Restore GraphQL API with Strawberry

[paula-mg]

Fixes #139

[coretl]

@callumforrester I am not tied to dataclass, but I have a requirement for positional args. I have a choice of directions for you:

1. Continue using dataclasses, add support to strawberry, probably using something like what I needed to do to autodoc_pydantic
2. Ditch dataclasses and make a BaseModel subclass with positional arg support. I think we could make this work both at runtime using something like this and at static analysis by overriding the dataclass_transform. The closed issue would suggest that pydantic would never accept such an approach upstream.

[coretl]

To elaborate on 2, this works for both static analysis and at runtime:

from __future__ import annotations

from abc import ABCMeta
from typing import Any

from pydantic import BaseModel, Field
from typing_extensions import dataclass_transform

# TODO: not sure about the others like NoInitField and PrivateAttr
@dataclass_transform(field_specifiers=(Field,))
class PosargsMeta(type(BaseModel), ABCMeta):
    def __new__(
        mcs,
        cls_name: str,
        bases: tuple[type[Any], ...],
        namespace: dict[str, Any],
        **kwargs: Any,
    ) -> type:
        cls = super().__new__(mcs, cls_name, bases, namespace, **kwargs)
        original_init = cls.__init__

        def patched_init(self, *args, **kwargs):
            for k, v in zip(cls.model_fields, args, strict=False):
                kwargs[k] = v
            original_init(self, **kwargs)

        cls.__init__ = patched_init
        return cls

class Spec(BaseModel, metaclass=PosargsMeta):
    pass

class Line(Spec):
    start: float = Field(description="Midpoint of the first point of the line")
    stop: float = Field(description="Midpoint of the last point of the line")
    num: int = Field(min=1, description="Number of frames to produce")

# pyright and pydantic are happy with this...
obj = Line(3, 4, 5)

[callumforrester]

I have no particular preference, we should discuss with @paula-mg since she'll be doing the work.

[paula-mg]

Hi @callumforrester, I started working on this a while ago, could I still be working on this issue?

[callumforrester]

@coretl what's the current state of scanspec?

[coretl]

We have no plans for modifications in the next few months, the next thing we need to do is make duration a special field on a Frame rather than just another key. Long term we need to look at exposing a function that gives the position at a given time.