from __future__ import annotations

import logging
from typing import Any

from .demo_pack import extract_photo_caption
from .index import tokenize
from .llm import generate_repair_checklist

logger = logging.getLogger(__name__)

def build_response(symptom: str, equipment_type: str, location: str, notes: str, photo_path: str | None, search_index: Any) -> tuple[str, list[list[Any]], dict[str, Any]]:
    query = " ".join([symptom, equipment_type, location, notes]).strip()
    if photo_path:
        photo_caption = extract_photo_caption(photo_path)
        query += " " + photo_caption
    else:
        photo_caption = ""
    section_hits = search_index.search_sections(query, top_k=5)
    job_hits = search_index.search_jobs(query, top_k=3)
    top_score = section_hits[0].score if section_hits else 0.0
    too_vague = len(tokenize(symptom)) < 4
    insufficient = too_vague or (not section_hits) or top_score < 0.10

    citations_rows: list[list[Any]] = []
    citation_lines = []
    reference_texts = []
    for hit in section_hits[:3]:
        title = hit.title
        citations_rows.append([f"{hit.score:.3f}", hit.kind, hit.record_id, title, hit.citation])
        citation_lines.append(f"- {title} — {hit.citation} (score {hit.score:.3f})")
        reference_texts.append(hit.text)

    logger.info(f"Generating response with LLM (insufficient={insufficient})...")
    generated_text, stats = generate_repair_checklist(
        symptom=symptom,
        equipment_type=equipment_type,
        location=location,
        notes=notes,
        photo_caption=photo_caption,
        references=reference_texts,
        insufficient=insufficient
    )
    
    body = generated_text + f"\n\n**Inference Stats:**\nModel: {stats['model_id']}\nAdapter: {stats['adapter']}\nTokens: {stats['total_tokens']} (P: {stats['prompt_tokens']}, C: {stats['completion_tokens']})"
    if citation_lines:
        body += "\n\nCitations:\n" + "\n".join(citation_lines)
    
    payload = {
        "status": "insufficient_evidence" if insufficient else "ok",
        "photo_caption": photo_caption,
        "retrieved_sections": [hit.citation for hit in section_hits],
        "retrieved_jobs": [hit.citation for hit in job_hits],
        "top_scores": [hit.score for hit in section_hits[:3]],
        "llm_stats": stats
    }
    return body, citations_rows, payload