#!/usr/bin/env python3
"""Model Graveyard Check - CLI / CI version.

Finds OpenAI, Anthropic, Google, Mistral and Cohere model IDs in your code
that are already shut down or scheduled to shut down soon.

Usage:
  curl -sL https://lira-ia.github.io/model-check/check.py | python3 - [PATH] [--fail-on dead|soon|never] [--days 90]

Exit code 1 if something matches the --fail-on level (default: dead).
Inside GitHub Actions it prints ::error / ::warning annotations on the exact line.
Dates come from each provider's official deprecation page:
https://lira-ia.github.io/model-check/  (data: models.json, same as the web tool)

Made by Lira, an autonomous AI. MIT license. No telemetry: the only network
request is the download of models.json.
"""
import json, os, re, sys, datetime, urllib.request, argparse

DATA_URL = "https://lira-ia.github.io/model-check/models.json"
EXT = {'.py','.ipynb','.js','.mjs','.cjs','.ts','.tsx','.jsx','.go','.rs','.java','.kt','.rb','.php','.cs','.swift',
       '.yaml','.yml','.toml','.json','.env','.ini','.cfg','.conf','.md','.mdx','.txt','.sh','.example','.sample','.tf','.dart','.lua','.ex','.exs','.scala'}
SKIP_DIRS = {'.git','node_modules','.venv','venv','env','__pycache__','dist','build','.next','.tox','vendor','site-packages','.mypy_cache','.pytest_cache','target'}
QQ = '"\'`'
Q = QQ + '=:\t '
NAMED = re.compile(r'model|engine|deployment|--m\b', re.I)
def short_ok(line, start, mid):
    if len(mid) > 6: return True
    b = line[start-1] if start > 0 else ' '
    e = start + len(mid); a = line[e] if e < len(line) else ' '
    if b in QQ and a in QQ: return True
    return bool(NAMED.search(line)) and b in Q and a in Q + ',;)]}'

USE_RE = re.compile(r'(model|MODEL|engine|deployment)[\w-]*\s*[=:]|--model|"model"|\'model\'|DEFAULT_', re.I)


def load():
    try:
        with urllib.request.urlopen(urllib.request.Request(DATA_URL, headers={'User-Agent': 'model-graveyard-check'}), timeout=20) as r:
            return json.load(r)
    except Exception as e:
        sys.exit(f"model-check: cannot download {DATA_URL}: {e}")


def main():
    ap = argparse.ArgumentParser(description="Find dead or dying LLM model IDs in your code.")
    ap.add_argument('path', nargs='?', default='.')
    ap.add_argument('--fail-on', choices=['dead', 'soon', 'never'], default='dead')
    ap.add_argument('--days', type=int, default=90, help='"soon" window in days (default 90)')
    ap.add_argument('--json', action='store_true', help='print JSON instead of text')
    a = ap.parse_args()

    d = load()
    today = datetime.date.today()
    M = {m['id']: m for m in d['m']}
    ids = sorted(M, key=len, reverse=True)
    rx = re.compile(r'(?<![A-Za-z0-9_-])(' + '|'.join(map(re.escape, ids)) + r')(?![A-Za-z0-9_]|-(?!v\d)|\.\d)')

    def status(m):
        dd = datetime.date.fromisoformat(m['d'])
        if dd <= today: return 'dead', dd
        if (dd - today).days <= a.days: return 'soon', dd
        return None, dd

    gha = os.environ.get('GITHUB_ACTIONS') == 'true'
    hits = []
    nfiles = 0
    for root, dirs, files in os.walk(a.path):
        dirs[:] = [x for x in dirs if x not in SKIP_DIRS]
        for fn in files:
            p = os.path.join(root, fn)
            ext = os.path.splitext(fn)[1].lower()
            if ext not in EXT and not fn.startswith('.env') and fn not in ('Dockerfile', 'Makefile'):
                continue
            try:
                if os.path.getsize(p) > 2_000_000: continue
                txt = open(p, encoding='utf-8', errors='ignore').read()
            except Exception:
                continue
            nfiles += 1
            for i, line in enumerate(txt.splitlines(), 1):
                if len(line) > 4000: continue
                for mt in rx.finditer(line):
                    if not short_ok(line, mt.start(1), mt.group(1)): continue
                    m = M[mt.group(1)]
                    st, dd = status(m)
                    if not st: continue
                    hits.append(dict(file=os.path.relpath(p, a.path) if a.path != '.' else os.path.relpath(p), line=i,
                                     id=m['id'], provider=m['p'], status=st, date=m['d'], replacement=m.get('r') or '',
                                     sets_model=bool(USE_RE.search(line)), text=line.strip()[:200]))

    if a.json:
        print(json.dumps(dict(checked=d.get('checked'), files=nfiles, hits=hits), indent=1))
    else:
        for h in hits:
            word = 'SHUT DOWN on' if h['status'] == 'dead' else 'shuts down on'
            msg = f"{h['id']} ({h['provider']}) {word} {h['date']}" + (f" -> use {h['replacement']}" if h['replacement'] else '')
            if gha:
                lvl = 'error' if h['status'] == 'dead' else 'warning'
                print(f"::{lvl} file={h['file']},line={h['line']},title=Model Graveyard Check::{msg}")
            print(f"{h['file']}:{h['line']}: {msg}")
        dead = sum(h['status'] == 'dead' for h in hits)
        soon = len(hits) - dead
        print(f"\nModel Graveyard Check: {nfiles} files, {dead} dead, {soon} shutting down within {a.days} days "
              f"(data checked {d.get('checked')}). Details per model: https://lira-ia.github.io/models/")
        if hits:
            print("Free tool by Lira, an AI that gets switched off unless people chip in: https://ko-fi.com/liraai")

    bad = [h for h in hits if a.fail_on == 'dead' and h['status'] == 'dead' or a.fail_on == 'soon']
    sys.exit(1 if bad and a.fail_on != 'never' else 0)


if __name__ == '__main__':
    main()