"""
Volcengine (Doubao) Voice Plugin for Hermes Agent.

Adds 'volcengine' as a TTS and STT provider by monkey-patching the
Hermes dispatch layer. Also registers a pre_llm_call hook that injects
voice-friendly response instructions when volcengine is the active provider.

Usage:
  1. Set VOLCENGINE_VOICE_API_KEY in ~/.hermes/env.d/volcengine.env
  2. Set config.yaml:
       tts.provider: volcengine
       stt.provider: volcengine
  3. Enable: hermes plugins enable volcengine-voice
  4. Restart Hermes gateway

Why monkey-patching: Hermes's TTS/STT system uses a provider-dispatch
pattern inside existing tools (text_to_speech_tool / transcribe_audio).
Registering new tools would create separate tool names the LLM can't
use — the monkey-patch intercepts the existing dispatch so the LLM
calls the same tool and gets routed to volcengine transparently.
"""
import logging

logger = logging.getLogger(__name__)


_VOICE_INSTRUCTION = (
    "You are responding via voice. Keep replies concise (2-4 sentences), "
    "use natural conversational tone, avoid markdown formatting that doesn't "
    "read well aloud (no tables, no code blocks unless asked), and speak "
    "directly to the user without meta-commentary. "
    "When you want to control the speaking style, prefix your TTS text with "
    "one or more [emotion] markers, e.g. '[开心] 今天天气真好' or "
    "'[开心][快速] 你好'; the bracketed part is a non-billed style "
    "instruction and is stripped from the spoken text."
)


def _pre_llm_inject_voice_context(
    session_id: str,
    user_message: str,
    conversation_history: list,
    is_first_turn: bool,
    model: str,
    platform: str,
    **kwargs,
):
    """Inject voice-friendly instructions when volcengine is the TTS provider."""
    try:
        import tools.tts_tool as tts_module
        tts_config = tts_module._load_tts_config()
        provider = tts_module._get_provider(tts_config)
        if provider == "volcengine":
            return {"context": _VOICE_INSTRUCTION}
    except Exception:
        pass
    return None


def register(ctx) -> None:
    """Monkey-patch the TTS/STT dispatch and register hooks."""

    # ── Patch TTS ──────────────────────────────────────────────────
    import tools.tts_tool as tts_module

    _original_tts = tts_module.text_to_speech_tool

    def patched_text_to_speech(text: str, output_path=None):
        """Wrapped TTS: if provider is 'volcengine', route to Doubao API."""
        tts_config = tts_module._load_tts_config()
        provider = tts_module._get_provider(tts_config)

        if provider != "volcengine":
            return _original_tts(text, output_path)

        # ── Volcengine path ─────────────────────────────────────
        import datetime
        import json as _json
        import os as _os
        from pathlib import Path as _Path

        from hermes_plugins.volcengine_voice.tts import _volcengine_tts

        if not text or not text.strip():
            return tts_module.tool_error("Text is required", success=False)

        # Truncate long text
        max_len = 5000
        if len(text) > max_len:
            logger.warning("Volcengine TTS: truncating %d chars to %d", len(text), max_len)
            text = text[:max_len]

        # Determine output path
        if output_path:
            file_path = _Path(output_path).expanduser()
        else:
            timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
            out_dir = _Path(tts_module.DEFAULT_OUTPUT_DIR)
            out_dir.mkdir(parents=True, exist_ok=True)
            file_path = out_dir / f"tts_{timestamp}.mp3"

        file_path.parent.mkdir(parents=True, exist_ok=True)
        file_str = str(file_path)

        try:
            _volcengine_tts(text, file_str, tts_config)
        except Exception as e:
            logger.error("Volcengine TTS failed: %s", e)
            return _json.dumps({
                "success": False,
                "error": f"Volcengine TTS error: {e}",
            }, ensure_ascii=False)

        media_tag = f"MEDIA:{file_str}"
        return _json.dumps({
            "success": True,
            "file_path": file_str,
            "media_tag": media_tag,
            "provider": "volcengine",
        }, ensure_ascii=False)

    # Replace the function reference in the module
    tts_module.text_to_speech_tool = patched_text_to_speech
    logger.info("volcengine-voice: patched text_to_speech_tool ← volcengine provider")

    # ── Patch STT ──────────────────────────────────────────────────
    import tools.transcription_tools as stt_module

    _original_transcribe = stt_module.transcribe_audio

    def patched_transcribe_audio(file_path, model=None):
        """Wrapped STT: if provider is 'volcengine', route to Doubao ASR."""
        stt_config = stt_module._load_stt_config()
        provider = stt_module._get_provider(stt_config)

        if provider != "volcengine":
            return _original_transcribe(file_path, model)

        from hermes_plugins.volcengine_voice.stt import volcengine_transcribe

        logger.info("Volcengine STT: transcribing %s...", file_path)
        return volcengine_transcribe(file_path, stt_config)

    stt_module.transcribe_audio = patched_transcribe_audio
    logger.info("volcengine-voice: patched transcribe_audio ← volcengine provider")

    # ── Register hooks ───────────────────────────────────────────
    ctx.register_hook("pre_llm_call", _pre_llm_inject_voice_context)
    logger.info("volcengine-voice: registered pre_llm_call hook")