Patch the inner source of python functions at runtime.
Patch the inner source of python functions at runtime.
A quick example, making a function that returns 1 instead return 9001:
>>> def sample(): ... return 1 >>> patchy.patch(sample, """\ ... @@ -1,2 +1,2 @@ ... def sample(): ... - return 1 ... + return 9001 ... """) >>> sample() 9001
Patchy works by replacing the code attribute of the function, leaving the function object itself the same. It’s thus more versatile than monkey patching, since if the function has been imported in multiple places they’ll also call the new behaviour.
pip install patchy
Python 3.5 to 3.8 supported.
If you’re monkey-patching an external library to add or fix some functionality, you will probably forget to check the monkey patch when you upgrade it. By using a patch against its source code, you can specify some context that you expect to remain the same in the function that will be checked before the source is applied.
I found this with some small but important patches to Django for a project. Since it takes a lot of energy to maintain a fork, writing monkey patches was the chosen quick solution, but then writing actual patches would be better.
The patches are applied with the standard patch commandline utility.
There are of course a lot of reasons against:
- It’s (relatively) slow (since it writes the source to disk and calls the patch command)
- If you have a patch file, why not just fork the library and apply it?
- At least with monkey-patching you know what end up with, rather than having the changes being done at runtime to source that may have changed.
All are valid arguments. However once in a while this might be the right solution.
The standard library function inspect.getsource() is used to retrieve the source code of the function, the patch is applied with the commandline utility patch, the code is recompiled, and the function’s code object is replaced the new one. Because nothing tends to poke around at code objects apart from dodgy hacks like this, you don’t need to worry about chasing any references that may exist to the function, unlike mock.patch.
A little special treatment is given to instancemethod, classmethod, and staticmethod objects to make sure the underlying function is what gets patched and that you don’t have to worry about the details.
Apply the patch patch_text to the source of function func. func may be either a function, or a string providing the dotted path to import a function.
If the patch is invalid, for example the context lines don’t match, ValueError will be raised, with a message that includes all the output from the patch utility.
Note that patch_text will be textwrap.dedent()’ed, but leading whitespace will not be removed. Therefore the correct way to include the patch is with a triple-quoted string with a backslash - """\ - which starts the string and avoids including the first newline. A final newline is not required and will be automatically added if not present.
import patchy def sample(): return 1 patchy.patch(sample, """\ @@ -2,2 +2,2 @@ - return 1 + return 2""") print(sample()) # prints 2
An alias for patch, so you can meme it up by calling patchy.mc_patchface().
Unapply the patch patch_text from the source of function func. This is the reverse of patch()ing it, and calls patch --reverse.
The same error and formatting rules apply as in patch().
import patchy def sample(): return 2 patchy.unpatch(sample, """\ @@ -2,2 +2,2 @@ - return 1 + return 2""") print(sample()) # prints 1
Takes the same arguments as patch. Usable as a context manager or function decorator to wrap code with a call to patch before and unpatch after.
Context manager example:
def sample(): return 1234 patch_text = """\ @@ -1,2 +1,2 @@ def sample(): - return 1234 + return 5678 """ with patchy.temp_patch(sample, patch_text): print(sample()) # prints 5678
Decorator example, using the same sample and patch_text:
@patchy.temp_patch(sample, patch_text) def my_func(): return sample() == 5678 print(my_func()) # prints True
replace(func, expected_source, new_source)
Check that function or dotted path to function func has an AST matching expected_source`, then replace its inner code object with source compiled from new_source. If the AST check fails, ValueError will be raised with current/expected source code in the message. In the author’s opinion it’s preferable to call patch() so your call makes it clear to see what is being changed about func, but using replace() is simpler as you don’t have to make a patch and there is no subprocess call to the patch utility.
Note both expected_source and new_source will be textwrap.dedent()’ed, so the best way to include their source is with a triple quoted string with a backslash escape on the first line, as per the example below.
If you want, you can pass expected_source=None to avoid the guard against your target changing, but this is highly unrecommended as it means if the original function changes, the call to replace() will continue to silently succeed.
import patchy def sample(): return 1 patchy.replace( sample, """\ def sample(): return 1 """, """\ def sample(): return 42 """ ) print(sample()) # prints 42
How to Create a Patch
Save the source of the function of interest (and nothing else) in a .py file, e.g. before.py:
def foo(): print("Change me")
Make sure you dedent it so there is no whitespace before the def, i.e. d is the first character in the file. For example if you wanted to patch the bar() method below:
class Foo(): def bar(self, x): return x * 2
…you would put just the method in a file like so:
def bar(self, x): return x * 2
However we’ll continue with the first example before.py since it’s simpler.
Copy that .py file, to e.g. after.py, and make the changes you want, such as:
def foo(): print("Changed")
Run diff, e.g. diff -u before.py after.py. You will get output like:
diff --git a/Users/chainz/tmp/before.py b/Users/chainz/tmp/after.py index e6b32c6..31fe8d9 100644 --- a/Users/chainz/tmp/before.py +++ b/Users/chainz/tmp/after.py @@ -1,2 +1,2 @@ def foo(): - print("Change me") + print("Changed")
The filenames are not necessary for patchy to work. Take only from the first @@ line onwards into the multiline string you pass to patchy.patch():
patchy.patch(foo, """\ @@ -1,2 +1,2 @@ def foo(): - print("Change me") + print("Changed") """)
- Drop Python 2 support, only Python 3.5-3.8 is supported now. Python 3.4 was dropped as it has reached its end of life.
- Converted setuptools metadata to configuration file. This meant removing the __version__ attribute from the package. If you want to inspect the installed version, use importlib.metadata.version("patchy") (docs / backport).
- Support cases where the function or class name is not in the freevars. This allows patching __init__ of a class, or patching a recursive module-level function.
- Support patching objects provided by dotted path string.
- Added new function patchy.replace(), that can be used to directly assign new source code to a function, without having to make a patch.
- Support freevars and methods that call mangled methods.
- Fixed setup.py to not fix the version of six required.
- Fixed install instruction in README.
- Added patchy.mc_patchface() as an alias for patchy.patch() because memes.
- Remove dependency on pylru by using a simpler caching strategy
- Pirate mascot!
- Patching caches the patched and unpatched versions, so unpatching and repeat patching/unpatching are both faster
- Patching doesn’t attach an attribute to the function object any more
- Fixed code compilation to use the __future__ flags from the function that is being patched
- Added unpatch method
- Added temp_patch context manager/decorator
- First release on PyPI, featuring patch function.
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size patchy-2.1.0-py3-none-any.whl (9.9 kB)||File type Wheel||Python version py3||Upload date||Hashes View|
|Filename, size patchy-2.1.0.tar.gz (152.9 kB)||File type Source||Python version None||Upload date||Hashes View|