[blogpost] http://williamedwardscoder.tumblr.com/post/33185451698/obiwan-typescript-for-python
Obiwan is a Python type-checker. You place descriptive type constraints in your function declarations and obiwan can check them for you at runtime.
A function can look like:
def example(a: int, b: float) -> number:
return a/b
My ambition is that this Obiwan syntax is widely adopted and eventually Python static type checkers support it and IDEs can do auto-complete ala Typescript.
To enable obiwan, you just call it:
from obiwan import *; install_obiwan_runtime_check()
you are now running obiwan! Runtime execution will be slower, but annotated functions will be checked for parameter correctness!
All strings in your function annotations are ingored; you can place documentation in annotations without impacting obiwan.
The dictionary and list checking code is based upon a tried-and-tested JSON validator.
The integration with Python 3 function annotations is new and the function and duck type checking is new. Improvements and patches welcome!
You can also describe dictionary parameters and what their expected attributes are:
def example2(obj: {"a":int, "b": float}) -> {"ret": number}:
return {"ret": a/b}
Checking can support the checking of optional and noneable attributes:
def example3(obj: {"a":int, optional("b"): float}):
...
Checks can contain dictionary and other attributes too:
def example4(person: {"name":str, "phone": {"type":str, "number":str}}):
...
You can specify alternative constraint types using sets:
def example5(x: {int,float}):
...
In fact, number type is just a set of int and float. And noneable is just a way of saying {...,None}
Lists mean that the attribute must be an array where each element matches the constraint e.g.:
def example6(numbers: [int]):
...
Tuples must map to lists or tuples (no destructive iterators!) with the appropriate types in each slot:
def nearest_point_on_line(line:((int,int),(int,int)),pt:(int,int)) -> (int,int):
Within tuples you can use any to indicate that a type needs not be checked, and you can use ellipsis as the last element in the type-defintion tuple to indicate that additional parameters are allowed:
def decode_data(data: str) -> (str,any,int,...):
It aids readability to use variables to hold definitions e.g.:
Point = (int,int)
def nearest_point_on_line(line:(Point,Point),pt:Point) -> Point:
and:
api_add_user = {
"name": str,
"admin": bool,
}
def add_user(user: api_add_user) -> int:
...
Utility functions to load and dump JSON are provided. These support a new template parameter and validate the input/output matches the constraint e.g.:
json.loads(tainted,template=[api_add_user])
In Python 3 everything is an object, even int and None. So you can't generically say that an argument or attribute must be an object. You have to say what its attributes should be. This follows the same style as validating dictionaries, but uses the duck type and keyword arguments to define:
def example7(a: duck(name=str,get_name=function)):
...
This means that a must be something with a name attribute of type string, and a function attribute called get_name.
You can of course use classes to:
class Person:
def get_name(self):
...
def example8(person: Person):
...
You can say that a parameter is callable using function:
def example9(callback: function):
...
If you want, you can describe the parameters that the function should take:
def example10(callback: function(int,str)):
...
However, all the functions passed to example8 must now be properly annotated with a matching annotation.
The special type any can be used if you do not want to check the type:
def example11(callback: function(int,any,number)):
...
You can also specify that a function should support further arguments using ellipsis:
def example12(callback: function(int,any,...)):
...
This will ensure that all callbacks have at least two parameters, the first being an int.
You can provide your own complex custom constraint checkers by subclassing the ObiwanCheck class; look at obiwan.StringCheck for inspiration.