deprecate-kwargs

A Tool for Deprecating (Keyword) Arguments for Backward Compatibility for Python Functions.

A decorator is implemented to deprecate old kwargs in a function, with signature and docstring modified accordingly. Instead of replacing the old kwargs with new ones, this decorator makes old and new kwargs both available, with warnings raised when old kwargs are passed.

• deprecate-kwargs
  • Installation
  • Usage Example
  • Benefits

Installation

Run

python -m pip install deprecate-kwargs

or install the latest version in GitHub using

python -m pip install git+https://github.com/DeepPSP/deprecate-kwargs.git

Usage Example

>>> from deprecate_kwargs import deprecate_kwargs
>>> @deprecate_kwargs([["new_arg_1", "old_arg_1"], ["new_arg_2", "old_arg_2"], ["new_kw", "old_kw"]])
>>> def some_func(old_arg_1: int, old_arg_2: int, old_kw: int = 3):
>>>     return (old_arg_1 + old_arg_2) * old_kw
>>> some_func.__signature__
<Signature (new_arg_1: int, new_arg_2: int, new_kw: int = 3)>
>>> some_func(10, 20, 3)
90
>>> some_func(new_arg_1=10, new_arg_2=20, new_kw=3)
90
>>> some_func(10, old_arg_2=20, old_kw=3)
PendingDeprecationWarning: (keyword) argument "old_arg_2" is deprecated, use "new_arg_2" instead
PendingDeprecationWarning: (keyword) argument "old_kw" is deprecated, use "new_kw" instead
90

Benefits

deprecate_kwargs is quite useful when one wants to change the name of an argument (or keyword argument) of some function, while keeping old codes using this function still working, hence is beneficial for backward compatibility. For example, say in version 0.1 of some package, there's a function

def some_deep_learning_model_trainer(learning_rate, ...):
    ...

And in version 0.2, someone wants to change learning_rate to lr. If it was replaced directly via

def some_deep_learning_model_trainer(lr, ...):
    ...

then old codes using this function bycalling some_deep_learning_model_trainer(learning_rate=1e-3, ...) would break. However, if the replacement is done using

@deprecate_kwargs([["lr", "learning_rate"]])
def some_deep_learning_model_trainer(learning_rate, ...):
    ...

then one can call this function using some_deep_learning_model_trainer(lr=1e-3), as well as some_deep_learning_model_trainer(learning_rate=1e-3, ...) only with a warning raised. In this way, old codes are rescued from breaking.