In December 2013, Giampaolo asked if it was possble to do property-like things in Python modules. It was. This was the result. Apparently a lot of Python folks really liked this functionality, or functionality like it, because in 2017 there were two PEPs that were put forward to add deferred attribute lookup to modules: https://www.python.org/dev/peps/pep-0549/ https://www.python.org/dev/peps/pep-0562/

Ultimately the simpler of the two (PEP 562) won out, and if you are using Python 3.7 or later, you can use the PEP 562 deferred attributes. As such, unless there is something major, this project is mostly for historical purposes.

Description

This package intends to offer a mechanism to add properties to your modules. The primary use-case of this functionality is to allow for the deferred execution of expensive-to-compute module/package level globals without needing to explicitly call a function.

Generally speaking, this module has two APIs that offer this functionality. You can use either of them or both of them, and everything will work more or less as you expect it to, as long as you follow the rules.

Note that this package supports the use of basically any descriptor at the module-level, not just properties.

Usage

The first method of using the package supports basically any descriptor being used as a decorator in the standard way. We show the use of property for the sake of brevity:

@property
def module_property(mod):
    '''mod is the module that this property was defined in'''

import mprop; mprop.init()

Because remembering to put the import/init call at the bottom of a module can be annoying, we've got a special decorator that works just like the property object, but handles the mprop.init() call for you:

from mprop import mproperty

@mproperty
def module_property2(mod):
    '''I work exactly the same as the earlier module property, but you
    don't need to make a subsequent call to ``mprop.init()``'''

Regardless of which method you use, if the name of your module is mod those that import the module can access the properties as normal:

# example.py
from mprop import mproperty

@mproperty
def prop(mod):
    return "I was called!"

def fcn():
    # I can access the property via:
    print _pmodule.prop

# test.py
import example

# the below should print "I was called!"
print example.prop
# this should also print "I was called!"
example.fcn()

Referencing properties from within the module

After initialization, your code may want to reference the global properties. If you try to access the properties directly, you will get a NameError unless you locally aliased the value, the initialization has not completed, or unless someone else injected a value with that name into the globals.

If you would like to directly access properties in the module, from within the module, you must reference them relative to the newly generated module. This is available from within your functions defined in the module via _pmodule, which is the "property-enhanced module" that offers property access.

If you find it necessary to require access to the original module object (which doesn't support properties), you can access _module from the global namspace.

How it works

The short version: we replace the standard Python module instance in sys.modules during module import, which allows us to ensure that the module is replaced everywhere it is used. We perform some magic to ensure that everything available in the original module is available in the replacement module (the replacement module's __dict__ is the module's globals), and we post-process everything in the module's global namespace to pull out descriptors as necessary.

As of February 1, 2019, in order to offer proper @mproperty.setter/deleter support, we've started initializing the module after import via the system profiler. More specifically, any time we get a call for @mproperty, we see if we've installed a system profiler to clean up after this module. In CPython via the sys.setprofile() handler, we look for return calls from module creation. These auto-nest, and auto-remove themselves, so by the time your final mprop-using module is done initializing, our profilers are removed.

We used to inject a special __getattribute__ method that auto-initializes on first access, but it was being triggered in cases where I didn't expect.

Python magic that goes on

• Using sys.getframe() to pull calling frame objects to get module globals and subsequently the module object itself
• Replace sys.modules entries
• Replace an instance __dict__
• Assign descriptors during runtime by modifying the base class of the replacement module
• Use the system profiler to discover when a module has finished loading, to start initialization