The huge number of lambdas had a tendency to muddle the first example; somebody should tell this gentleman about the `operator` module. Other than that, looks interesting! The decorator syntax may be a more apparent way to handle assertions about arguments in situations where numerous (or repetitive ones) are required. Hooks can be similarly handy.
> The huge number of lambdas had a tendency to muddle the first example; somebody should tell this gentleman about the `operator` module.
The lambdas are how you specify the constraints. How does operator module help here?
@ensure_args(a=lambda x: x > 10 and x < 100)
def foo(a):
pass
Here, the constraint is `a` should be greater than 10 and less than 100. Where does operator module comes in here?
And the first example is using multiple lambdas to show the multiple use cases:
1. `a=lambda x: x > 10` is for the positional argument `a`.
2. `b=r'^?-\d+(\.\d+)$'` is to show if you need the arg to match a regex, you can directly pass it.
3. c=lambda x: x < 5) # `c` will be picked from `kwargs`.
And here, the comment clarifies that c isn't there in the positional args. But since arguments are looked for in both kwargs and positional args, this constraint will work.
Won't that defeat the whole point? Contract based programming means your function assumes certain properties and acts accordingly. If you turn off the contract enforcers, your function will still be acting under the assumptions that the input meets the contract and will err.
I think the key word here is "assumption." zope.interface doesn't enforce its contracts by default; one must explicitly invoke verifyObject() to verify the invariants and interfaces of a given object. But, for speed, verifyObject() is usually omitted in interface-heavy applications. (I admit that my personal projects always verifyObject(), but only for rigorous plugin validation.)
The waters are murkier here. Apart from checks, you also have
@transform_args(a=lambda x: x*x)
def foo(a):
pass
This isn't validation and can't be turned off. The idea behind the project was the function should concentrate on logic and validations should happen declaratively. If the `verifyObject` sort of call comes inside the function, it impacts the declarative goal.
The current validators throw Exception by default. They can be optionally take an `error_handler` to which they pass the errors. Bringing in inside the function will mean manually accumulating the errors from various validators.
Also, these validators are mostly for conditions which must be met. Say you are writing a sqrt function, then the pre-condition is the number shouldn't be negative and that isn't something that can be turned off.
I don't know how zope.interface works but this isn't supposed to be optional validation. It is basically about taking the manual checks from inside the function to above the function with convenient declarative syntax.
I liked contracts in Racket, so ported some of it to Python. These checks can very well be done inside the called function but decorators enable a more declarative style.
Very nice. I did something very similar for an RPC server I wrote, but it's less general and relies on the ORM (SQLAlchemy). Essentially, it turns database id's into the mapped objects, checks the objects against the constraints declared as arguments to the decorator (much like this module), and then passes the mapped objects in to the original function in place of the id. It's proven to be a very useful pattern, but I'll probably use this library instead for future projects.
There also are a few modules out there that implement design-by-contract's pre-/post-conditions for class instances, for example http://code.activestate.com/recipes/436834/
