attributes of the resolved module to find name, before looking for submodules.

Module globals and the attributes on a module object are the same thing. On import, Python adds submodules as attributes (so globals) to the parent module, but you are free to overwrite those attributes, as you did in your code. However, when you then use an import with the parent.a module path, attributes do not come into play.

From the Submodules section of the Python import system reference documentation:

When a submodule is loaded using any mechanism [...] a binding is placed in the parent module’s namespace to the submodule object. For example, if package spam has a submodule foo, after importing spam.foo, spam will have an attribute foo which is bound to the submodule.

Your import parent.a as _a statement adds two names to the parent namespace; first a is added pointing to the parent.a submodule, and then _a is also set, pointing to the same object.

Your next line replaces the name a with a binding to the 'some string' object.

The Searching section of the same details how Python goes about finding a module when you import:

To begin the search, Python needs the fully qualified name of the module [...] being imported.

[...]

This name will be used in various phases of the import search, and it may be the dotted path to a submodule, e.g. foo.bar.baz. In this case, Python first tries to import foo, then foo.bar, and finally foo.bar.baz. If any of the intermediate imports fail, a ModuleNotFoundError is raised.

then further on

The first place checked during import search is sys.modules. This mapping serves as a cache of all modules that have been previously imported, including the intermediate paths. So if foo.bar.baz was previously imported, sys.modules will contain entries for foo, foo.bar, and foo.bar.baz. Each key will have as its value the corresponding module object.

During import, the module name is looked up in sys.modules and if present, the associated value is the module satisfying the import, and the process completes. [...] If the module name is missing, Python will continue searching for the module.

So when trying to import parent.a all that matters is that sys.modules['parent.a'] exists. sys.modules['parent'].a is not consulted.

Only from module import ... would ever look at attributes. From the import statement documentation:

The from form uses a slightly more complex process:

1. find the module specified in the from clause, loading and initializing it if necessary;
2. for each of the identifiers specified in the import clauses:
   1. check if the imported module has an attribute by that name
   2. if not, attempt to import a submodule with that name and then check the imported module again for that attribute
   3. [...]

So from parent import _a would work, as would from parent import a, and you'd get the parent.a submodule and the 'some string' object, respectively.

Note that sys.modules is writable, if you must have import parent._a work, you can always just alter sys.modules directly:

sys.modules['parent._a'] = sys.modules['parent.a']  # make parent._a an alias for parent.a
import parent._a  # works now

I think I have a coherent understanding of this problem now, just documenting my findings in case others run into this.

What Martijn said above is mostly true, expanding on that answer, import parent.a as _a is a two step process. The first step is module lookup of parent.a, which never goes through attribute lookup, and then it does a binding onto sys.modules, and then an attribute binding of the module to attribute a in parent. In fact this is all you get if you only use import parent.a. This part is described thoroughly by the previous answer.

The second part as _a does an attribute lookup of parent.a, and binds it onto the name _a. So to answer my original question, now if I go outside and start an interactive Python interpreter, now parent.a has been overwritten to the string in __init__.py, and import parent.a as the_a; the_a would get me the string. In fact, this is the same as import parent.a; parent.a. Both the_a and parent.a are the results of attribute lookup. I could still get the submodule by parent._a or sys.modules["parent.a"].

To answer my follow up question:

Even more confusing, if I remove the import statement in __init__.py, everything is back normal again (import parent.a; type(parent.a) is module). But why is this different? The a name in parent namespace is still a string.

This is when I import parent.a in the outside interactive Python interpreter, it first evaluates __init__.py, which does the overwriting of parent.a to a string. But the import hasn't finished yet, it goes on importing the submodule parent.a, and since we are still in the importing part, we don't do attribute lookups, and so we find the correct submodule. When all this is done, it binds the submodule to a of parent, thus overwriting the string that was overwriting the submodule, and making it all correct again.

This sounds very confusing, but remember (https://docs.python.org/3/reference/import.html#submodules):

When a submodule is loaded using any mechanism (e.g. importlib APIs, the import or import-from statements, or built-in __import__()) a binding is placed in the parent module’s namespace to the submodule object. For example, if package spam has a submodule foo, after importing spam.foo, spam will have an attribute foo which is bound to the submodule. Let’s say you have the following directory structure:

An import parent.a first runs all the module set-up code, and then binds the name.