How to reload a module's function in Python?

问题:

Following up on this question regarding reloading a module, how do I reload a specific function from a changed module?

pseudo-code:

from foo import bar

if foo.py has changed:
    reload bar

回答1:

What you want is possible, but requires reloading two things... first reload(foo), but then you also have to reload(baz) (assuming baz is the name of the module containing the from foo import bar statement).

As to why... When foo is first loaded, a foo object is created, containing a bar object. When you import bar into the baz module, it stores a reference to bar. When reload(foo) is called, the foo object is blanked, and the module re-executed. This means all foo references are still valid, but a new bar object has been created... so all references that have been imported somewhere are still references to the old bar object. By reloading baz, you cause it to reimport the new bar.

---

Alternately, you can just do import foo in your module, and always call foo.bar(). That way whenever you reload(foo), you'll get the newest bar reference.

回答2:

Hot reloading is not something you can do in Python reliably without blowing up your head. You literally cannot support reloading without writing code special ways, and trying to write and maintain code that supports reloading with any sanity requires extreme discipline and is too confusing to be worth the effort. Testing such code is no easy task either.

The solution is to completely restart the Python process when code has changed. It is possible to do this seamlessly, but how depends on your specific problem domain.

回答3:

As of today, the proper way of doing this is:

import sys, importlib
importlib.reload(sys.modules['foo'])
from foo import bar

Tested on python 2.7, 3.5, 3.6.

回答4:

While reloading of functions is not a feature of the reload function, it is still possible. I would not recommend doing it in production, but here is how it works: The function you want to replace is a object somewhere in memory, and your code might hold many references to it (and not to the name of the function). But what that function actually does when called is not saved in that object, but in another object that, in turn, is referenced by the function object, in its attribute __code__. So as long as you have a reference to the function, you can update its code:

Module mymod:

from __future__ import print_function
def foo():
    print("foo")

Other module / python session:

>>> import mymod
>>> mymod.foo()
foo
>>> old_foo_ref = mymod.foo
>>> # edit mymod.py to make function "foo" print "bar"
>>> reload(mymod)
<module 'mymod' from 'mymod.py'>
>>> old_foo_ref()   # old reference is running old code
foo
>>> mymod.foo()     # reloaded module has new code
bar
>>> old_foo_ref.__code__ = mymod.foo.__code__
>>> old_foo_ref()   # now the old function object is also running the new code
bar
>>> 

Now, in case you do not have a reference to the old function (i.e. a lambda passed into another function), you can try to get a hold of it with the gc module, by searching the list of all objects:

>>> def _apply(f, x):
...     return lambda: f(x)
... 
>>> tmp = _apply(lambda x: x+1, 1)  # now there is a lambda we don't have a reference to
>>> tmp()
2
>>> import gc
>>> lambdas = [obj for obj in gc.get_objects() if getattr(obj,'__name__','') == '<lambda>'] # get all lambdas
[<function <lambda> at 0x7f315bf02f50>, <function <lambda> at 0x7f315bf12050>]     
# i guess the first one is the one i passed in, and the second one is the one returned by _apply
# lets make sure:
>>> lambdas[0].__code__.co_argcount
1  # so this is the "lambda x: ..."
>>> lambdas[1].__code__.co_argcount
0  # and this is the "lambda: ..."
>>> lambdas[0].__code__ = (lambda x: x+2).__code__  # change lambda to return arg + 2
>>> tmp()
3  # 
>>> 

回答5:

You can't reload a method from a module but you can load the module again with a new name, say foo2 and say bar = foo2.bar to overwrite the current reference.

Note that if bar has any dependencies on other things in foo or any other side effects, you will get into trouble. So while it works, it only works for the most simple cases.

回答6:

You will have to use reload to reload the module, since you can't reload just the function:

>>> import sys
>>> reload(sys)
<module 'sys' (built-in)>
>>>

回答7:

If you are the author of foo.py, you can write:

with open('bar.py') as f: bar_code=compile(f.read(),'bar.py','exec')

def bar(a,b):
    exec(bar_code)

def reload_bar():
    global bar_code
    with open('bar.py') as f: bar_code=compile(f.read(),'bar.py','exec') 

Then, in your pseudocode, reload if bar.py has changed. This approach is especially nice when bar lives in the same module as the code that wants to hot-reload it rather than the OP's case where it lives in a different module.