"""
Asks a local llama-swap model to suggest a topic, subtopic, and
destination folder for a file, given its filename, current folder, and
(if available) an extracted content excerpt.

Talks to llama-swap's OpenAI-compatible endpoint over plain HTTP using
only the stdlib — nothing about a file's name or content ever leaves the
machine this endpoint points at.
"""

import json
import os
import urllib.error
import urllib.request

DEFAULT_ENDPOINT = os.environ.get("VAULT_RENAMER_LLM_ENDPOINT", "http://localhost:8080/v1/chat/completions")
DEFAULT_MODEL = os.environ.get("VAULT_RENAMER_LLM_MODEL", "daily")
TIMEOUT_SECONDS = 30

PROMPT_TEMPLATE = """You are helping organise a personal file vault. Given a file's name, its current folder, an optional content excerpt, and the list of real destination folders that already exist, suggest a short topic, a short subtopic, and which existing folder it belongs in.

Rules:
- topic and subtopic: a few lowercase words each, no punctuation, describing what the file is actually about — not a restatement of the filename.
- dest_folder: MUST be exactly one of the folders listed below, copied verbatim. If genuinely unsure, pick the closest match rather than inventing a new folder.
- Reply with ONLY a JSON object, nothing else, no markdown fence: {{"topic": "...", "subtopic": "...", "dest_folder": "..."}}

Current filename: {filename}
Current folder: {current_folder}
Content excerpt (may be empty): {excerpt}

Real destination folders:
{folders}
"""


class LLMSuggestError(Exception):
    """Callers should catch this and treat it as 'no suggestion available'
    for this file — never let a model/network hiccup crash the batch."""


def suggest(filename: str, current_folder: str, excerpt: str, folders: list,
            endpoint: str = DEFAULT_ENDPOINT, model: str = DEFAULT_MODEL) -> dict:
    prompt = PROMPT_TEMPLATE.format(
        filename=filename,
        current_folder=current_folder or "(root)",
        excerpt=(excerpt or "")[:1500],
        folders="\n".join(folders),
    )

    payload = {
        "model": model,
        "messages": [{"role": "user", "content": prompt}],
        "temperature": 0.2,
        "max_tokens": 200,
    }

    req = urllib.request.Request(
        endpoint,
        data=json.dumps(payload).encode("utf-8"),
        headers={"Content-Type": "application/json"},
        method="POST",
    )

    try:
        with urllib.request.urlopen(req, timeout=TIMEOUT_SECONDS) as resp:
            data = json.loads(resp.read().decode("utf-8"))
    except urllib.error.URLError as e:
        raise LLMSuggestError(f"Couldn't reach {endpoint} — is llama-swap running? ({e})")
    except TimeoutError:
        raise LLMSuggestError(f"No response within {TIMEOUT_SECONDS}s — model may still be loading.")

    try:
        content = data["choices"][0]["message"]["content"]
    except (KeyError, IndexError):
        raise LLMSuggestError(f"Unexpected response shape from {endpoint}: {data}")

    parsed = _extract_json(content)
    if parsed is None:
        raise LLMSuggestError(f"Model didn't reply with valid JSON: {content[:200]!r}")

    for key in ("topic", "subtopic", "dest_folder"):
        parsed.setdefault(key, "")

    if parsed["dest_folder"] not in folders:
        # The model invented a folder that doesn't exist — never apply a
        # destination that isn't real.
        parsed["dest_folder"] = ""

    return parsed


def _extract_json(text: str):
    text = text.strip()
    if text.startswith("```"):
        text = text.strip("`")
        if text.lower().startswith("json"):
            text = text[4:]
        text = text.strip()
    try:
        return json.loads(text)
    except json.JSONDecodeError:
        pass
    start, end = text.find("{"), text.rfind("}")
    if start != -1 and end != -1 and end > start:
        try:
            return json.loads(text[start:end + 1])
        except json.JSONDecodeError:
            return None
    return None