context_parser.py

from __future__ import annotations

import re
from typing import Any, Dict, List, Optional


def _to_bool(value: str) -> bool:
    return str(value or "").strip().lower() in {"true", "1", "yes", "y"}


def _looks_like_int(value: str) -> bool:
    return bool(re.fullmatch(r"-?\d+", str(value or "").strip()))


def split_unity_message(text: str) -> Dict[str, Any]:
    """
    Parses several Unity-style message formats and always returns a stable dict.

    Supported cases:
    1. Plain user text
    2. Hidden-context prefix + USER:/PROMPT:/MESSAGE:
    3. Structured multiline payload like:
         hint
         x/5 = 12
         0
         solve
         False
         answer
         answer
         algebra
         Quantitative
    """
    raw = (text or "").strip()

    result: Dict[str, Any] = {
        "hidden_context": "",
        "user_text": raw,
        "question_text": "",
        "hint_stage": 0,
        "user_last_input_type": "",
        "built_on_previous_turn": False,
        "help_mode": "",
        "intent": "",
        "topic": "",
        "category": "",
    }

    if not raw:
        return result

    # Case 1: hidden/system context followed by USER:/PROMPT:/MESSAGE:
    tagged_match = re.search(r"(?is)^(.*?)(?:\buser\b|\bprompt\b|\bmessage\b)\s*:\s*(.+)$", raw)
    if tagged_match:
        hidden = (tagged_match.group(1) or "").strip()
        user = (tagged_match.group(2) or "").strip()
        result["hidden_context"] = hidden
        result["user_text"] = user
        return result

    # Case 2: exact structured Unity payload block
    lines = [line.strip() for line in raw.splitlines() if line.strip()]
    if len(lines) >= 9 and _looks_like_int(lines[2]) and lines[4].lower() in {"true", "false"}:
        result["user_text"] = lines[0]
        result["question_text"] = lines[1]
        result["hint_stage"] = int(lines[2])
        result["user_last_input_type"] = lines[3]
        result["built_on_previous_turn"] = _to_bool(lines[4])
        result["help_mode"] = lines[5]
        result["intent"] = lines[6]
        result["topic"] = lines[7]
        result["category"] = lines[8]
        return result

    # Case 3: field-based payload
    def _extract_field(name: str) -> str:
        match = re.search(rf"(?im)^\s*{re.escape(name)}\s*[:=]\s*(.+?)\s*$", raw)
        return (match.group(1) or "").strip() if match else ""

    question_text = _extract_field("question") or _extract_field("question_text")
    user_text = _extract_field("user") or _extract_field("message") or _extract_field("prompt")
    hint_stage_text = _extract_field("hint_stage")
    user_last_input_type = _extract_field("user_last_input_type")
    built_on_previous_turn = _extract_field("built_on_previous_turn")
    help_mode = _extract_field("help_mode")
    intent = _extract_field("intent")
    topic = _extract_field("topic")
    category = _extract_field("category")

    if any([
        question_text,
        user_text,
        hint_stage_text,
        user_last_input_type,
        built_on_previous_turn,
        help_mode,
        intent,
        topic,
        category,
    ]):
        result["question_text"] = question_text
        result["user_text"] = user_text or raw
        result["hint_stage"] = int(hint_stage_text) if _looks_like_int(hint_stage_text) else 0
        result["user_last_input_type"] = user_last_input_type
        result["built_on_previous_turn"] = _to_bool(built_on_previous_turn)
        result["help_mode"] = help_mode
        result["intent"] = intent
        result["topic"] = topic
        result["category"] = category
        return result

    # Fallback: plain message
    return result


def _extract_options(text: str) -> List[str]:
    if not text:
        return []

    lines = [line.strip() for line in text.splitlines() if line.strip()]
    options: List[str] = []

    for line in lines:
        if re.match(r"^[A-E][\)\.\:]\s*", line, flags=re.I):
            options.append(re.sub(r"^[A-E][\)\.\:]\s*", "", line, flags=re.I).strip())

    if options:
        return options

    matches = re.findall(r"(?:^|\s)([A-E])[\)\.\:]\s*(.*?)(?=(?:\s+[A-E][\)\.\:])|$)", text, flags=re.I | re.S)
    if matches:
        return [m[1].strip() for m in matches if m[1].strip()]

    return []


def extract_game_context_fields(text: str) -> Dict[str, Any]:
    raw = (text or "").strip()

    result: Dict[str, Any] = {
        "question": "",
        "options": [],
        "difficulty": None,
        "category": None,
        "money": None,
        "has_choices": False,
        "looks_like_quant": False,
    }

    if not raw:
        return result

    q_match = re.search(r"\bquestion\s*[:=]\s*(.+?)(?=\n[A-Za-z_ ]+\s*[:=]|\Z)", raw, flags=re.I | re.S)
    if q_match:
        result["question"] = q_match.group(1).strip()

    opt_match = re.search(r"\b(?:options|choices|answers)\s*[:=]\s*(.+?)(?=\n[A-Za-z_ ]+\s*[:=]|\Z)", raw, flags=re.I | re.S)
    if opt_match:
        result["options"] = _extract_options(opt_match.group(1))

    if not result["options"]:
        result["options"] = _extract_options(raw)

    result["has_choices"] = len(result["options"]) > 0

    difficulty_match = re.search(r"\bdifficulty\s*[:=]\s*([A-Za-z0-9_\- ]+)", raw, flags=re.I)
    if difficulty_match:
        result["difficulty"] = difficulty_match.group(1).strip()

    category_match = re.search(r"\b(?:category|topic)\s*[:=]\s*([A-Za-z0-9_\- /]+)", raw, flags=re.I)
    if category_match:
        result["category"] = category_match.group(1).strip()

    money_match = re.search(r"\b(?:money|balance|bank)\s*[:=]\s*([\-]?\d+(?:\.\d+)?)", raw, flags=re.I)
    if money_match:
        try:
            result["money"] = float(money_match.group(1))
        except Exception:
            pass

    lower = raw.lower()
    result["looks_like_quant"] = any(
        token in lower
        for token in [
            "solve",
            "equation",
            "percent",
            "%",
            "ratio",
            "probability",
            "mean",
            "median",
            "algebra",
            "integer",
            "triangle",
            "circle",
        ]
    )

    return result


def detect_intent(text: str, incoming_help_mode: Optional[str] = None) -> str:
    forced = (incoming_help_mode or "").strip().lower()
    if forced in {
        "answer",
        "hint",
        "instruction",
        "walkthrough",
        "step_by_step",
        "explain",
        "method",
        "definition",
        "concept",
    }:
        return forced

    t = (text or "").strip().lower()

    if not t:
        return "answer"

    if (
        re.search(r"\bdefine\b", t)
        or re.search(r"\bdefinition\b", t)
        or re.search(r"\bwhat does\b", t)
        or re.search(r"\bwhat is meant by\b", t)
    ):
        return "definition"

    if re.search(r"\bhint\b", t) or re.search(r"\bclue\b", t) or re.search(r"\bnudge\b", t):
        return "hint"

    if (
        re.search(r"\bfirst step\b", t)
        or re.search(r"\bnext step\b", t)
        or re.search(r"\bwhat should i do first\b", t)
        or re.search(r"\bgive me the first step\b", t)
        or re.search(r"\bspecific step\b", t)
    ):
        return "instruction"

    if (
        re.search(r"\bwalk ?through\b", t)
        or re.search(r"\bstep by step\b", t)
        or re.search(r"\bfull working\b", t)
        or re.search(r"\bwork through\b", t)
    ):
        return "walkthrough"

    if re.search(r"\bexplain\b", t) or re.search(r"\bwhy\b", t):
        return "explain"

    if (
        re.search(r"\bmethod\b", t)
        or re.search(r"\bapproach\b", t)
        or re.search(r"\bhow do i solve\b", t)
        or re.search(r"\bhow to solve\b", t)
        or re.search(r"\bhow do i do this\b", t)
    ):
        return "method"

    if (
        re.search(r"\bconcept\b", t)
        or re.search(r"\bprinciple\b", t)
        or re.search(r"\brule\b", t)
        or re.search(r"\bwhat is the idea\b", t)
    ):
        return "concept"

    if (
        re.search(r"\bsolve\b", t)
        or re.search(r"\bwhat is\b", t)
        or re.search(r"\bfind\b", t)
        or re.search(r"\bgive (?:me )?the answer\b", t)
        or re.search(r"\bjust the answer\b", t)
        or re.search(r"\banswer only\b", t)
        or re.search(r"\bcalculate\b", t)
    ):
        return "answer"

    return "answer"


def intent_to_help_mode(intent: str) -> str:
    if intent in {"walkthrough", "step_by_step"}:
        return "walkthrough"
    if intent in {"explain", "method", "concept"}:
        return "explain"
    if intent == "hint":
        return "hint"
    if intent in {"definition", "instruction"}:
        return intent
    return "answer"