Pydantic serialization

[adrianschneider94]

cadquery-pydantic

https://github.com/sangl-spezialtransformatoren/cadquery-pydantic

A package that holds pydantic schemas for CadQuery classes. With this, you'll be able to use CadQuery objects just like a pydantic model and serialize/deserialize them to json.

Warning The serialization format is not stable yet, future versions won't be able to parse objects serialized with the current implementation.

import cadquery as cq
from pydantic import TypeAdapter
from cadquery_pydantic import patch_cadquery

patch_cadquery()

# Direct serialization of a CadQuery object
box = cq.Workplane("XY").box(1, 1, 1)
json_result = TypeAdapter(cq.Workplane).dump_json(box)

# Deserialize back to a CadQuery object
loaded_box = TypeAdapter(cq.Workplane).validate_json(json_result)

This facilitates caching with backends like a database or redis and the usage in APIs via FastAPI etc. Details can be found in the Readme.

[adam-urbanczyk]

Thanks for sharing, interesting. Any reason (e.g. some esoteric limitation of CQ) for using monkey patching instead of https://docs.pydantic.dev/latest/concepts/types/#handling-third-party-types ?

[adrianschneider94]

It makes the code on the consuming side less verbose when you can use the classes directly instead of having to annotate them. From the cadquery side there is no issue.
It would be quite easy to provide classes for annotation as an alternative API though.

[adam-urbanczyk]

Am I missing something (not a Pydantic user), or is this what you refer to as verbose?

class Model(BaseModel):
    third_party_type: PydanticShape

[adrianschneider94]

Yes, when you have many fields, you will read "Pydantic" a lot, without information conveyed (or you need to take care of import alises etc.). Also you need to add the Annotation to every subclass you write and have mixed imports of cadquery and cadquery-pydantic.

Monkey patching can happen once and all other imports will be directly from cadquery. This is the closest to what would be the cleanest approach in my eyes, where the schemas are defined in the cadquery code, directly on the classes.

Still, it probably makes sense, to provide both mechanisms, so the package user can decide.