from dataclasses import dataclass, field, make_dataclass
from enum import Enum

from src.about import Tasks


def fields(raw_class):
    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]


# These classes are for user facing column names,
# to avoid having to change them all around the code
# when a modif is needed
@dataclass
class ColumnContent:
    name: str
    type: str
    displayed_by_default: bool
    hidden: bool = False
    never_hidden: bool = False

## Leaderboard columns
auto_eval_column_dict = []


def column_field(name: str, type: str, displayed_by_default: bool, hidden: bool = False, never_hidden: bool = False):
    return (
        ColumnContent,
        field(
            default_factory=lambda: ColumnContent(
                name,
                type,
                displayed_by_default,
                hidden,
                never_hidden,
            )
        ),
    )


# Init
auto_eval_column_dict.append(["model_type_symbol", *column_field("T", "str", True, never_hidden=True)])
auto_eval_column_dict.append(["model", *column_field("Model", "markdown", True, never_hidden=True)])
auto_eval_column_dict.append(["average", *column_field("Average ⬆️", "number", True)])
for task in Tasks:
    auto_eval_column_dict.append([task.name, *column_field(task.value.col_name, "number", True)])
auto_eval_column_dict.append(["model_type", *column_field("Type", "str", False)])
auto_eval_column_dict.append(["architecture", *column_field("Architecture", "str", False)])
auto_eval_column_dict.append(["weight_type", *column_field("Weight type", "str", False, True)])
auto_eval_column_dict.append(["precision", *column_field("Precision", "str", False)])
auto_eval_column_dict.append(["license", *column_field("Hub License", "str", False)])
auto_eval_column_dict.append(["params", *column_field("#Params (B)", "number", False)])
auto_eval_column_dict.append(["likes", *column_field("Hub ❤️", "number", False)])
auto_eval_column_dict.append(["still_on_hub", *column_field("Available on the hub", "bool", False)])
auto_eval_column_dict.append(["revision", *column_field("Model sha", "str", False, False)])

# We use make dataclass to dynamically fill the scores from Tasks
AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
for field_name, _, field_def in auto_eval_column_dict:
    setattr(AutoEvalColumn, field_name, field_def.default_factory())

## For the queue columns in the submission tab
@dataclass(frozen=True)
class EvalQueueColumn:  # Queue column
    model = ColumnContent("model", "markdown", True)
    revision = ColumnContent("revision", "str", True)
    precision = ColumnContent("precision", "str", True)
    submitted_time = ColumnContent("submitted_time", "str", True)
    status = ColumnContent("status", "str", True)

## All the model information that we might need
@dataclass
class ModelDetails:
    name: str
    display_name: str = ""
    symbol: str = "" # emoji


class ModelType(Enum):
    PT = ModelDetails(name="pretrained", symbol="P")
    FT = ModelDetails(name="fine-tuned", symbol="F")
    IFT = ModelDetails(name="instruction-tuned", symbol="I")
    RL = ModelDetails(name="RL-tuned", symbol="R")
    Unknown = ModelDetails(name="", symbol="?")

    def to_str(self, separator=" "):
        return f"{self.value.symbol}{separator}{self.value.name}"

    @staticmethod
    def from_str(type):
        if "fine-tuned" in type or "🔶" in type:
            return ModelType.FT
        if "pretrained" in type or "🟢" in type:
            return ModelType.PT
        if "RL-tuned" in type or "🟦" in type:
            return ModelType.RL
        if "instruction-tuned" in type or "⭕" in type:
            return ModelType.IFT
        return ModelType.Unknown

class WeightType(Enum):
    Adapter = ModelDetails("Adapter")
    Original = ModelDetails("Original")
    Delta = ModelDetails("Delta")

class Precision(Enum):
    float16 = ModelDetails("float16")
    bfloat16 = ModelDetails("bfloat16")
    Unknown = ModelDetails("?")

    def from_str(precision):
        if precision in ["torch.float16", "float16"]:
            return Precision.float16
        if precision in ["torch.bfloat16", "bfloat16"]:
            return Precision.bfloat16
        return Precision.Unknown

# Column selection
COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]

EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]

BENCHMARK_COLS = [t.value.col_name for t in Tasks]