"""End-to-end test for the full CityQuest AI pipeline.

Run with ``CITYQUEST_FAST_TEST=1`` to skip GGUF model downloads
and use mock generation for LLM-dependent steps.

Usage:
    python test_end_to_end.py
    CITYQUEST_FAST_TEST=1 python test_end_to_end.py
"""

import os
import sys
import json
import uuid
from pathlib import Path

# ── Fast-test guard ──────────────────────────────────────────────────────────
if os.environ.get("CITYQUEST_FAST_TEST"):
    print("[test] FAST TEST - skipping GGUF model download and LLM inference")
    os.environ["CITYQUEST_SKIP_MODEL"] = "1"

# ── Test tracking ────────────────────────────────────────────────────────────
passed = 0
failed = 0
errors: list[str] = []


def check(label: str, condition: bool, detail: str = ""):
    global passed, failed
    if condition:
        passed += 1
        print(f"  ✓ PASS: {label}")
    else:
        failed += 1
        msg = f"✗ FAIL: {label} — {detail}" if detail else f"✗ FAIL: {label}"
        errors.append(msg)
        print(f"  {msg}")


def main():
    global passed, failed, errors
    print("\n" + "=" * 80)
    print("CITYQUEST AI — END-TO-END PIPELINE TEST")
    print("=" * 80)

    # ── Imports ──────────────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("MODULE IMPORTS")
    print("-" * 80)

    try:
        from app.services.retrieval import load_games_dataset, normalize_game_record, retrieve_examples
        from app.services.generator import generate_game, generate_game_mock, build_generation_prompt
        from app.services.validator import validate_game, repair_game
        from app.services.schema_validator import validate_game_schema, create_minimal_game_template
        from app.services.tracing import log_event, load_events, log_generation_trace
        from app.services.journal import create_journal_entry, save_journal_entry, summarize_journal, load_journal_entries, detect_mood, assess_story_value
        from app.services.scoring import compute_scores
        from app.services.story import build_story_packet, generate_story, _template_short_recap, _template_long_summary
        from app.services.minicpm import generate_recap as minicpm_recap, generate_summary as minicpm_summary
        from app.services.image_gen import generate_poster, generate_poster_sync
        check("All service modules imported", True)
    except Exception as e:
        check("All service modules imported", False, str(e))
        # Don't continue if core imports fail
        _summary()
        sys.exit(1)

    # ── T1: Dataset loading ──────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T1: DATASET LOADING & NORMALIZATION")
    print("-" * 80)

    try:
        raw = load_games_dataset("app/data/games_dataset.json")
        check("Dataset loaded", len(raw) > 0, f"Got {len(raw)} records")
        check("Dataset has 70 records", len(raw) == 70, f"Got {len(raw)}")

        records = [normalize_game_record(r) for r in raw]
        check("Records normalized", len(records) == len(raw))

        # Verify structure of first record
        r0 = records[0]
        check("Normalized record has id", bool(r0.get("id")))
        check("Normalized record has game_type", bool(r0.get("game_type")))
        check("Normalized record has tasks list", isinstance(r0.get("tasks"), list))
        check("Normalized record has rules list", isinstance(r0.get("rules"), list))

        # Check game type distribution
        game_types = set(r["game_type"] for r in records)
        check("All game types present", game_types == {"scavenger_hunt", "hide_and_seek", "tag"},
              f"Got {game_types}")
    except Exception as e:
        check("Dataset loading", False, str(e))

    # ── T2: Retrieval ────────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T2: RETRIEVAL GROUNDING")
    print("-" * 80)

    try:
        config_adults = {
            "game_type": "scavenger_hunt",
            "city": "Paris", "area": "Le Marais",
            "location_type": "mixed", "duration_minutes": 60,
            "num_players": 4, "difficulty": "medium",
            "age_group": "adults", "energy_level": "medium",
        }

        retrieved = retrieve_examples(config_adults, records, k=3)
        check("Retrieval returns k=3 results", len(retrieved) == 3)

        # Verify structure
        check("Retrieved has id", all(r.get("id") for r in retrieved))
        check("Retrieved has retrieval_score", all("retrieval_score" in r for r in retrieved))
        check("Retrieved has rules_summary", all("rules_summary" in r for r in retrieved))
        check("Retrieved has task_patterns", all("task_patterns" in r for r in retrieved))

        # Test different config types
        config_kids = {"game_type": "hide_and_seek", "city": "Paris", "area": "Park",
                        "location_type": "park", "duration_minutes": 45, "num_players": 5,
                        "difficulty": "easy", "age_group": "kids", "energy_level": "high"}
        r_kids = retrieve_examples(config_kids, records, k=5)
        check("Kids config returns k=5", len(r_kids) == 5)

        config_teens = {"game_type": "tag", "city": "Paris", "area": "Square",
                         "location_type": "mixed", "duration_minutes": 30, "num_players": 8,
                         "difficulty": "hard", "age_group": "teens", "energy_level": "high"}
        r_teens = retrieve_examples(config_teens, records, k=3)
        check("Teens config returns k=3", len(r_teens) == 3)
    except Exception as e:
        check("Retrieval", False, str(e))

    # ── T3: Game Generation (mock) ────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T3: GAME GENERATION (MOCK PATH)")
    print("-" * 80)

    try:
        game = generate_game_mock(config_adults, retrieved)
        check("Game generated", game is not None)
        check("Game has game_id", bool(game.get("game_id")))
        check("Game has title", bool(game.get("title")))
        check("Game has theme", bool(game.get("theme")))
        check("Game has setup with city", game.get("setup", {}).get("city") == "Paris")
        check("Game has tasks list", len(game.get("tasks", [])) > 0)
        check("Game has rules list", len(game.get("rules", [])) > 0)
        check("Game has safety section", bool(game.get("safety")))
        check("Game has story_seed", bool(game.get("story_seed")))

        # Schema validation
        is_valid_schema, schema_errors = validate_game_schema(game)
        check("Game passes JSON schema", is_valid_schema, f"Errors: {schema_errors[:3]}")

        # Build generation prompt
        prompt = build_generation_prompt(config_adults, retrieved)
        check("Prompt built", len(prompt) > 100, f"Got {len(prompt)} chars")
    except Exception as e:
        check("Game generation", False, str(e))

    # ── T4: Validation ───────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T4: VALIDATION CHECKS")
    print("-" * 80)

    try:
        is_valid, failures = validate_game(game, config_adults)
        check("Generated game passes validation", is_valid, f"Failures: {failures[:3]}")

        # Test that invalid games are caught
        bad_game = {
            "game_id": "", "title": "", "theme": "",
            "setup": {}, "rules": [], "tasks": [],
            "global_hints": [], "score_rules": [],
            "tie_breaker": "", "safety": {}, "story_seed": {},
        }
        is_bad_valid, bad_failures = validate_game(bad_game, config_adults)
        check("Empty game fails validation", not is_bad_valid, f"Got {len(bad_failures)} issues")
        check("Empty game has issues", len(bad_failures) > 5)
    except Exception as e:
        check("Validation", False, str(e))

    # ── T5: Repair ───────────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T5: REPAIR PASS")
    print("-" * 80)

    try:
        bad_game = {
            "game_id": "", "title": "", "theme": "",
            "setup": {}, "rules": [], "tasks": [],
            "global_hints": [], "score_rules": [],
            "tie_breaker": "", "safety": {}, "story_seed": {},
        }
        _, failures = validate_game(bad_game, config_adults)
        repaired = repair_game(bad_game, failures, config_adults)

        check("Repair returns dict", isinstance(repaired, dict))
        check("Repaired has game_id", bool(repaired.get("game_id")), f"Got '{repaired.get('game_id')}'")
        check("Repaired has title", bool(repaired.get("title")))
        check("Repaired has tasks", len(repaired.get("tasks", [])) > 0)
        check("Repaired has rules", len(repaired.get("rules", [])) > 0)
        check("Repaired has safety zone", bool(repaired.get("safety", {}).get("allowed_zone")))

        is_repaired_valid, repaired_failures = validate_game(repaired, config_adults)
        check("Repaired game passes validation", is_repaired_valid, f"Remaining: {repaired_failures[:3]}")
    except Exception as e:
        check("Repair", False, str(e))

    # ── T6: Event Logging ────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T6: EVENT LOGGING")
    print("-" * 80)

    try:
        session_id = f"test-{uuid.uuid4().hex[:8]}"

        ev1 = log_event(session_id, "task_revealed", {"task_id": "t1", "title": "Find the mural"})
        check("Event logged", ev1 is not None)
        check("Event has event_id", bool(ev1.get("event_id")))
        check("Event has timestamp", bool(ev1.get("timestamp")))
        check("Event has payload", "task_id" in ev1.get("payload", {}))

        log_event(session_id, "task_completed", {"task_id": "t1", "summary": "Done"}, team_id="team-a")
        log_event(session_id, "task_completed", {"task_id": "t2", "summary": "Done"}, team_id="team-a")
        log_event(session_id, "hint_used", {"task_id": "t2", "summary": "Used hint"}, team_id="team-a")
        log_event(session_id, "task_skipped", {"task_id": "t3", "summary": "Skipped"}, team_id="team-a")
        log_event(session_id, "photo_uploaded", {"photo_id": "p1", "summary": "Selfie"}, team_id="team-a")
        log_event(session_id, "journal_recorded", {"journal_id": "j1", "mood": "excited"}, team_id="team-a")

        events = load_events(session_id)
        check("All events loaded", len(events) >= 6, f"Got {len(events)} events")

        event_types = [e["event_type"] for e in events]
        check("Has task_completed", "task_completed" in event_types)
        check("Has hint_used", "hint_used" in event_types)
        check("Has task_skipped", "task_skipped" in event_types)
        check("Has photo_uploaded", "photo_uploaded" in event_types)
    except Exception as e:
        check("Event logging", False, str(e))

    # ── T7: Journal ──────────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T7: JOURNAL PIPELINE")
    print("-" * 80)

    try:
        transcript = "We found the mural near the canal, it was incredible! The whole team cheered. This was definitely the highlight of our game - a close call that turned into our best moment."
        entry = create_journal_entry(
            transcript=transcript,
            session_id=session_id,
            team_id="team-a",
            task_id="t1",
            location_note="Canal area",
        )
        check("Journal entry created", entry is not None)
        check("Journal has journal_id", bool(entry.get("journal_id")))
        check("Journal has session_id", entry.get("session_id") == session_id)

        # Mood detection
        mood = detect_mood(transcript)
        check("Mood detected", bool(mood), f"Got '{mood}'")
        check("Mood is expected", mood in ("excited", "funny", "confused", "tense", "lucky", "chaotic"),
              f"Got '{mood}'")

        # Story value
        story_value = assess_story_value(transcript)
        check("Story value assessed", bool(story_value), f"Got '{story_value}'")

        # Summarization
        summary = summarize_journal(transcript, task_id="t1")
        check("Summary has moment_summary", bool(summary.get("moment_summary")))
        check("Summary has tags", len(summary.get("tags", [])) > 0)
        check("Summary has story_value", bool(summary.get("story_value")))

        entry.update(summary)
        save_journal_entry(entry)

        journals = load_journal_entries(session_id)
        check("Journal persisted", len(journals) >= 1, f"Got {len(journals)} entries")
    except Exception as e:
        check("Journal pipeline", False, str(e))

    # ── T8: Scoring ──────────────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T8: DETERMINISTIC SCORING")
    print("-" * 80)

    try:
        events = load_events(session_id)
        scores = compute_scores(events, game)

        check("Scores returned", scores is not None)
        check("Has team_scores", len(scores.get("team_scores", [])) > 0)
        check("Has winner", bool(scores.get("winner")))
        check("Has scoring_explanation", len(scores.get("scoring_explanation", [])) > 0)

        ts = scores["team_scores"][0]
        check("Team score has points", ts.get("points", 0) >= 0)
        check("Team score has completed_tasks", ts.get("completed_tasks", 0) >= 0)
        check("Team score has hints_used", ts.get("hints_used", 0) >= 0)
        check("Team score has total_tasks", ts.get("total_tasks", 0) > 0)

        # Verify hint penalty
        check("Hint penalty applied", ts["hints_used"] > 0)

        # Scoring breakdown is present
        has_breakdown = any(s.get("scoring_breakdown") for s in scores["team_scores"])
        check("Has scoring breakdown", has_breakdown)
    except Exception as e:
        check("Scoring", False, str(e))

    # ── T9: Story Recap (template) ──────────────────────────────────────
    print("\n" + "-" * 80)
    print("T9: STORY RECAP (TEMPLATE FALLBACK)")
    print("-" * 80)

    try:
        events = load_events(session_id)
        packet = build_story_packet(
            game=game,
            events=events,
            scores=scores,
            journal_entries=load_journal_entries(session_id),
        )
        check("Story packet built", packet is not None)
        check("Packet has game_info", bool(packet.get("game_info")))
        check("Packet has winner", bool(packet.get("winner")))
        check("Packet has final_scores", len(packet.get("final_scores", [])) > 0)
        check("Packet has task_outcomes", len(packet.get("task_outcomes", [])) > 0)

        # Template recap
        short = _template_short_recap(packet)
        check("Template short recap generated", len(short) > 50, f"Got {len(short)} chars")

        long = _template_long_summary(packet)
        check("Template long summary generated", len(long) > 100, f"Got {len(long)} chars")

        # Full generate_story with fallback
        story = generate_story(packet, session_id=session_id)
        check("generate_story() returns dict", isinstance(story, dict))
        check("Has short_recap", bool(story.get("short_recap")))
        check("Has long_summary", bool(story.get("long_summary")))
        check("Has poster_prompt", bool(story.get("poster_prompt")))
        check("Has story_packet", bool(story.get("story_packet")))
    except Exception as e:
        check("Story recap", False, str(e))

    # ── T10: MiniCPM module ─────────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T10: MINICPM MODULE (lazy-load only — no download in fast test)")
    print("-" * 80)

    try:
        from app.services.minicpm import MINICPM_MODEL_ID, MINICPM_GGUF_FILE, generate_recap, generate_summary

        check("MiniCPM constants loaded", bool(MINICPM_MODEL_ID) and bool(MINICPM_GGUF_FILE))
        check("MiniCPM model ID is correct", "MiniCPM" in MINICPM_MODEL_ID)
        check("MiniCPM GGUF file ends with .gguf", MINICPM_GGUF_FILE.endswith(".gguf"))

        # In fast test, skip actual model loading
        if os.environ.get("CITYQUEST_FAST_TEST"):
            check("MiniCPM module imports (fast test — model skipped)", True)
        else:
            # Try recapping — will likely fail without model, that's OK
            packet = build_story_packet(
                game=game, events=events, scores=scores,
                journal_entries=load_journal_entries(session_id),
            )
            result = generate_recap(packet)
            # May be None if no model, but must not raise
            check("MiniCPM generate_recap() runs without exception", True)
    except Exception as e:
        check("MiniCPM module", False, str(e))

    # ── T11: Image Gen module ───────────────────────────────────────────
    print("\n" + "-" * 80)
    print("T11: IMAGE GEN MODULE")
    print("-" * 80)

    try:
        from app.services.image_gen import generate_poster, generate_poster_sync, FLUX_MODEL_ID

        check("Image gen constants loaded", bool(FLUX_MODEL_ID))

        # Test with skip-model env var — must return None gracefully
        result = generate_poster("A test poster prompt")
        check("generate_poster() with skip returns None (no model loaded)", result is None)

        path_or_none, status = generate_poster_sync("test-session", "A test poster prompt")
        check("generate_poster_sync() returns tuple", isinstance(path_or_none, type(None)) and isinstance(status, str))
        check("Status message is informative", "unavailable" in status or "Failed" in status)
    except Exception as e:
        check("Image gen module", False, str(e))

    # ── Summary ──────────────────────────────────────────────────────────
    _summary()


def _summary():
    total = passed + failed
    print(f"\nRESULTS: {passed}/{total} tests passed", end="")
    if failed > 0:
        print(f"{passed}/{total} tests passed - ALL CLEAR")
    else:
        print(" — ALL PASSED! 🎉")
    print("=" * 80)

    if errors:
        print("\nFailed tests:")
        for e in errors:
            print(f"  {e}")

    # Exit with error code if any test failed
    if failed > 0:
        sys.exit(1)


if __name__ == "__main__":
    main()