serdataclasses
Serialize (serdataclasses.dump
) and deserialize (serdataclasses.load
) from/to strongly-typed, native Python data structures.
Motivation
No typing-focused serde library in Python satisfies me. Call me needy, but when I must translate loosely-typed data structures (like list
and dict
) to strongly-typed data structures (like NamedTuple
and dataclasses.dataclass
), I want the following capabilities:
- Effortless deserialization of unstructured Python types into structured, type-hinted Python types (
dataclasses.dataclass
,typing.NamedTuple
) - Effortless serialization of structured, type-hinted Python objects into unstructured Python types (eg, the reverse)
- Clear error messages when serde fails at runtime
- No inherited, non-standard types. dataclasses, NamedTuples, and other standard Python types are bread and butter
- Editor support: I like my autocompletion, so I jump through lots of hoops to make this library compatible with Jedi
- Handle optional properties with a domain-specific
serdataclasses.OptionalProperty
- Provide option to automatically convert primitive types, but avoid converting ambiguous types (
Union
,TypeVar
, etc). Handle ofOptional
andserdataclasses.OptionalProperty
- Require no 3rd party dependencies; Python 3.8+
Installation
# With pip
pip install serdataclasses
# With poetry
poetry add serdataclasses
Usage
See examples folder if you'd like to get your hands dirty. Otherwise, keep reading for a complete, real-world example.
Example: Libraries and Books
Assume that you receive a JSON
list of libraries containing each library's name and a list of each library's books.
[
{
"name": "Clark County Library",
"books": [
{
"title": "Hello, World!",
"author": "Susy Smith",
"year": 1929,
"tags": ["boring"]
},
{
"title": "The great showman",
"author": "Beth John"
},
{
"title": "My favorite pony",
"author": null
}
]
},
{
"name": "Only 1 book here",
"books": [
{
"title": "The great fun time",
"author": "Smitty",
"year": 1950,
"tags": ["swell"]
}
]
}
]
Now you want to ingest this document into Python. Your first step is probably to deserialize the JSON string (or file) into Python data structures.
import json
from pprint import pprint
with open("libraries.json", "r") as infile:
libraries_raw = json.load(infile)
pprint(libraries_raw)
print(type(libraries_raw))
print(type(libraries_raw[0]))
Assuming the JSON is read from a file called libraries.py
, the preceding script will print:
[{'books': [{'author': 'Susy Smith',
'tags': ['boring'],
'title': 'Hello, World!',
'year': 1929},
{'author': 'Beth John', 'title': 'The great showman'},
{'author': None, 'title': 'My favorite pony'}],
'name': 'Clark County Library'},
{'books': [{'author': 'Smitty',
'tags': ['swell'],
'title': 'The great fun time',
'year': 1950}],
'name': 'Only 1 book here'}]
<class 'list'>
<class 'dict'>
Some observations:
- Python's native
json
module deserializes the JSON string / document into Python's primitive (or primitive-like) types -
null
is translated to Python'sNone
- The first list element is a dictionary. So Python appears to have translated the JSON into a list of dictionaries.
- There is little inherent structure to the Python objects deserialized by the JSON module. By this, I mean that we have no way of knowing whether the dictionaries contain keys that we expect or are structured improperly. Should books also have an
"isbn"
field? Does code we write that useslibraries_raw
expect an"isbn"
field? What happens if there are missing tags?
The first 3 items are merely facts; serdataclasses
accepts these facts and builds on them. The 4th item in this list is THE problem that serdataclasses
is designed to solve. If we take the above Python dictionary and associate it with a Python variable named LIBRARIES
, we can define a strongly-typed Python container that serdataclasses
can use to ingest LIBRARIES
.
from dataclasses import dataclass, field
from pprint import pprint
from typing import List, Optional
import serdataclasses
from serdataclasses import OptionalProperty
[
{
"books": [
{
"author": "Susy Smith",
"tags": ["boring"],
"title": "Hello, World!",
"year": 1929,
},
{"author": "Beth John", "title": "The great showman"},
{"author": None, "title": "My favorite pony"},
],
"name": "Clark County Library",
},
{
"books": [
{
"author": "Smitty",
"tags": ["swell"],
"title": "The great fun time",
"year": 1950,
}
],
"name": "Only 1 book here",
},
]
@dataclass
class Book:
author: Optional[str]
title: str
year: OptionalProperty[int]
tags: List[str] = field(default_factory=list)
@dataclass
class Library:
books: List[Book]
name: str
LIBRARIES_LOADED = serdataclasses.load(LIBRARIES, List[Library])
print(LIBRARIES_LOADED[0].name)
print(LIBRARIES_LOADED[0].books[1].author)
pprint(serdataclasses.dump(LIBRARIES_LOADED))
Running the above script, we get the following output to the terminal:
[{'books': [{'author': 'Susy Smith',
'tags': ['boring'],
'title': 'Hello, World!',
'year': 1929},
{'author': 'Beth John', 'tags': [], 'title': 'The great showman'},
{'author': None, 'tags': [], 'title': 'My favorite pony'}],
'name': 'Clark County Library'},
{'books': [{'author': 'Smitty',
'tags': ['swell'],
'title': 'The great fun time',
'year': 1950}],
'name': 'Only 1 book here'}]
Notice how we have the following features:
- Data structures are loaded, recursively, without you needing to write anything more than a couple standard Python classes.
- For missing properties / dictionary keys (for example,
Book.tags
), we can set a default value in our dataclass using standard Python andserdataclasses
adds the default value to our structure - For missing properties without default values, serdataclasses intelligently omits them when re-serializing the result. There is also an option to
serdataclasses.load
that allows you to convert missing values toNone
and keep the keys in the output. For all other desired default values, just usedataclasses.field
; no need to re-invent the wheel!
What about additional validation, you may ask? Again, just use dataclasses! Assume that, for some reason, no book can possibly be published before 1930, and that a book published before 1930 invalidates all data. No problem, just use the standard method __post_init__
on the relevant dataclass!
from dataclasses import dataclass, field
from pprint import pprint
from typing import List, Optional
import serdataclasses
from serdataclasses import OptionalProperty
LIBRARIES = [
{
"books": [
{
"author": "Susy Smith",
"tags": ["boring"],
"title": "Hello, World!",
"year": 1929,
},
{"author": "Beth John", "title": "The great showman"},
{"author": None, "title": "My favorite pony"},
],
"name": "Clark County Library",
},
{
"books": [
{
"author": "Smitty",
"tags": ["swell"],
"title": "The great fun time",
"year": 1950,
}
],
"name": "Only 1 book here",
},
]
@dataclass
class Book:
author: Optional[str]
title: str
year: OptionalProperty[int]
tags: List[str] = field(default_factory=list)
def __post_init__(self) -> None:
if self.year and self.year < 1930:
raise ValueError(
f"Received illegal year {self.year}, cannot be before 1930"
)
@dataclass
class Library:
books: List[Book]
name: str
LIBRARIES_LOADED = serdataclasses.load(LIBRARIES, List[Library])
Running this script should give you the following error message (traceback omitted because it can get somewhat long)
serdataclasses.errors.DeserializeError: Received illegal year 1929, cannot be before 1930
4. "<class '__main__.Book'>": "{'author': 'Susy Smith', 'tags': ['boring'], 'title': 'Hello, World!', 'year': 1929}"
3. 'typing.List[__main__.Book]': "[{'author': 'Susy Smith', 'tags': ['boring'], 'title': 'Hello, World!', 'year': 1929}, {'author': 'Beth John', 'title': 'The great showman'}, {'author': None, 'title': 'My favorite pony'}]"
2. "<class '__main__.Library'>": "{'books': [{'author': 'Susy Smith', 'tags': ['boring'], 'title': 'Hello, World!', 'year': 1929}, {'author': 'Beth John', 'title': 'The great showman'}, {'author': None, 'title': 'My favorite pony'}], 'name': 'Clark County Library'}"
1. 'typing.List[__main__.Library]': "[{'books': [{'author': 'Susy Smith', 'tags': ['boring'], 'title': 'Hello, World!', 'year': 1929}, {'author': 'Beth John', 'title': 'The great showman'}, {'author': None, 'title': 'My favorite pony'}], 'name': 'Clark County Library'}, {'books': [{'author': 'Smitty', 'tags': ['swell'], 'title': 'The great fun time', 'year': 1950}], 'name': 'Only 1 book here'}]"
The error message begins with the error message received, followed by a reverse presentation container types paired with the data they attempted to deserialize. This structure makes it incredibly easy to see not only what your error is, but where exactly it occurred in both the data serdataclasses.load
receives but also in the types serdataclasses.load
uses to attempt to deserialize the received data.
In serde, when working with resources external to your system, errors are inevitable. These error messages should hopefully make debugging your errors less annoying.
Frequent issues
My JSON keys contain whitespace, etc
Simple solution: use typeing.TypeDict
's backwards-compatibility syntax.
from pprint import pprint
from typing import List, TypedDict
import serdataclasses
from serdataclasses import OptionalProperty
DATA = [
{
"weird, key": 1,
"normal": 2,
},
{
"normal": 3,
},
]
DataItem = TypedDict(
"DataItem",
{
"weird, key": OptionalProperty[int],
"normal": int,
},
)
LOADED = serdataclasses.load(DATA, List[DataItem])
print("Loaded data:")
pprint(LOADED)
print("Re-serialized data:")
pprint(serdataclasses.dump(LOADED))
This prints the following to the console.
Loaded data:
[{'normal': 2, 'weird, key': 1},
{'normal': 3, 'weird, key': <Undefined property>}]
Re-serialized data:
[{'normal': 2, 'weird, key': 1}, {'normal': 3}]
Try changing values in your JSON data; you'll get runtime errors if your data does not conform to the above schema. Additionally, mypy
should call out any misused variable keys / types. In short, this has enabled a type-safe load and a perfectly sane dump.
Local Development
Local development for this project is simple.
Dependencies
Install the following tools manually.
Recommended
Set up development environment
make setup
Run Tests
make test
Notes
- Initially inspired by undictify and a PR I helped with. serdataclasses's goals are different; it's focused on serde instead of general function signature overrides.
- I also notice some striking similarities with a library called typedload (great minds think alike, I guess :p). I renamed my top-level functions to "load" and "dump" in typedload's homage. Unfortunately, as of version
1.20
, typedload does not handle all types of dataclasses elegantly (mainly, InitVar). Since typedload supports Python 3.5+, it never will elegantly handle all dataclasses without lots of unfortunate conditionals in the codebase. If you must use Python 3.7-, I suggest looking into typedload.
Written by
Samuel Roeca samuel.roeca@gmail.com