模块可以具有与对象相同的属性吗？

With python properties, I can make it such that

obj.y 

calls a function rather than just returning a value.

Is there a way to do this with modules? I have a case where I want

module.y 

to call a function, rather than just returning the value stored there.

Only instances of new-style classes can have properties. You can make Python believe such an instance is a module by stashing it in sys.modules[thename] = theinstance. So, for example, your m.py module file could be:

import sys

class _M(object):
    def __init__(self):
        self.c = 0
    def afunction(self):
        self.c += 1
        return self.c
    y = property(afunction)

sys.modules[__name__] = _M()

I would do this in order to properly inherit all the attributes of a module, and be correctly identified by isinstance()

import types

class MyModule(types.ModuleType):
    @property
    def y(self):
        return 5


>>> a=MyModule("test")
>>> a
<module 'test' (built-in)>
>>> a.y
5

And then you can insert this into sys.modules:

sys.modules[__name__] = MyModule(__name__)  # remember to instantiate the class

As PEP 562 has been implemented in Python >= 3.7, now we can do this

file: module.py

def __getattr__(name):
    if name == 'y':
        return 3
    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")

other = 4

usage:

>>> import module
>>> module.y
3
>>> module.other
4
>>> module.nosuch
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "module.py", line 4, in __getattr__
    raise AttributeError(f"module '{__name__}' has no attribute '{name}'")
AttributeError: module 'module' has no attribute 'nosuch'

Note that if you omit the raise AttributeError in the __getattr__ function, it means the function ends with return None, then the module.nosuch will get a value of None.

Based on John Lin’s answer:

def module_property(func):
    """Decorator to turn module functions into properties.
    Function names must be prefixed with an underscore."""
    module = sys.modules[func.__module__]

    def base_getattr(name):
        raise AttributeError(
            f"module '{module.__name__}' has no attribute '{name}'")

    old_getattr = getattr(module, '__getattr__', base_getattr)

    def new_getattr(name):
        if f'_{name}' == func.__name__:
            return func()
        else:
            return old_getattr(name)

    module.__getattr__ = new_getattr
    return func

Usage (note the leading underscore), in the_module.py:

@module_property
def _thing():
    return 'hello'

Then:

import the_module

print(the_module.thing)  # prints 'hello'

The leading underscore is necessary to differentiate the property-ized function from the original function. I couldn’t think of a way to reassign the identifier, since during the time of the decorator execution, it has not been assigned yet.

Note that IDEs won’t know that the property exists and will show red wavies.

A typical use case is: enriching a (huge) existing module with some (few) dynamic attributes – without turning all module stuff into a class layout. Unfortunately a most simple module class patch like sys.modules[__name__].__class__ = MyPropertyModule fails with TypeError: __class__ assignment: only for heap types. So module creation needs to be rewired.

This approach does it without Python import hooks, just by having some prolog on top of the module code:

# propertymodule.py
""" Module property example """

if '__orgmod__' not in globals():

    # constant prolog for having module properties / supports reload()

    print "PropertyModule stub execution", __name__
    import sys, types
    class PropertyModule(types.ModuleType):
        def __str__(self):
            return "<PropertyModule %r from %r>" % (self.__name__, self.__file__)
    modnew = PropertyModule(__name__, __doc__)
    modnew.__modclass__ = PropertyModule        
    modnew.__file__ = __file__
    modnew.__orgmod__ = sys.modules[__name__]
    sys.modules[__name__] = modnew
    exec sys._getframe().f_code in modnew.__dict__

else:

    # normal module code (usually vast) ..

    print "regular module execution"
    a = 7

    def get_dynval(module):
        return "property function returns %s in module %r" % (a * 4, module.__name__)    
    __modclass__.dynval = property(get_dynval)

Usage:

>>> import propertymodule
PropertyModule stub execution propertymodule
regular module execution
>>> propertymodule.dynval
"property function returns 28 in module 'propertymodule'"
>>> reload(propertymodule)   # AFTER EDITS
regular module execution
<module 'propertymodule' from 'propertymodule.pyc'>
>>> propertymodule.dynval
"property function returns 36 in module 'propertymodule'"

Note: Something like from propertymodule import dynval will produce a frozen copy of course – corresponding to dynval = someobject.dynval

A short answer: use proxy_tools

The package attempts to provide @module_property functionality.

It installs with

pip install proxy_tools

Using a slight modification of @Marein’s example, in the_module.py we put

from proxy_tools import module_property

@module_property
def thing():
    print(". ", end='')  # Prints ". " on each invocation
    return 'hello'

Now from another script, I can do

import the_module

print(the_module.thing)
# . hello

Unexpected behavior

This solution is not without caveats. Namely, the_module.thing is not a string! It is a proxy_tools.Proxy object whose special methods have been overridden so that it mimicks a string. Here are some basic tests which illustrate the point:

res = the_module.thing
# [No output!!! Evaluation doesn't occur yet.]

print(type(res))
# <class 'proxy_tools.Proxy'>

print(isinstance(res, str))
# False

print(res)
# . hello

print(res + " there")
# . hello there

print(isinstance(res + "", str))
# . True

print(res.split('e'))
# . ['h', 'llo']

Internally, the original function is stored to the_module.thing._Proxy__local:

print(res._Proxy__local)
# <function thing at 0x7f729c3bf680>

Further thoughts

Honestly, I’m baffled about why modules don’t have this functionality built in. I think the crux of the matter is that the_module is an instance of the types.ModuleType class. Setting a “module property” amounts to setting a property on an instance of this class, rather than on the types.ModuleType class itself. For more details, see this answer.

We can actually implement properties on types.ModuleType as follows, although the results are not great. We can’t directly modify built-in types, but we can curse them:

# python -m pip install forbiddenfruit
from forbiddenfruit import curse
from types import ModuleType
# curse has the same signature as setattr.
curse(ModuleType, "thing2", property(lambda module: f'hi from {module.__name__}'))

This gives us a property which exists over all modules. It’s a bit unwieldly, since we break the setting behavior across all modules:

import sys

print(sys.thing2)
# hi from sys

sys.thing2 = 5
# AttributeError: can't set attribute