import json
from io import BytesIO, IOBase, StringIO
from typing import Type, TypeVar

from pydantic import BaseModel, TypeAdapter
from ruamel.yaml import YAML
from sqlalchemy import Text, TypeDecorator

T = TypeVar("T", bound=BaseModel)


def from_yaml(model_type: Type[T], raw: str | bytes | IOBase) -> T:
    if isinstance(raw, str):
        stream = StringIO(raw)
    elif isinstance(raw, IOBase):
        stream = raw
    else:
        stream = BytesIO(raw)
    reader = YAML(typ="safe", pure=True)
    return TypeAdapter(model_type).validate_python(reader.load(stream))


def to_yaml(model: BaseModel) -> str:
    stream = StringIO()
    writer = YAML(typ="safe", pure=True)
    writer.default_flow_style = False
    writer.indent(mapping=None, sequence=None, offset=None)
    writer.sort_base_mapping_type_on_output = False
    writer.dump(json.loads(model.model_dump_json(exclude_none=True)), stream)
    stream.seek(0)
    return stream.read()


class PydanticYaml(TypeDecorator):
    impl = Text

    def __init__(self, model: type[BaseModel], *args, **kwargs):
        super().__init__(args, kwargs)
        self.model = model

        # Fixes table generation bug in :meth:`visit_TEXT` in
        # :class:`sqlalchemy.dialects.mysql.base.MySQLTypeCompiler` where length
        # is set to an empty tuple and fails the conditional statement check.
        self.impl.length = None

    def process_bind_param(self, value, dialect):
        if value is None:
            return value
        if not isinstance(value, BaseModel):
            assert ValueError("Value must be type BaseModel.")
        return to_yaml(value)

    def process_result_value(self, value, dialect):
        if value is not None:
            value = from_yaml(self.model, value)
        return value

"""
# Usage example.

from pydantic import BaseModel
from sqlalchemy.orm import DeclarativeBase


class Base(DeclarativeBase):
    pass


class PydanticType(BaseModel):
    field: str

        
class Entity(Base):
    id: Mapped[int] = mapped_column(primary_key=True)
    data: Mapped[PydanticType] = mapped_column(PydanticYaml(PydanticType))
"""