'pyoverload' overloads the functions by simply using typehints and adding decorator '@overload'.
Project description
pyoverload
Introduction
pyoverload
is a package affiliated to project PyCAMIA
. It is a powerful overloading tools to provide easy overload for python v3.6+
. pyoverload
provide multiple usages. The simplest one, however, can be easily implemented as follows.
>>> from pyoverload import overload
>>> @overload
... def func(x: int):
... print("func1", x)
...
>>> @overload
... def func(x: str):
... print("func2", x)
...
>>> func(1)
func1 1
>>> func("1")
func2 1
pyoverload
has all of following appealing features:
- Support of
Jedi
auto-completion by keyword decorator@overload
. This means all main-stream python IDE can hint you the overloaded functions you have defined. - Multiple usages that are user friendly for all kinds of users, including
C/Java
language system users and those who are used tosingledispatch
based overload. Also, easy collector of ordinary python functions is also provided. - Support of all kinds of functions, including functions, methods, class methods and static methods. One simple implementation for all.
- String types supported. This means that one can use
"numpy.ndarray"
to mark a numpy array without importing the whole package. - Sufficient built-in types are provided for easy representations such as
List[Int]
,Dict@{str: int}
orList<<int>>[10]
. - Available usage listing when no overload function matches the input arguments.
- Type constraint for an ordinary function using
@params
decorator.
Installation
This package can be installed by pip install pyoverload
or moving the source code to the directory of python libraries (the source code can be downloaded on github or PyPI).
pip install pyoverload
Usages
Usage 1: Decorator Fashion
One can use @overload
before the functions with the same function name to build an overloaded function. When the function is called, the inputs will be handed out to the suitable implementation.
The types of the input arguments are specified by the typehints available in python3.6+
. All known types can be added after the colon. For package classes like np.ndarray
, please use a string to represent it. For more types, one can use types from package types
or pyoverload.typehint
.
For usage of pyoverload.typehint
, please refer to section Typehints for more information.
All implementations of the overloaded function are referenced in the order of definition, but the implementation ends with __default__
or __0__
will be used when no usage is available. Note that there are four underlines for this notation, two on each side.
>>> from pyoverload import *
>>> import numpy as np
>>> @overload
... def func__default__(x):
... print("func1", x)
...
>>> @overload
... def func(x: int):
... print("func2", x)
...
>>> @overload
... def func(x: str):
... print("func3", x)
...
>>> @overload
... def func(x: List<<Int>>[4]):
... print("func4", x)
...
>>> @overload
... def func(x: 'np.ndarray'):
... print("func5", x)
...
>>> func(1)
func2 1
>>> func("1")
func3 1
>>> func([1,2,3,4])
func4 [1, 2, 3, 4]
>>> func(np.array([1,2,3,4]))
func5 [1 2 3 4]
>>> func(1.)
func1 1.0
Note that the auto-completion by Jedi
can only work for this usage.
Usage 2: Registering Fashion
After using @overload
decorator, apart from using @overload
to decorate functions with the same name, one can also use the decorator with the function name @{fill in the function name}
to decorate other functions with relevant names like func1
, func_str
, first_func
for function func
. However, omitting sign _
is recommended for these functions.
In this fashion, the default function is the one decorated with @overload
though it can still be changed by adding __default__
or __0__
tags in the decorated function names. All typehints are the same as the first usage.
The following example realized the first three functions in the usage 1 example in a reimplementation.
>>> from pyoverload import overload
>>> @overload
... def func(x):
... print("func1", x)
...
>>> @func
... def func2(x: int):
... print("func2", x)
...
>>> @func
... def _(x: str):
... print("func3", x)
...
Note that usage 1 and usage 2 can be used together though you may need to specify the default function manually like in usage 1 if needed. The last example is rewrote in a combined style.
>>> from pyoverload import overload
>>> @overload
... def func__default__(x):
... print("func1", x)
...
>>> @overload
... def func(x: int):
... print("func2", x)
...
>>> @func
... def _(x: str):
... print("func3", x)
...
Usage 3: Collector Fashion
The last possible usage can not be used along with the previous two, or at least this is not tested by the developer and is not recommended. Another decorator @override
is used as a collector.
In this usage, all functions should have different names and all functions with typehints should add decorator @params
to activate the typehint regularization. Collector @override
takes additional function list as the arguments indicating these functions should be packed into a single function.
Note that the last function in the function list is the default function.
>>> from pyoverload import override, params
>>> @params
... def func_all(x):
... print("func1", x)
...
>>> @params
... def func_int(x: int):
... print("func2", x)
...
>>> @params
... def func_str(x: str):
... print("func3", x)
...
>>> @override(func_int, func_str, func_all)
... def func(): pass
...
Theoretically, decorator @override
can also be used in usages 1&2, but this is not recommended either.
Implementation List
When an overloaded function receives arguments that are not suitable for all implementations, the error information will tell you which ones are correct.
>>> from pyoverload import overload
>>> @overload
... def func(x: int):
... print("func1", x)
...
>>> @overload
... def func(x: str):
... print("func2", x)
...
>>> func(1.)
Traceback (most recent call last):
[...omitted...]
NameError: No func() matches arguments 1.0. All available usages are:
func(x:int)
func(x:str)
This function is available for all two Implementations but none of them takes 1.
.
Typehints
Decorator @params
enables functions to reject inputs with wrong types by raising TypeHintError
. One can use it directly to decorate functions with python typehints or one can add some arguments to it. In the following example, we apply the condition that a
is a function, b
is an integer, k
is a series of integers while function test_func
needs to return an iterable type of real numbers with length 2
.
>>> from pyoverload import *
>>> @params(Func, Int, +Int, __return__ = Real[2])
... def test_func(a, b=2, *k):
... print(a, b, k)
... return k
...
>>> test_func(lambda x: 1, 3, 4, 5)
<function <lambda> at 0x7fbdb2027f70> 3 (4, 5)
(4, 5)
The basic types in pyoverload.typehint
are Bool
, Int
, Float
, Str
, Set
, List
, Tuple
, Dict
, Callable
, Function
, Method
, Lambda
, Functional
, Real
, Null
, Sequence
, Array
, Iterable
, Scalar
, IntScalar
, FloatScalar
. Among which,
callable
contains all callable objects including callable classes whileFunc
consists all kinds of actual functions.Function
, however, only refer to explicitly defined functions, andMethod
andLambda
refer to the class methods and anonymous functions respectively. These three types are all contained in typeFunctional
.Real
is apyoverload.typehint.Type
whilereal
, a list[int, float]
, is not.null
is the type of elementNone
whileNull
is thepyoverload.typehint.Type
form of it.sequence
is[tuple, list, set]
, whileSequence
is thepyoverload.typehint.Type
form.Array
is the type of package based tensors. Only the tensors ofnumpy
,torch
,tensorflow
andbatorch
are currently supported. UseArray.Numpy
,Array.Torch
,Array.Tensorflow
andArray.Batorch
to identify specific packages.Iterable
includesArray
,Sequence
anddict
.- Three types of
Scalar
s support the array variables.
All these types are subclasses of type
and instances of pyoverload.typehint.Type
which will be abbreviated as Type
in the following introduction.
One can use Type(int)
to convert a python type like int
to a Type
or use Type(int, float)
to combine multiple existing types. Recurrence is also acceptable, Type(Type(int, float), Type(str))
is equivalent to Type(int, float, str)
. It will be better if a keyword name
is also assigned.
For a Type
, List
for example, we can do the following operations. Except the first four, these usages are designed for iterable types only.
-
+List
: This indicates that this is an extendable argument, which means it decorates arguments after*
. It was used to specify*args
arguments in@params
but currently deprecated though adding+
would not lead to failures. -
~List
: This invert the typehint, meaning that all non-list types. -
List|Tuple
: This is equivalent toType(List, Tuple)
indicating list or tuples. -
Type(A)&Callable
: The and operator. Note that both indicator should beType
s.>>> from pyoverload import * >>> class A: pass ... >>> class B(A): pass ... >>> class C(A): ... def __call__(self): pass ... >>> isoftype(B(), Type(A)&Callable) False >>> isoftype(C(), Type(A)&Callable) True
-
List[5]
: This indicates the length or shape of the variable. For other types such asInt[5]
, this indicates an iterable of length 5 with integer elements. Note thatArray[3, 4]
represents an array with shape3 x 4
, but this representation is not available for nested lists. UseList[List[4]][3]
orList<<List[4]>>[3]
instead. -
List@[int, str]
: This only works for ordered sequenceList
andTuple
. It indicates that the list has 2 elements, and the first one is an integer while the second is a string. Note that this is not equivalent toList[int, str]
which represents a list with elements either integer or string. -
List@int
orDict@{str: int}
: For all Iterables,A@B
indicates typeA
with all elements of typeB
.Dict@{str:int}
represents a dictionary with string keys and integer values. -
List[int]
:=List@int
. A list with integer elements. Note that one can useList[Int|float]
to directly identify a list of elements that are eitherint
orfloat
.List[int, float]
andDict[str: int, int:str]
are also valid for multiple element types or key-value pairs. Using list in the blankets will return to the effect of@
operator:List[[int, float]]=List@[int, float]
. -
List<<int>>[10]
:=(List@int)[10]=List[int][10]
The first one of the two types should be aType
. The representation refers to an integer list of length 10 which is equivalent toList[10]@int
. The length or shape is not specified if a pair of empty blankets is given. UseList<<Type(int, Float)>>[]
to represent multiple candidates andDict<<(str,int)>>[]=Dict[str:int]
to represent a dictionary. -
len(List[10, 20])
: Functionlen
returns the length of the array.200
should be the result for the given example.
Suggestions
- The speed of type check for
pyoverload.typehint.Type
is not very fast, hence please try your best to use builtin types, types from moduletypes
or list of types to do the typehint. - The
overload
takes extra time for delivering the arguments, hence using it for functions require fast speed is not recommended. - Do use
@params
for functions not overloaded but needs typehint constraints instead of using@overload
without actually has multiple implementations. This is because@params
is way faster.
Acknowledgment
@Yuncheng Zhou: Developer
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for pyoverload-1.1.27-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2c6f75f3232dcc8ee77cfe41c1e0e0de2a4c6f20f1a52a9d5270c570d082c4d7 |
|
MD5 | c2530bbe2da5c3bd95cf772102f09d3b |
|
BLAKE2b-256 | 0c8a544e7ad52a8f6e1f5b824a549d660c5b58c390fd34476aed5bd45e05b65d |