Phantom types for Python will help you make illegal states unrepresentable and avoid shotgun parsing by enabling you to practice "Parse, don't validate".
$ python3 -m pip install phantom-types
There are a few extras available that can be used to either enable a feature or install a compatible version of a third-party library.
Extra name | Feature |
---|---|
[dateutil] |
Installs python-dateutil. Required for parsing strings with TZAware and TZNaive . |
[phonenumbers] |
Installs phonenumbers. Required to use phantom.ext.phonenumbers . |
[pydantic] |
Installs pydantic. |
[hypothesis] |
Installs hypothesis. |
[all] |
Installs all of the above. |
$ python3 -m pip install phantom-types[all]
By introducing a phantom type we can define a pre-condition for a function argument.
from phantom import Phantom
from phantom.predicates.collection import contained
class Name(str, Phantom, predicate=contained({"Jane", "Joe"})): ...
def greet(name: Name):
print(f"Hello {name}!")
Now this will be a valid call.
greet(Name.parse("Jane"))
... and so will this.
joe = "Joe"
assert isinstance(joe, Name)
greet(joe)
But this will yield a static type checking error.
greet("bird")
To be clear, the reason the first example passes is not because the type checker somehow
magically knows about our predicate, but because we provided the type checker with proof
through the assert
. All the type checker cares about is that runtime cannot continue
executing past the assertion, unless the variable is a Name
. If we move the calls
around like in the example below, the type checker would give an error for the greet()
call.
joe = "Joe"
greet(joe)
assert isinstance(joe, Name)
By combining phantom types with a runtime type-checker like beartype or typeguard, we can achieve the same level of security as you'd gain from using contracts.
import datetime
from beartype import beartype
from phantom.datetime import TZAware
@beartype
def soon(dt: TZAware) -> TZAware:
return dt + datetime.timedelta(seconds=10)
The soon
function will now validate that both its argument and return value is
timezone aware, e.g. pre- and post conditions.
Phantom types are ready to use with pydantic and have integrated
support out-of-the-box. Subclasses of Phantom
work with both
pydantic's validation and its schema generation.
class Name(str, Phantom, predicate=contained({"Jane", "Joe"})):
@classmethod
def __schema__(cls) -> Schema:
return super().__schema__() | {
"description": "Either Jane or Joe",
"format": "custom-name",
}
class Person(BaseModel):
name: Name
created: TZAware
print(json.dumps(Person.schema(), indent=2))
The code above outputs the following JSONSchema.
{
"title": "Person",
"type": "object",
"properties": {
"name": {
"title": "Name",
"description": "Either Jane or Joe",
"format": "custom-name",
"type": "string"
},
"created": {
"title": "TZAware",
"description": "A date-time with timezone data.",
"type": "string",
"format": "date-time"
}
},
"required": ["name", "created"]
}
Install development requirements, preferably in a virtualenv:
$ python3 -m pip install .[all,test,type-check]
Run tests:
$ pytest
# or
$ make test
Run type checker:
$ mypy
Linters and formatters are set up with goose, after installing it you can run it as:
# run all checks
$ goose run --select=all
# or just a single hook
$ goose run mypy --select=all
In addition to static type checking, the project is set up with pytest-mypy-plugins to test that exposed mypy types work as expected, these checks will run together with the rest of the test suite, but you can single them out with the following command.
$ make test-typing