"""Renderer selection registry for PMFlow AI turn runtime.

The registry chooses renderer candidates from canonical TurnState. It does not
inspect user text and it does not generate answers. Runtime-owned renderer
functions execute the selected candidates and may later be moved behind the
same keys without changing the turn loop.
"""

from __future__ import annotations

from dataclasses import dataclass
from typing import Literal

from ..schemas.pmflow_ai_turn_runtime import PMFlowAITurnState
from .pmflow_ai_static_contract_registry import PMFlowAIStaticContractRegistry


RendererKey = Literal[
    "static_contract",
    "collection_snapshot",
    "project_operating_packet",
    "open_judgement",
    "llm_state_compose",
    "safe_fallback",
]


@dataclass(frozen=True)
class PMFlowAIAnswerRendererCandidate:
    renderer_key: RendererKey
    reason: str
    priority: int


class PMFlowAIAnswerRendererRegistry:
    """Select deterministic and generative renderer candidates for a turn."""

    def select_candidates(self, turn_state: PMFlowAITurnState) -> list[PMFlowAIAnswerRendererCandidate]:
        candidates: list[PMFlowAIAnswerRendererCandidate] = []
        if PMFlowAIStaticContractRegistry().resolve_for_turn_state(turn_state) is not None:
            candidates.append(
                PMFlowAIAnswerRendererCandidate(
                    renderer_key="static_contract",
                    reason="权威真源是静态产品合同，优先确定性表达。",
                    priority=10,
                )
            )
        if _can_render_collection(turn_state):
            candidates.append(
                PMFlowAIAnswerRendererCandidate(
                    renderer_key="collection_snapshot",
                    reason="TurnState 已包含集合观察或集合 draft，可确定性展示列表。",
                    priority=20,
                )
            )
        if _has_project_operating_packet(turn_state):
            candidates.append(
                PMFlowAIAnswerRendererCandidate(
                    renderer_key="project_operating_packet",
                    reason="TurnState 已包含项目运行态 packet，可确定性表达项目状态。",
                    priority=30,
                )
            )
        if not list(turn_state.evidence_state.known_facts or []):
            candidates.append(
                PMFlowAIAnswerRendererCandidate(
                    renderer_key="open_judgement",
                    reason="当前没有可吸收 facts，尝试基于开放判断状态给出受控初判。",
                    priority=40,
                )
            )
        candidates.extend(
            [
                PMFlowAIAnswerRendererCandidate(
                    renderer_key="llm_state_compose",
                    reason="使用 TurnState 进行受控合成。",
                    priority=90,
                ),
                PMFlowAIAnswerRendererCandidate(
                    renderer_key="safe_fallback",
                    reason="无 renderer 产出时使用客户安全 fallback。",
                    priority=100,
                ),
            ]
        )
        return sorted(candidates, key=lambda item: item.priority)


def _can_render_collection(turn_state: PMFlowAITurnState) -> bool:
    selected = turn_state.selected_frame
    if selected is not None and selected.operation_kind == "list_collection":
        return True
    if str(turn_state.answer_state.answer_shape or "") == "fact_list":
        return True
    draft = dict(turn_state.customer_packet_draft or {})
    if isinstance(draft.get("items"), list) or isinstance(draft.get("count"), int):
        return True
    return any(
        observation.observation_type == "collection_snapshot"
        and (bool(observation.items) or observation.count is not None)
        for observation in list(turn_state.observations or [])
    )


def _has_project_operating_packet(turn_state: PMFlowAITurnState) -> bool:
    structured = dict(turn_state.evidence_state.structured_evidence or {})
    if isinstance(structured.get("project_operating_evidence_packet"), dict):
        return True
    packets = structured.get("project_operating_evidence_packets")
    return any(isinstance(packet, dict) for packet in list(packets or []))