resguard 0.12

No project description provided

resguard

This module provides function for parsing response data, based on dataclass defined schemas.

The user define arbitrary schema using dataclass. One dataclass can refer to others to represent nested structures.

>>> @dataclass
... class Foo:
...     pass

>>> @dataclass
... class Bar:
...     foo: Foo

While made with parsing json decoded data from REST responses in mind, the approach is pretty generic and may work for other use cases.

So suppose that you're in charging to do another API client.. if you started doing this once you know that you'll gonna work with JSON and that JSON become plain dicts and lists in python, it's easy to lose the track of these objects and start to spread KeyError and IndexError handlers all over the codebase.

It became usual to me to write representation of the response data as objects and instantiating these objects, and with objects I can have some type checking, mutch better than with dicts... and can track what the fields

But writing ad-hoc classes and parsers from dict -> myobject became boring too.. so I created this! Much more declarative and type checking friendly

So let's write an API to cat facts, we can find the docs here https://alexwohlbruck.github.io/cat-facts/docs/endpoints/facts.html

We're implementing the /facts/random endpoint. The documentation said that it will respond like this:

	{
		"_id": "591f9894d369931519ce358f",
		"__v": 0,
		"text": "A female cat will be pregnant for approximately 9 weeks - between 62 and 65 days from conception to delivery.",
		"updatedAt": "2018-01-04T01:10:54.673Z",
		"deleted": false,
		"source": "api",
		"used": false
	}

So is a list of facts, a fact can be defined like this

>>> from datetime import datetime
>>> @dataclass
... class Fact:
...     _id: str
...     __v: int
...     text: str
...     updatedAt: datetime
...     deleted: bool
...     source: str
...     used: bool
...     user: Optional[str]

To parse a respone you call parse_dc, where dc stands for dataclass. You call it with the dataclass and the response data:

>>> import requests as r
>>> url = "https://cat-fact.herokuapp.com"
>>> res = r.get(f"{url}/facts/random")
>>> parse_dc(Fact, res.json())
Traceback (most recent call last):
...
TypeError: Unknow field type for Fact. Expected one of (_id,_Fact__v,text,updatedAt,deleted,source,used,user)

You may notice that I put a user: Optional[str] on the Fact definition too. This is how you express optional fields, that may or may not be present on response. Missing optinal fields become None in dataclass

What happens here is that the documentation is outdated, there are a type field that was not expected in response. parse_dc raise a TypeError if anything goes out of rails. Let's see in response what we have in type field

>>> type_ = res.json()['type']
>>> type_, type(type_)
('cat', <class 'str'>)

This may happen again and again. Some APIs are freaking crazy, they may add some fields in some responses and another not. If you pass ,ignore_unknows=True) to parse_dc it will not raise type errors if an unexpected field arrives. If you want this behavior by default you can memoise parse_dc as

from functools import partial
parse_dc = partial(parse_dc, ignore_unknows=True)

So let's update our Fact definition

>>> @dataclass
... class Fact:
...     _id: str
...     __v: int
...     text: str
...     updatedAt: datetime
...     deleted: bool
...     source: str
...     used: bool
...     user: Optional[str]
...     type: str # <- we added this

And parse again

>>> parse_dc(Fact, res.json())
Traceback (most recent call last):
...
TypeError:  in dataclass Fact while trying to construct value from datetime...

Know it stopped in updatedAt: datetime.. Well, for a dataclass field F: T and value V, parse_dc will call T(V) at somepoint. So is important that construtors can work with data that came from the response, but also we don't want to lose typechecking. The solution here is to wrap the problematic object in a class and override its constructor. I override __new__ instead of __init__ here because I want to control the returned object not how it's initializated.

>>> class datetimestr(datetime):
...     def __new__(cls, datestr):
...         return datetime.strptime(datestr, "%Y-%m-%dT%H:%M:%S.%fZ")

>>> @dataclass
... class Fact:
...     _id: str
...     __v: int
...     updatedAt: datetimestr
...     createdAt: datetimestr
...     deleted: bool
...     source: str
...     used: bool
...     type: str
...     user: Optional[str]
...     text: str
...
>>> parse_dc(Fact, res.json())
Traceback (most recent call last):
...
TypeError: Unknow field status for Fact. ...

Oh no, another missing field. While I write this tutorial, we just got two outdated fields... this is why the library is called resguard. It's an abbreviation for response guard, it guard you from dealing to bad responses all over your code... Back to the tutorial, let't see what we have in status field:

>>> res.json()['status']
{'verified': True, 'sentCount': 1}

So it's a dictionary, let's create another dataclass to represent this and update our Facts dataclass to include status field

>>> @dataclass
... class Status:
...     verified: bool
...     sentCount: int

>>> @dataclass
... class Fact:
...     _id: str
...     __v: int
...     updatedAt: datetimestr
...     createdAt: datetimestr
...     deleted: bool
...     source: str
...     used: bool
...     type: str
...     user: str
...     text: str
...     status: Status # <- we added this

Here we go again

>>> parse_dc(Fact, res.json())
Fact(_id=..., _Fact__v=..., updatedAt=..., createdAt=..., deleted=..., source=..., used=..., type='cat', user=..., text=..., status=Status(verified=True, sentCount=1))

And it worked!

To avoid the class boilerplate is possible to use the @create_base decorator, it takes a function and replaces it by a class which is subclass of the sole argument

So you can replace this

>>> class datetimestr(datetime):
...     def __new__(cls, datestr):
...         return datetime.strptime(datestr, "%Y-%m-%dT%H:%M:%S.%fZ")

by this

>>> @create_base(datetime)
... def datetimestr(s):
...     return datetime.strptime(s, r"%d/%m/%Y")

Since datetimestr is a subtype of datetime it typechecks for datetime.

Now what if we want go to the oposite direction, given somejson, construct a dataclass. Well resguard can be invoked as curl something | python -m resguard fromjson and it will output a dataclass definition for that JSON.

The type inference is pretty simple, but it is already better than writing all that dataclasses by hand. Let's see it in action

>>> print(print_dc(fromjson("Root", '{"foo": "foo", "bar": { "bar": "bar" }}')))
@dataclass
class bar:
   bar: str
<BLANKLINE>
<BLANKLINE>
@dataclass
class Root:
   foo: str
   bar: bar
<BLANKLINE>

To use it from command line (much simpler)

curl -s https://cat-fact.herokuapp.com/facts/random | python -m resguard fromjson
@dataclass
class status:
   verified: bool
   sentCount: int


@dataclass
class Root:
   used: bool
   source: str
   type: str
   deleted: bool
   _id: str
   __v: int
   text: str
   updatedAt: str
   createdAt: str
   status: status
   user: str

That's it, check below for function docs

parse_dc(cls: resguard.Dataclass, data: dict, ignore_unknows=False) -> resguard.Dataclass

Given an arbitrary dataclass and a dict this function will recursively parse the data, checking data types against cls dataclass.

It raises TypeError if something goes wrong. It tries to improve the common errors by reraising then with better messages.

First declare some dataclasses that define the data that you want to parse.

>>> @dataclass
... class Foo:
...     name: Literal[0, 1]
...
>>> @dataclass
... class Bar:
...     l: List[int]
...     foo: Dict[str, int]
...     Foo: Foo
...     age: Optional[int] = None

Now suppose that you get this data from a network response. I'm expecting it to be plain json parsed to dicts, lists and so on, but no objects, just decoded json:

>>> data = {"foo": {"bar": 1}, "l": [], "Foo": {"name": 1}}

Hmm, it seems to match, lets try to parse this

>>> parse_dc(Bar, data)
Bar(l=[], foo={'bar': 1}, Foo=Foo(name=1), age=None)

You can see that it creates nested dataclasses too, cool. But this was easy, this was the happy path, what about the not so happy path. Let's change data, and see how parse_dc handle errors

>>> data["badkey"] = "bad things"
>>> parse_dc(Bar, data)
Traceback (most recent call last):                                           
...
TypeError: Unknow field badkey for Bar. Expected one of (l,foo,Foo,age)

Hmm... interesting... It knows that badkey is not in Bar dataclass definition and it shows what are the expected keys. Let's try another thing

>>> @dataclass
... class Foo:
...     foo: int
>>> parse_dc(Foo, {"foo": "an string"})
Traceback (most recent call last):
...
TypeError: in dataclass Foo, 'an string' is not int: invalid literal for int() with base 10: 'an string'

So I passed a bad (by little margin) value in "foo" key. It expects an int and received an string. The "invalid literal ..." part is from an error that raises when parce_dc tries to pass "an string" to int(), it handles the error and reraise as TypeError. The idea is that calle should catch TypeError.

Now if it uses the dataclass field type as constructor, this works

>>> @dataclass
... class Foo:
...     foo: str 
>>> parse_dc(Foo, {"foo": 1}).foo
'1'

It works because str(1) just .. works .. So think about str as the Any type from typing module. Almost anything can be encoded as string, so take care of yours, since they point to holes on type checking, but provide a nice generic system

create_base(base)

A function decorator. It replace the function by a class which call the decorated function in its new method, for example

>>> from datetime import datetime
>>> @create_base(datetime)
... def date_br(s):
...     return datetime.strptime(s, r"%d/%m/%Y")
>>> issubclass(date_br, datetime)
True
>>> date_br("01/01/2001")
datetime.datetime(2001, 1, 1, 0, 0)

unpack_union(union: Union[~T, Any, NoneType]) -> ~T

Takes an Unin and return another union with the same arguments as input, but with None and Any filtered

>>> unpack_union(Optional[str])
<class 'str'>

>>> unpack_union(List[str])
<class 'str'>

It respect concrete types

>>> unpack_union(int)
<class 'int'>

If the input is a literal, it returns itself. Literals are types and values at same time, like enums

>>> unpack_union(1)
1
>>> unpack_union([1,2])
[1, 2]

Dataclass(*args, **kwds)

Dataclass static type https://stackoverflow.com/a/55240861/652528