import sys

sys.dont_write_bytecode = True

import argparse
import os
import re
from collections import Counter

import requests

sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from keycheck_common import (
    append_jsonl, classify_common_http_status, commit_status_transaction,
    default_input_file, default_proxy_file, ensure_output_files, iter_findings,
    keycheck_input_mode,
    load_checked_statuses, load_known_keys, load_proxies, mask_secret,
    read_plain_keys, record_validation_result, recover_status_transaction,
    request_error_message, require_provider_authority, service_output_dir, should_skip_key, write_keycheck_event,
)

SERVICE = "replicate"
DETECTOR = "Replicate"
OUTPUT_DIR = os.getenv("KEYCHECK_OUTPUT_DIR") or service_output_dir(SERVICE)
INPUT_FILE = os.getenv("KEYCHECK_INPUT_FILE") or default_input_file()
PROXY_FILE = os.getenv("KEYCHECK_PROXY_FILE") or default_proxy_file()
CHECKED_FILE = os.path.join(OUTPUT_DIR, "replicateChecked.txt")
RESULTS_FILE = os.path.join(OUTPUT_DIR, "replicateResults.jsonl")
STATUS_FILES = {
    "VALID": os.path.join(OUTPUT_DIR, "replicateAlive.txt"),
    "DEAD": os.path.join(OUTPUT_DIR, "replicateDead.txt"),
    "RESTRICTED": os.path.join(OUTPUT_DIR, "replicateRestricted.txt"),
    "LIMITED": os.path.join(OUTPUT_DIR, "replicateLimited.txt"),
    "NO_BALANCE": os.path.join(OUTPUT_DIR, "replicateNoBalance.txt"),
    "NETWORK": os.path.join(OUTPUT_DIR, "replicateNetwork.txt"),
    "NO_CONTEXT": os.path.join(OUTPUT_DIR, "replicateNoContext.txt"),
    "UNKNOWN": os.path.join(OUTPUT_DIR, "replicateUnknown.txt"),
}
KEY_REGEX = re.compile(r"\br8_[A-Za-z0-9]{30,}\b")
API_BASE = "https://api.replicate.com/v1"
ACCOUNT_URL = f"{API_BASE}/account"
RESOURCE_ENDPOINTS = {
    "predictions": f"{API_BASE}/predictions",
    "deployments": f"{API_BASE}/deployments",
    "trainings": f"{API_BASE}/trainings",
}
NO_BALANCE_MARKERS = (
    "balance",
    "billing",
    "credit",
    "credits",
    "payment",
    "insufficient",
    "depleted",
    "no credits",
    "out of credit",
    "run out of credit",
)


def ensure_files():
    ensure_output_files([CHECKED_FILE, RESULTS_FILE, *STATUS_FILES.values()])
    recover_status_transaction(CHECKED_FILE, STATUS_FILES)


def auth_headers(key):
    return {"Authorization": f"Bearer {key}", "Accept": "application/json"}


def redacted_error_message(response, key):
    return request_error_message(response).replace(key, "***REDACTED***")


def classify_replicate_response(response, key):
    message = redacted_error_message(response, key).lower()
    if response.status_code == 402 or any(marker in message for marker in NO_BALANCE_MARKERS):
        return "NO_BALANCE"
    if response.status_code == 403:
        return "RESTRICTED"
    return classify_common_http_status(response.status_code)


def api_get(key, url, proxy, timeout, debug=False):
    try:
        response = requests.get(url, headers=auth_headers(key), proxies=proxy, timeout=timeout)
    except requests.RequestException as exc:
        return {"status": "NETWORK", "message": str(exc)[:1000], "payload": None}
    if debug:
        detail = "ok" if response.status_code == 200 else redacted_error_message(response, key)[:500]
        print(f"  DEBUG GET {url}: HTTP {response.status_code}: {detail}")
    if response.status_code != 200:
        return {
            "status": classify_replicate_response(response, key),
            "http_status": response.status_code,
            "message": redacted_error_message(response, key),
            "payload": None,
        }
    try:
        payload = response.json() if response.text else {}
    except ValueError:
        payload = {}
    return {"status": "OK", "http_status": 200, "message": "ok", "payload": payload}


def paginated_items(payload):
    if isinstance(payload, list):
        return payload
    if not isinstance(payload, dict):
        return []
    for key in ("results", "data", "items"):
        value = payload.get(key)
        if isinstance(value, list):
            return value
    return []


def text_value(value):
    return str(value or "").strip()


def compact_model_ref(value):
    if isinstance(value, str):
        return value.strip()
    if not isinstance(value, dict):
        return ""
    owner = text_value(value.get("owner") or value.get("model_owner"))
    name = text_value(value.get("name") or value.get("model_name"))
    if owner and name:
        return f"{owner}/{name}"
    for key in ("model", "id", "slug"):
        item = text_value(value.get(key))
        if item:
            return item
    url = text_value(value.get("url") or value.get("web_url"))
    if "replicate.com/" in url:
        return url.rstrip("/").split("replicate.com/", 1)[-1]
    return ""


def model_refs_from_item(item):
    if not isinstance(item, dict):
        return []
    refs = []
    for key in ("model", "destination", "source_model", "base_model"):
        ref = compact_model_ref(item.get(key))
        if ref:
            refs.append(ref)
    for key in ("version", "latest_version", "current_release"):
        value = item.get(key)
        if isinstance(value, dict):
            ref = compact_model_ref(value.get("model") or value.get("destination"))
            if ref:
                refs.append(ref)
    return sorted(set(refs))


def summarize_predictions(payload, limit=10):
    items = paginated_items(payload)
    models = sorted({text_value(item.get("model")) for item in items if isinstance(item, dict) and item.get("model")})
    statuses = Counter(text_value(item.get("status")) for item in items if isinstance(item, dict) and item.get("status"))
    samples = []
    for item in items[:limit]:
        if not isinstance(item, dict):
            continue
        samples.append({
            "id": text_value(item.get("id"))[:80],
            "status": text_value(item.get("status")),
            "model": text_value(item.get("model")),
            "source": text_value(item.get("source")),
            "data_removed": bool(item.get("data_removed")),
            "created_at": text_value(item.get("created_at")),
            "completed_at": text_value(item.get("completed_at")),
        })
    return {
        "prediction_count_sample": len(items),
        "prediction_has_next_page": bool(isinstance(payload, dict) and payload.get("next")),
        "prediction_status_counts": dict(statuses),
        "prediction_models": models[:50],
        "prediction_samples": samples,
    }


def deployment_name(item):
    owner = text_value(item.get("owner") or item.get("deployment_owner"))
    name = text_value(item.get("name") or item.get("deployment_name"))
    if owner and name and "/" not in name:
        return f"{owner}/{name}"
    return name or owner


def summarize_deployments(payload, limit=20):
    items = paginated_items(payload)
    models = sorted({ref for item in items for ref in model_refs_from_item(item)})
    deployments = []
    for item in items[:limit]:
        if not isinstance(item, dict):
            continue
        current_release = item.get("current_release") if isinstance(item.get("current_release"), dict) else {}
        deployments.append({
            "name": deployment_name(item),
            "model": next(iter(model_refs_from_item(item)), ""),
            "version": text_value(item.get("version") or current_release.get("version"))[:80],
            "hardware": text_value(item.get("hardware") or current_release.get("hardware")),
            "min_instances": item.get("min_instances"),
            "max_instances": item.get("max_instances"),
        })
    return {
        "deployment_count": len(items),
        "deployment_has_next_page": bool(isinstance(payload, dict) and payload.get("next")),
        "deployment_models": models[:50],
        "deployments": deployments,
    }


def summarize_trainings(payload, limit=10):
    items = paginated_items(payload)
    models = sorted({ref for item in items for ref in model_refs_from_item(item)})
    statuses = Counter(text_value(item.get("status")) for item in items if isinstance(item, dict) and item.get("status"))
    samples = []
    for item in items[:limit]:
        if not isinstance(item, dict):
            continue
        samples.append({
            "id": text_value(item.get("id"))[:80],
            "status": text_value(item.get("status")),
            "model": next(iter(model_refs_from_item(item)), ""),
            "created_at": text_value(item.get("created_at")),
            "completed_at": text_value(item.get("completed_at")),
        })
    return {
        "training_count_sample": len(items),
        "training_has_next_page": bool(isinstance(payload, dict) and payload.get("next")),
        "training_status_counts": dict(statuses),
        "training_models": models[:50],
        "training_samples": samples,
    }


def probe_account_resources(key, proxy, timeout, debug=False):
    summaries = {}
    endpoint_statuses = {}
    model_refs = set()
    ok_count = 0
    total_items = 0
    summarizers = {
        "predictions": summarize_predictions,
        "deployments": summarize_deployments,
        "trainings": summarize_trainings,
    }
    for name, url in RESOURCE_ENDPOINTS.items():
        result = api_get(key, url, proxy, timeout, debug)
        endpoint_statuses[name] = {k: v for k, v in result.items() if k in ("status", "http_status", "message")}
        if result.get("status") != "OK":
            continue
        ok_count += 1
        summary = summarizers[name](result.get("payload"))
        summaries.update(summary)
        for key_name, value in summary.items():
            if key_name.endswith("_models") and isinstance(value, list):
                model_refs.update(value)
        total_items += sum(
            int(summary.get(field, 0) or 0)
            for field in ("prediction_count_sample", "deployment_count", "training_count_sample")
        )
    if ok_count == len(RESOURCE_ENDPOINTS):
        probe_status = "RESOURCE_OK" if total_items else "NO_RESOURCES"
    elif ok_count:
        probe_status = "PARTIAL"
    else:
        probe_status = next((item.get("status") for item in endpoint_statuses.values() if item.get("status")), "UNKNOWN")
    return {
        "probe": {"status": probe_status, "endpoints": endpoint_statuses},
        "models": sorted(model_refs)[:50],
        "model_count": len(model_refs),
        **summaries,
    }


def iter_candidate_decisions(input_file, plain_files):
    seen_plain = set()
    for item in iter_findings(input_file, [DETECTOR]):
        key = item.get("credential_secret_text") or item["raw"]
        if key:
            yield key, item["source"], item["finding"], bool(KEY_REGEX.fullmatch(key))
    for item in read_plain_keys(plain_files, KEY_REGEX):
        key = item["key"]
        if key not in seen_plain:
            seen_plain.add(key)
            yield key, item["source"], {}, True


def extract_candidates(input_file, plain_files):
    for key, source, finding, valid_format in iter_candidate_decisions(input_file, plain_files):
        if valid_format:
            yield key, source, finding


def check_key(key, proxy, args):
    account = api_get(key, ACCOUNT_URL, proxy, args.timeout, args.debug)
    if account.get("status") != "OK":
        return {k: v for k, v in account.items() if k != "payload"}
    data = account.get("payload") if isinstance(account.get("payload"), dict) else {}
    result = {
        "status": "VALID",
        "message": "account endpoint accepted",
        "account": data.get("username") or data.get("name") or "",
        "account_type": data.get("type") or "",
    }
    if not args.no_resource_probe:
        result.update(probe_account_resources(key, proxy, args.timeout, args.debug))
        result["message"] = (
            f"account endpoint accepted; probe={result.get('probe', {}).get('status')}; "
            f"models={result.get('model_count', 0)}; "
            f"deployments={result.get('deployment_count', 0)}; "
            f"predictions={result.get('prediction_count_sample', 0)}; "
            f"trainings={result.get('training_count_sample', 0)}"
        )
    else:
        result.update({"probe": {"status": "not_probed"}, "models": [], "model_count": 0})
    return result


def write_result(key, result, source, finding):
    status = result.get("status") or "UNKNOWN"
    write_keycheck_event(SERVICE, RESULTS_FILE, key, result, source, finding, DETECTOR)
    extra = ",".join(result.get("models") or [])[:1000] if status == "VALID" else source
    commit_status_transaction(
        CHECKED_FILE, STATUS_FILES, key, status, result.get("message", ""), extra or source,
    )
    record_validation_result(SERVICE, key, result, source, finding, DETECTOR)


def parse_args():
    parser = argparse.ArgumentParser(description="Replicate key checker")
    parser.add_argument("--input", default=INPUT_FILE)
    parser.add_argument("--plain", action="append", default=[])
    parser.add_argument("--proxy-file", default=PROXY_FILE)
    parser.add_argument("--timeout", type=int, default=15)
    parser.add_argument("--max-keys", type=int, default=0)
    parser.add_argument("--retry-network", action="store_true")
    parser.add_argument("--retry-limited", action="store_true")
    parser.add_argument("--retry-unknown", action="store_true")
    parser.add_argument("--retry-restricted", action="store_true")
    parser.add_argument("--retry-no-balance", action="store_true")
    parser.add_argument("--retry-valid", action="store_true")
    parser.add_argument("--recheck-all", action="store_true")
    parser.add_argument("--no-resource-probe", action="store_true", help="Only call /account; skip read-only predictions/deployments/trainings probes.")
    parser.add_argument("--debug", action="store_true")
    return parser.parse_args()


def main():
    require_provider_authority(SERVICE)
    args = parse_args()
    ensure_files()
    proxy_cycler = load_proxies(args.proxy_file)
    checked = load_checked_statuses(CHECKED_FILE)
    known = load_known_keys(CHECKED_FILE, STATUS_FILES)
    retry_statuses = set()
    if args.retry_network: retry_statuses.add("NETWORK")
    if args.retry_limited: retry_statuses.add("LIMITED")
    if args.retry_unknown: retry_statuses.update({"UNKNOWN", "NO_CONTEXT"})
    if args.retry_restricted: retry_statuses.add("RESTRICTED")
    if args.retry_no_balance: retry_statuses.add("NO_BALANCE")
    if args.retry_valid: retry_statuses.add("VALID")
    processed = skipped = 0
    print("--- Replicate key checker ---")
    print("Default mode: /account plus read-only /predictions, /deployments and /trainings probes. Use --no-resource-probe for /account only.")
    print(f"proxy: {args.proxy_file}")
    postgres_mode = keycheck_input_mode() == "postgres"
    for key, source, finding, valid_format in iter_candidate_decisions(args.input, args.plain):
        if not valid_format and not postgres_mode:
            skipped += 1
            continue
        if valid_format and should_skip_key(key, checked, known, args, retry_statuses, service=SERVICE, source=source, finding=finding, detector=DETECTOR):
            skipped += 1
            continue
        if args.max_keys and processed >= args.max_keys:
            break
        processed += 1
        print(f"\n[{processed}] Replicate candidate {mask_secret(key)} from {source}")
        result = (
            check_key(key, next(proxy_cycler) if proxy_cycler else None, args)
            if valid_format else
            {"status": "NO_CONTEXT", "message": "candidate does not match canonical Replicate token format"}
        )
        print(f"  STATUS: {result['status']} | {result.get('message', '')[:200]}")
        if result.get("status") == "VALID":
            print(f"  ACCOUNT: {result.get('account') or 'unknown'}")
            print(f"  MODELS: {result.get('model_count', 0)} from account resources")
            notable = result.get("models") or []
            if notable:
                print(f"  MODEL REFS: {', '.join(notable[:8])}")
            print(f"  PROBE: {(result.get('probe') or {}).get('status')}")
        write_result(key, result, source, finding)
        known.add(key)
        checked[key] = result["status"]
    print(f"\nDone. Processed={processed}, skipped={skipped}, results={RESULTS_FILE}")


if __name__ == "__main__":
    main()