"""Per-file summarisation for project folder integration."""
from __future__ import annotations

from dataclasses import dataclass

from langchain_core.messages import HumanMessage, SystemMessage

from app.core.langfuse_client import (
    compile_prompt,
    extract_usage,
    get_prompt_or_fallback,
)
from app.core.llm import get_llm

_TEXT_FALLBACK = (
    "You are summarising a file for an AI assistant that helps the user manage a project.\n"
    "Produce a single sentence (<=30 words, <=200 chars) that captures the file's purpose "
    "and most important detail.\nFile extension: {ext}\nFile name: {name}\nContent (truncated if long):\n{content}"
)
_IMAGE_FALLBACK = (
    "You are summarising an image attached to a project folder.\n"
    "Produce a single sentence (<=30 words, <=200 chars) describing what the image shows "
    "and any obvious purpose (logo, screenshot, diagram, photo of a whiteboard, etc.)."
)
_MAX_INPUT_CHARS = 6000


@dataclass
class IndexResult:
    summary: str
    tokens_used: int


async def _llm_text(messages: list) -> object:
    """Make the LLM call for text summarisation.

    Defined as a standalone async function so tests can patch it cleanly
    without needing to mock the LLM object itself.
    """
    llm = get_llm(model="gpt-4o-mini", temperature=0.2)
    return await llm.ainvoke(messages)


async def summarize_text(*, content: str, ext: str, name: str) -> IndexResult:
    """Return a compact summary of a text file.

    Parameters
    ----------
    content:
        Raw text content of the file (will be truncated to _MAX_INPUT_CHARS).
    ext:
        File extension including the leading dot, e.g. ``".md"``.
    name:
        File name, e.g. ``"kickoff.md"``.
    """
    template, prompt_obj = get_prompt_or_fallback("folder_file_summary_text", _TEXT_FALLBACK)
    truncated = content[:_MAX_INPUT_CHARS]
    compiled = compile_prompt(template, prompt_obj, ext=ext, name=name, content=truncated)
    messages = [
        SystemMessage(content=compiled),
        HumanMessage(content="Summarise this file."),
    ]
    response = await _llm_text(messages)
    usage = extract_usage(response)
    summary = (response.content or "").strip()[:500]
    return IndexResult(summary=summary, tokens_used=usage.get("total", 0))