typedload 2.41

Load and dump data from json-like format into typed data structures

typedload

Load and dump json-like data into typed data structures in Python3, enforcing a schema on the data.

This module provides an API to load dictionaries and lists (usually loaded from json) into Python's NamedTuples, dataclass, sets, enums, and various other typed data structures; respecting all the type-hints and performing type checks or casts when needed.

It can also dump from typed data structures to json-like dictionaries and lists.

It is very useful for projects that use Mypy and deal with untyped data like json, because it guarantees that the data will follow the specified schema.

It is released with a GPLv3 license but it is possible to ask for LGPLv3.

Example

For example this dictionary, loaded from a json:

data = {
    'users': [
        {
            'username': 'salvo',
            'shell': 'bash',
            'sessions': ['pts/4', 'tty7', 'pts/6']
        },
        {
            'username': 'lop'
        }
    ],
}

Can be treated more easily if loaded into this type:

@dataclasses.dataclass
class User:
    username: str
    shell: str = 'bash'
    sessions: list[str] = dataclasses.field(default_factory=list)

class Logins(NamedTuple):
    users: list[User]

And the data can be loaded into the structure with this:

t_data = typedload.load(data, Logins)

And then converted back:

data = typedload.dump(t_data)

Supported types

Since this is not magic, not all types are supported.

The following things are supported:

• Basic python types (int, str, bool, float, NoneType)
• NamedTuple
• Enum
• Optional[SomeType]
• list[SomeType]
• dict[TypeA, TypeB]
• tuple[TypeA, TypeB, TypeC] and tuple[SomeType, ...]
• set[SomeType]
• TypeA | TypeB
• dataclass
• attr.s
• ForwardRef (Refer to the type in its own definition)
• Literal
• TypedDict
• datetime.date, datetime.time, datetime.datetime
• re.Pattern
• Path
• IPv4Address, IPv6Address
• typing.Any
• typing.NewType
• uuid.UUID
• typedload.moretypes

Unions

typedload works fine with untagged unions. However using Literal fields to tag them makes it much faster.

Using Mypy

Mypy and similar tools work without requiring any plugins.

# This is treated as Any, no checks done.
data = json.load(f)

# This is treated as dict[str, int]
# but there will be runtime errors if the data does not
# match the expected format
data = json.load(f)  # type: dict[str, int]

# This is treated as dict[str, int] and an exception is
# raised if the actual data is not dict[str, int]
data = typedload.load(json.load(f), dict[str, int])

So when using Mypy, it makes sense to make sure that the type is correct, rather than hoping the data will respect the format.

Extending

Type handlers can easily be added, and existing ones can be replaced, so the library is fully cusomizable and can work with any type.

Inheriting a base class is not required.

Install

• pip install typedload
• apt install python3-typedload
• Latest and greatest .deb file is in releases

Documentation

• Online documentation
• In the docs/ directory

The tests are hard to read but provide more in depth examples of the capabilities of this module.

Used by

As dependency, typedload is used by those entities. Feel free to add to the list.

• Lyft
• Several universities around the world (via Relational)
• People who love IRC (via localslackirc)
• No clue but it gets thousands of downloads per day according to pypi