"""
Response Parser Module

This module provides low-coupling, high-cohesion parsing utilities for extracting
tool calls from LLM responses and converting them to OpenAI-compatible format.

Design principles:
- Single Responsibility: Each function handles one specific parsing task
- Testability: Pure functions that are easy to unit test
- Type Safety: Uses Pydantic models for validation
"""

import re
import json
import logging
from typing import Optional, List, Dict, Any
from uuid import uuid4

from app.models import ResponseMessage, ToolCall, ToolCallFunction

logger = logging.getLogger(__name__)


# Constants for tool call parsing
# Using XML-style tags for clarity and better compatibility with JSON
# LLM should emit:<tool_call>{"name": "...", "arguments": {...}}</tool_call>
TOOL_CALL_START_TAG = "{"
TOOL_CALL_END_TAG = "}"


class ToolCallParseError(Exception):
    """Raised when tool call parsing fails."""
    pass


class ResponseParser:
    """
    Parser for converting LLM text responses into structured ResponseMessage objects.

    This class encapsulates all parsing logic for tool calls, making it easy to test
    and maintain. It follows the Single Responsibility Principle by focusing solely
    on parsing responses.
    """

    def __init__(self, tool_call_start_tag: str = TOOL_CALL_START_TAG,
                 tool_call_end_tag: str = TOOL_CALL_END_TAG):
        """
        Initialize the parser with configurable tags.

        Args:
            tool_call_start_tag: The opening tag for tool calls (default: {...")
            tool_call_end_tag: The closing tag for tool calls (default: ...})
        """
        self.tool_call_start_tag = tool_call_start_tag
        self.tool_call_end_tag = tool_call_end_tag
        self._compile_regex()

    def _compile_regex(self):
        """Compile the regex pattern for tool call extraction."""
        # Escape special regex characters in the tags
        escaped_start = re.escape(self.tool_call_start_tag)
        escaped_end = re.escape(self.tool_call_end_tag)
        # Match from start tag to end tag (greedy), including both tags
        # This ensures we capture the complete JSON object
        self._tool_call_pattern = re.compile(
            f"{escaped_start}.*{escaped_end}",
            re.DOTALL
        )

    def _extract_valid_json(self, text: str) -> Optional[str]:
        """
        Extract a valid JSON object from text that may contain extra content.

        This handles cases where non-greedy regex matching includes incomplete JSON.

        Args:
            text: Text that should contain a JSON object

        Returns:
            The extracted valid JSON string, or None if not found
        """
        text = text.lstrip()  # Only strip leading whitespace

        # Find the first opening brace (the start of JSON)
        start_idx = text.find('{')
        if start_idx < 0:
            return None

        text = text[start_idx:]  # Start from the first opening brace

        # Find the matching closing brace by counting brackets
        brace_count = 0
        in_string = False
        escape_next = False

        for i, char in enumerate(text):
            if escape_next:
                escape_next = False
                continue

            if char == '\\' and in_string:
                escape_next = True
                continue

            if char == '"':
                in_string = not in_string
                continue

            if not in_string:
                if char == '{':
                    brace_count += 1
                elif char == '}':
                    brace_count -= 1
                    if brace_count == 0:
                        # Found matching closing brace
                        return text[:i+1]

        return None

    def parse(self, llm_response: str) -> ResponseMessage:
        """
        Parse an LLM response and extract tool calls if present.

        This is the main entry point for parsing. It handles both:
        1. Responses with tool calls (wrapped in tags)
        2. Regular text responses

        Args:
            llm_response: The raw text response from the LLM

        Returns:
            ResponseMessage with content and optionally tool_calls

        Example:
            >>> parser = ResponseParser()
            >>> response = parser.parse('Hello world')
            >>> response.content
            'Hello world'

            >>> response = parser.parse('Check the weather.<invo>{"name": "weather", "arguments": {...}}<invoke>')
            >>> response.tool_calls[0].function.name
            'weather'
        """
        if not llm_response:
            return ResponseMessage(content=None)

        try:
            match = self._tool_call_pattern.search(llm_response)

            if match:
                return self._parse_tool_call_response(llm_response, match)
            else:
                return self._parse_text_only_response(llm_response)

        except Exception as e:
            logger.warning(f"Failed to parse LLM response: {e}. Returning as text.")
            return ResponseMessage(content=llm_response)

    def _parse_tool_call_response(self, llm_response: str, match: re.Match) -> ResponseMessage:
        """
        Parse a response that contains tool calls.

        Args:
            llm_response: The full LLM response
            match: The regex match object containing the tool call

        Returns:
            ResponseMessage with content and tool_calls
        """
        # The match includes start and end tags, so strip them
        matched_text = match.group(0)
        tool_call_str = matched_text[len(self.tool_call_start_tag):-len(self.tool_call_end_tag)]

        # Extract valid JSON by finding matching braces
        json_str = self._extract_valid_json(tool_call_str)
        if json_str is None:
            # Fallback to trying to parse the entire string
            json_str = tool_call_str

        try:
            tool_call_data = json.loads(json_str)

            # Extract content before the tool call tag
            parts = llm_response.split(self.tool_call_start_tag, 1)
            content = parts[0].strip() if parts[0] else None

            # Create the tool call object
            tool_call = self._create_tool_call(tool_call_data)

            return ResponseMessage(
                content=content,
                tool_calls=[tool_call]
            )

        except json.JSONDecodeError as e:
            raise ToolCallParseError(f"Invalid JSON in tool call: {tool_call_str}. Error: {e}")

    def _parse_text_only_response(self, llm_response: str) -> ResponseMessage:
        """
        Parse a response with no tool calls.

        Args:
            llm_response: The full LLM response

        Returns:
            ResponseMessage with content only
        """
        return ResponseMessage(content=llm_response.strip())

    def _create_tool_call(self, tool_call_data: Dict[str, Any]) -> ToolCall:
        """
        Create a ToolCall object from parsed data.

        Args:
            tool_call_data: Dictionary containing 'name' and optionally 'arguments'

        Returns:
            ToolCall object

        Raises:
            ToolCallParseError: If required fields are missing
        """
        name = tool_call_data.get("name")
        if not name:
            raise ToolCallParseError("Tool call missing 'name' field")

        arguments = tool_call_data.get("arguments", {})

        # Generate a unique ID for the tool call
        tool_call_id = f"call_{name}_{str(uuid4())[:8]}"

        return ToolCall(
            id=tool_call_id,
            type="function",
            function=ToolCallFunction(
                name=name,
                arguments=json.dumps(arguments)
            )
        )

    def parse_streaming_chunks(self, chunks: List[str]) -> ResponseMessage:
        """
        Parse a list of streaming chunks and aggregate into a ResponseMessage.

        This method handles streaming responses where tool calls might be
        split across multiple chunks.

        Args:
            chunks: List of content chunks from streaming response

        Returns:
            Parsed ResponseMessage
        """
        full_content = "".join(chunks)
        return self.parse(full_content)

    def parse_native_tool_calls(self, llm_response: Dict[str, Any]) -> ResponseMessage:
        """
        Parse a response that already has native OpenAI-format tool calls.

        Some LLMs natively support tool calling and return them in the standard
        OpenAI format. This method handles those responses.

        Args:
            llm_response: Dictionary response from LLM with potential tool_calls field

        Returns:
            ResponseMessage with parsed tool_calls or content
        """
        if "tool_calls" in llm_response and llm_response["tool_calls"]:
            # Parse native tool calls
            tool_calls = []
            for tc in llm_response["tool_calls"]:
                tool_calls.append(ToolCall(
                    id=tc.get("id", f"call_{str(uuid4())[:8]}"),
                    type=tc.get("type", "function"),
                    function=ToolCallFunction(
                        name=tc["function"]["name"],
                        arguments=tc["function"]["arguments"]
                    )
                ))

            return ResponseMessage(
                content=llm_response.get("content"),
                tool_calls=tool_calls
            )
        else:
            # Fallback to text parsing
            content = llm_response.get("content", "")
            return self.parse(content)


# Convenience functions for backward compatibility and ease of use

def parse_response(llm_response: str) -> ResponseMessage:
    """
    Parse an LLM response using default parser settings.

    This is a convenience function for simple use cases.

    Args:
        llm_response: The raw text response from the LLM

    Returns:
        ResponseMessage with parsed content and tool calls
    """
    parser = ResponseParser()
    return parser.parse(llm_response)


def parse_response_with_custom_tags(llm_response: str,
                                    start_tag: str,
                                    end_tag: str) -> ResponseMessage:
    """
    Parse an LLM response using custom tool call tags.

    Args:
        llm_response: The raw text response from the LLM
        start_tag: Custom start tag for tool calls
        end_tag: Custom end tag for tool calls

    Returns:
        ResponseMessage with parsed content and tool calls
    """
    parser = ResponseParser(tool_call_start_tag=start_tag, tool_call_end_tag=end_tag)
    return parser.parse(llm_response)