Skip to main content

Type-checked function calls at runtime

Project description


CI (License MIT 1.0)


Python library providing type-checked function calls at runtime

Table of contents


Let's start with a toy example:

def times_two(value):
    return 2 * value

value = 3
result = times_two(value)
print(f'{value} * 2 == {result}')

This is fine, it outputs output: 3 * 2 = 6. But what if value accidentally is '3' instead of 3? The output will become output: 3 * 2 = 33, which might not be desired.

So you add something like

if not isinstance(value, int):
    raise TypeError(...)

to times_two. This will raise an TypeError instead, which is better. But you still only recognize the mistake when actually running the code. Catching it earlier in the development process might be better. Luckily Python allows to opt-in for static typing by offering type annotations. So you add them and mypy (or your IDE) will tell you about the problem early.

def times_two(value: int) -> int:
    return 2 * value

value = '3'
result = times_two(value) # error: Argument 1 to "times_two"
                          # has incompatible type "str"; expected "int"
print(f'{value} * 2 == {result}')

But you may get into a situation in which there is no useful static type information, because of values:

  • coming from external non-typed functions (so actually they are of type Any)
  • were produced by a (rogue) function that returns different types depending on some internal decision (Union[T, V])
  • being provided as a Dict[str, Any]
  • etc.
def times_two(value: int) -> int:
    return 2 * value

def get_value() -> Any:
    return '3'

value = get_value()
result = times_two(value)
print(f'{value} * 2 == {result}')

At least with the appropriate settings, mypy should dutifully complain, and now you're left with two options:

  • Drop type-checking (for example by adding # type: ignore to the end of the result = times_two(value) line): This however catapults you back into the insane world where 2 * 3 == 33.
  • You manually add type checks before the call (or inside of times_two) like if not isinstance(value, int):: This of course does not provide static type checking (because of the dynamic nature of value), but at least guarantees sane runtime behavior.

But the process of writing that boilerplate validation code can become quite cumbersome if you have multiple parameters/functions to check. Also it is not very DRY since you already have the needed type information in our function signature and you just duplicated it in the check condition.

This is where undictify comes into play. Simply decorate your times_two function with @type_checked_call():

from undictify import type_checked_call

def times_two(value: int) -> int:
    return 2 * value

And the arguments of times_two will be type-checked with every call at runtime automatically. A TypeError will be raised if needed.

This concept of runtime type-checks of function calls derived from static type annotations is quite simple, however it is very powerful and brings some highly convenient consequences with it.

Use case: JSON deserialization

Imagine your application receives a JSON string representing an entity you need to handle:

tobias_json = '''
        "id": 1,
        "name": "Tobias",
        "heart": {
            "weight_in_kg": 0.31,
            "pulse_at_rest": 52
        "friend_ids": [2, 3, 4, 5]

tobias = json.loads(tobias_json)

Now you start to work with it. Somewhere deep in your business logic you have:

name_length = len(tobias['name'])

But that's only fine if the original JSON string was well-behaved. If it had "name": 4, in it, you would get:

    name_length = len(tobias['name'])
TypeError: object of type 'int' has no len()

at runtime, which is not nice. So you start to manually add type checking:

if isinstance(tobias['name'], str):
    name_length = len(tobias['name'])
    # todo: handle the situation somehow

You quickly realize that you need to separate concerns better, in that case the business logic and the input data validation.

So you start to do all checks directly after receiving the data:

tobias = json.loads(...
if isinstance(tobias['id'], int):
if isinstance(tobias['name'], str):
if isinstance(... # *yawn*

and then transfer it into a type-safe class instance:

class Heart:
    weight_in_kg: float
    pulse_at_rest: int

class Human:
    id: int
    name: str
    nick: Optional[str]
    heart: Heart
    friend_ids: List[int]

Having the safety provided by the static type annotations (and probably checking your code with mypy) is a great because of all the:

  • bugs that don't make it into PROD
  • manual type checks (and matching unit tests) that you don't have to write
  • help your IDE can now offer
  • better understanding people get when reading your code
  • easier and more confident refactorings

But again, writing all that boilerplate code for data validation is tedious (and not DRY).

So you decide to use a library that does JSON schema validation for you. But now you have to manually adjust the schema every time your entity structure changes, which still is not DRY, and thus also brings with it all the typical possibilities to make mistakes.

Undictify can help here too! Annotate the classes @type_checked_constructor and their constructors will be wrapped in type-checked calls.

class Heart:
class Human:

(They do not need to be dataclasses. Deriving from NamedTuple works too.)

Undictify will type-check the construction of objects of type Heart and Human automatically. (This works for normal classes with a manually written __init__ function too. You just need to provide the type annotations to its parameters.) So you can use the usual dictionary unpacking syntax, to safely convert your untyped dictionary (i.e., Dict[str, Any]) resulting from the JSON string into your statically typed class:

tobias = Human(**json.loads(tobias_json))

(Btw this application is the origin of the name of this library.)

It throws exceptions with meaningful details in their associated values in case of errors like:

  • missing a field
  • a field having the wrong type
  • etc.

It also supports optional values being omitted instead of being None explicitly (as shown in the example with the nick field).


Sometimes, e.g., in case of unpacking a dictionary resulting from a JSON string, you might want to just skip the fields in the dictionary that your function / constructor does not take as a parameter. For these cases undictify provides @type_checked_call(skip=True).

It also supports valid type conversions via @type_checked_call(convert=True), which might for example come in handy when processing the arguments of an HTTP request you receive for example in a get handler of a flask_restful.Resource class:

def target_function(some_int: int, some_str: str)

class WebController(Resource):
    def get(self) -> Any:
        # request.args is something like {"some_int": "4", "some_str": "hi"}
        result = target_function(**flask.request.args)

The values in the MultiDict request.args are all strings, but the logic behind @type_checked_call(convert=True) tries to convert them into the desired target types with reasonable exceptions in case the conversion is not possible.

This way a request to http://.../foo?some_int=4&some_str=hi would be handled normally, but http://.../foo?some_int=four&some_str=hi would raise an appropriate TypeError.

Additional flexibility is offered for cases in which you would like to not type-check all calls of a specific function / class constructor, but only some. You can use type_checked_call() at call site instead of adding the annotation for those:

from undictify import type_checked_call

def times_two(value: int) -> int:
    return 2 * value

value: Any = '3'
resutl = type_checked_call()(times_two)(value)

And last but not least, custom converters for specified parameters are also supported:

import json
from dataclasses import dataclass
from datetime import datetime

from undictify import type_checked_constructor, optional_converter

def parse_timestamp(datetime_repr: str) -> datetime:
    return datetime.strptime(datetime_repr, '%Y-%m-%dT%H:%M:%SZ')

@type_checked_constructor(converters={'some_timestamp': optional_converter(parse_timestamp)})
class Foo:
    some_timestamp: datetime

json_repr = '{"some_timestamp": "2019-06-28T07:20:34Z"}'
my_foo = Foo(**json.loads(json_repr))

In case the converter should be applied even if the source type already matches the destination type, use mandatory_converter instead of optional_converter.

Requirements and Installation

You need Python 3.7 or higher.

python3 -m pip install undictify

Or, if you like to use latest version from this repository:

git clone
cd undictify
python3 -m pip install .


Distributed under the MIT License. (See accompanying file LICENSE or at

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

undictify-0.11.3.tar.gz (23.4 kB view hashes)

Uploaded source

Built Distribution

undictify-0.11.3-py3-none-any.whl (19.3 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page