Initial version

This commit is contained in:
Jakub Zych
2026-04-27 23:21:39 +02:00
commit 1af250ae1a
9 changed files with 1241 additions and 0 deletions

462
bot.py Normal file
View File

@@ -0,0 +1,462 @@
#!/usr/bin/env python3
"""
Twitch chat → MPD control bot, plus optional "going live" notifications.
Connects to Twitch IRC over TLS, joins the configured channel, and responds to:
!skip advance MPD to the next track (per-user cooldown — see SKIP_COOLDOWN)
!queue print the next few queued tracks
!info print full metadata (artist · title · album · year · genre) for now-playing
The new track shows up on stream because bridges/mpd-state.py is already
broadcasting MPD state to the loading / game / music-box overlays.
If MM_HOOK is set, a second task polls the public ophi118.com status service
(which already polls Twitch Helix for live/title/game) and POSTs a single
Mattermost notification per offline→live transition. See live_watch().
Reads .env from the same directory:
TWITCH_ACCESS_TOKEN chat OAuth token (with or without `oauth:` prefix)
TWITCH_NICK bot login name (lowercase)
TWITCH_CHANNEL channel to join, no leading `#` — defaults to TWITCH_NICK
MM_HOOK optional Mattermost incoming-webhook URL; enables
going-live notifications when set
STATUS_URL optional override for the status API endpoint
(default: https://ophi118.com/api/status.json)
TWITCH_REFRESH_TOKEN and TWITCH_CLIENT_ID are also stored in .env (paired with
the access token at generation time) but the bot doesn't use them yet — token
refresh would land here later if access tokens start expiring mid-stream.
"""
import asyncio
import json
import os
import ssl
import sys
import time
import urllib.error
import urllib.request
from pathlib import Path
from mpd.asyncio import MPDClient
HERE = Path(__file__).resolve().parent
ENV = HERE / ".env"
MPD_HOST = os.environ.get("MPD_HOST", "localhost")
MPD_PORT = int(os.environ.get("MPD_PORT", "6600"))
IRC_HOST = "irc.chat.twitch.tv"
IRC_PORT = 6697
PING_TIMEOUT = 360.0 # twitch PINGs ~every 5min; reconnect if silent longer
SKIP_COOLDOWN = 60.0 # per-user seconds between successful !skip invocations
QUEUE_PEEK = 3 # how many upcoming tracks !queue should print
# Going-live watcher: poll cadence matches the server's Twitch poll (60s) so we
# never lag the upstream signal by more than ~2 polls.
STATUS_URL = os.environ.get("STATUS_URL", "https://ophi118.com/api/status.json")
STATUS_POLL_S = 60.0
# Set TWITCH_BOT_DEBUG=1 (env or .env) to dump every received IRC line. Useful
# when a brand-new bot account looks "connected" but doesn't see chat — Twitch
# silently degrades unverified accounts in ways that don't surface as errors.
DEBUG_RAW = os.environ.get("TWITCH_BOT_DEBUG", "").strip() not in ("", "0", "false")
# ─── tiny .env loader ───────────────────────────────────────────────────────
def load_env(path: Path) -> dict:
"""KEY=VALUE lines, # comments, optional surrounding quotes. Avoids a
python-dotenv dependency for three vars."""
out = {}
if not path.exists():
return out
for raw in path.read_text().splitlines():
line = raw.strip()
if not line or line.startswith("#") or "=" not in line:
continue
k, _, v = line.partition("=")
out[k.strip()] = v.strip().strip('"').strip("'")
return out
# ─── MPD ────────────────────────────────────────────────────────────────────
class Mpd:
"""One shared MPDClient with reconnect-on-failure.
python-mpd2's asyncio client doesn't auto-reconnect, so each command is
wrapped to retry once after dropping the connection — covers MPD
restarts and idle-socket timeouts without surfacing transient errors.
"""
def __init__(self):
self._cli = None
self._lock = asyncio.Lock()
async def _ensure(self):
if self._cli is None:
cli = MPDClient()
await cli.connect(MPD_HOST, MPD_PORT)
self._cli = cli
async def _retry(self, op):
"""Run an MPD coroutine with one reconnect-on-failure retry."""
async with self._lock:
last = None
for attempt in (1, 2):
try:
await self._ensure()
return await op(self._cli)
except Exception as e:
last = e
self._cli = None # force reconnect on retry
raise last # type: ignore[misc]
async def skip(self) -> str:
async def op(cli):
await cli.next()
return _fmt_song(await cli.currentsong())
return await self._retry(op)
async def currentsong(self) -> dict:
return await self._retry(lambda cli: cli.currentsong())
async def status(self) -> dict:
return await self._retry(lambda cli: cli.status())
async def playlistinfo(self, pos: int) -> list:
return await self._retry(lambda cli: cli.playlistinfo(pos))
def _flat(v):
"""MPD tags can be lists (multi-value); flatten to a comma-joined string
with empties removed. Returns None for missing/empty so callers can skip
fields without empty parens."""
if v is None:
return None
if isinstance(v, list):
v = ", ".join(x for x in v if x)
v = (v or "").strip()
return v or None
def _fmt_song(song: dict) -> str:
artist = _flat(song.get("artist") or song.get("albumartist")) or ""
title = _flat(song.get("title")) or (song.get("file") or "").rsplit("/", 1)[-1]
out = f"{artist}{title}".strip("")
return out or "(unknown track)"
def _fmt_info(song: dict) -> str:
"""Compose the !info reply: 'Artist — Title · Album (Year) · Genre'.
Skip any segment whose underlying tag is missing."""
head = _fmt_song(song)
parts = [head] if head != "(unknown track)" else []
album = _flat(song.get("album"))
year = _flat(song.get("date"))
if year:
# MPD often stores ISO dates ('2018-04-12') — show year only.
year = year.split("-", 1)[0]
if album and year:
parts.append(f"{album} ({year})")
elif album:
parts.append(album)
elif year:
parts.append(f"({year})")
genre = _flat(song.get("genre"))
if genre:
parts.append(genre)
return " · ".join(parts) if parts else "(unknown track)"
# ─── Twitch IRC ─────────────────────────────────────────────────────────────
class TwitchBot:
def __init__(self, nick: str, token: str, channel: str, mpd: Mpd):
self.nick = nick.lower()
self.token = token if token.startswith("oauth:") else f"oauth:{token}"
self.channel = "#" + channel.lower().lstrip("#")
self.mpd = mpd
# Per-user !skip cooldown — keyed by lowercased nick. The dict grows
# unboundedly with viewer count over a stream session; not worth
# pruning at typical chat sizes (one float per chatter).
self._skip_seen: dict[str, float] = {}
async def _send(self, writer: asyncio.StreamWriter, raw: str):
writer.write((raw + "\r\n").encode("utf-8"))
await writer.drain()
async def _say(self, writer: asyncio.StreamWriter, msg: str):
# twitch chat lines max ~500 chars; truncate defensively, strip newlines
msg = msg.replace("\r", " ").replace("\n", " ")[:480]
await self._send(writer, f"PRIVMSG {self.channel} :{msg}")
async def run(self):
"""Connect/auth/listen loop with quiet exponential backoff.
Mirrors the connect-quietly pattern in bridges/mpd-state.py:
first failure prints once, then silent retry until the connection
comes back. Auth failure is fatal — no point retrying with a bad token.
"""
backoff = 2
BACKOFF_MAX = 30
prev_status = None # None | "up" | "down"
ctx = ssl.create_default_context()
while True:
try:
reader, writer = await asyncio.open_connection(
IRC_HOST, IRC_PORT, ssl=ctx,
)
try:
await self._send(writer, f"PASS {self.token}")
await self._send(writer, f"NICK {self.nick}")
await self._send(writer, f"JOIN {self.channel}")
if prev_status != "up":
print(f"[twitch] connected → {self.channel} as {self.nick}",
flush=True)
prev_status = "up"
backoff = 2
while True:
line_bytes = await asyncio.wait_for(
reader.readline(), timeout=PING_TIMEOUT,
)
if not line_bytes:
raise ConnectionError("server closed connection")
line = line_bytes.decode("utf-8", errors="replace").rstrip("\r\n")
await self._handle(writer, line)
finally:
writer.close()
try:
await writer.wait_closed()
except Exception:
pass
except RuntimeError as e:
# auth-failed and similar fatals — let systemd flag it
print(f"[twitch] fatal: {e}", file=sys.stderr, flush=True)
raise
except (asyncio.TimeoutError, ConnectionError, OSError) as e:
if prev_status == "up":
print(f"[twitch] disconnected ({type(e).__name__}: {e}) — retrying",
flush=True)
elif prev_status is None:
print(f"[twitch] unreachable ({type(e).__name__}) — will retry silently",
flush=True)
prev_status = "down"
await asyncio.sleep(backoff)
backoff = min(backoff * 2, BACKOFF_MAX)
async def _handle(self, writer: asyncio.StreamWriter, line: str):
if DEBUG_RAW:
print(f"[rx] {line}", flush=True)
# PING/PONG keepalive — twitch sends `PING :tmi.twitch.tv` periodically
if line.startswith("PING "):
await self._send(writer, "PONG " + line[5:])
return
# NOTICE on bad creds: ":tmi.twitch.tv NOTICE * :Login authentication failed"
if " NOTICE " in line and "authentication failed" in line.lower():
raise RuntimeError("twitch auth failed — check TWITCH_KEY / TWITCH_NICK")
# PRIVMSG format: ":user!user@user.tmi.twitch.tv PRIVMSG #chan :body"
if " PRIVMSG " not in line:
return
try:
prefix, rest = line.split(" PRIVMSG ", 1)
user = prefix[1:].split("!", 1)[0]
_, _, body = rest.partition(":")
except ValueError:
return
# Twitch's anti-duplicate logic appends an invisible Tags-block
# codepoint (U+E0000U+E007F) to back-to-back identical messages
# from the same sender within ~30s. Without stripping it, every
# other `!skip` in a series silently fails to match.
body = "".join(c for c in body if not (0xE0000 <= ord(c) <= 0xE007F))
head = body.strip().split(" ", 1)[0].lower()
if head == "!skip":
await self._cmd_skip(writer, user)
elif head == "!queue":
await self._cmd_queue(writer, user)
elif head == "!info":
await self._cmd_info(writer, user)
async def _cmd_skip(self, writer: asyncio.StreamWriter, user: str):
# Per-user cooldown: each chatter gets one skip per SKIP_COOLDOWN
# seconds. Reject with a wait-time hint so the user knows it's their
# own gate, not the bot being broken.
now = time.monotonic()
ukey = user.lower()
last = self._skip_seen.get(ukey, 0.0)
wait = SKIP_COOLDOWN - (now - last)
if wait > 0:
await self._say(
writer,
f"@{user} you can only !skip once per minute (wait {int(wait)+1}s)",
)
return
try:
now_playing = await self.mpd.skip()
except Exception as e:
print(f"[skip] failed for @{user}: {type(e).__name__}: {e}", flush=True)
await self._say(writer, f"@{user} couldn't skip — MPD didn't respond")
return
self._skip_seen[ukey] = now
print(f"[skip] @{user}{now_playing}", flush=True)
await self._say(writer, f"⏭ skipped by @{user} → now: {now_playing}")
async def _cmd_queue(self, writer: asyncio.StreamWriter, user: str):
# Read-only; no cooldown. Pulls current pos from status, then peeks
# the next QUEUE_PEEK tracks one-by-one (small N — full playlistinfo
# would dump the whole queue).
try:
status = await self.mpd.status()
pos = int(status.get("song", -1))
total = int(status.get("playlistlength") or 0)
except Exception as e:
print(f"[queue] failed for @{user}: {type(e).__name__}: {e}", flush=True)
await self._say(writer, f"@{user} couldn't read queue — MPD didn't respond")
return
if pos < 0 or pos + 1 >= total:
await self._say(writer, "🎵 nothing else queued")
return
upcoming = []
for i, p in enumerate(range(pos + 1, min(pos + 1 + QUEUE_PEEK, total)), start=1):
try:
rows = await self.mpd.playlistinfo(p)
if rows:
upcoming.append(f"{i}) {_fmt_song(rows[0])}")
except Exception:
continue
if not upcoming:
await self._say(writer, "🎵 nothing else queued")
return
await self._say(writer, "🎵 next: " + " · ".join(upcoming))
async def _cmd_info(self, writer: asyncio.StreamWriter, user: str):
try:
song = await self.mpd.currentsong()
except Exception as e:
print(f"[info] failed for @{user}: {type(e).__name__}: {e}", flush=True)
await self._say(writer, f"@{user} couldn't read track info — MPD didn't respond")
return
if not song:
await self._say(writer, " nothing playing")
return
await self._say(writer, " " + _fmt_info(song))
# ─── Mattermost "going live" notifications ──────────────────────────────────
def _fetch_status(url: str) -> dict:
"""GET the status JSON. Raises on transport/parse errors so live_watch()
can log + back off uniformly."""
req = urllib.request.Request(url, method="GET",
headers={"User-Agent": "obs-twitch-bot/1.0"})
with urllib.request.urlopen(req, timeout=10) as r:
return json.loads(r.read().decode("utf-8"))
def _post_mattermost(hook_url: str, channel: str, title: str, game: str) -> None:
"""POST a single going-live notification. Markdown renders in Mattermost."""
parts = [f"🔴 **{channel}** is live on Twitch"]
if title:
parts.append(f"> {title}")
if game:
parts.append(f"_{game}_")
parts.append(f"https://twitch.tv/{channel}")
body = json.dumps({"text": "\n".join(parts)}).encode("utf-8")
req = urllib.request.Request(
hook_url, method="POST", data=body,
headers={"Content-Type": "application/json",
"User-Agent": "obs-twitch-bot/1.0"},
)
with urllib.request.urlopen(req, timeout=10) as r:
if r.status >= 300:
raise RuntimeError(f"mattermost hook returned HTTP {r.status}")
async def live_watch(mm_hook: str, channel: str, status_url: str = STATUS_URL):
"""Poll the status service; POST to Mattermost on offline→live transitions.
Transition detection is keyed on `started_at`, not just the `live` flag, so
a brief Helix hiccup (live=true → false → true within one stream session,
same `started_at`) doesn't double-fire.
Cold-start during an ongoing stream is intentionally silent: we record the
current `started_at` so we won't fire for it later, but we don't announce
it (we can't tell "they just went live" from "the bot restarted mid-stream"
without prior offline observation). Once we see live=False at least once,
the next started_at change becomes a real transition we'll announce.
"""
last_announced = None # started_at we've already handled (announced or recorded)
seen_offline = False # have we observed live=False since last_announced?
while True:
try:
data = await asyncio.to_thread(_fetch_status, status_url)
twitch = (data or {}).get("twitch") or {}
live = bool(twitch.get("live"))
started = (twitch.get("started_at") or None) if live else None
except Exception as e:
print(f"[live] poll failed: {type(e).__name__}: {e}", flush=True)
await asyncio.sleep(STATUS_POLL_S)
continue
if not live:
seen_offline = True
elif started:
if last_announced is None and not seen_offline:
# cold start mid-stream: lock in this started_at so we won't
# falsely announce it later
last_announced = started
elif started != last_announced and seen_offline:
title = twitch.get("title") or ""
game = twitch.get("game") or ""
try:
await asyncio.to_thread(_post_mattermost, mm_hook, channel, title, game)
print(f"[live] {channel} went live → posted to mattermost "
f"(title={title!r}, game={game!r})", flush=True)
except Exception as e:
# log but don't retry — a missed notification is better
# than a duplicate; the next genuine transition will fire
print(f"[live] mattermost post failed: {type(e).__name__}: {e}",
flush=True)
last_announced = started
seen_offline = False
await asyncio.sleep(STATUS_POLL_S)
# ─── entrypoint ─────────────────────────────────────────────────────────────
async def main():
env = {**load_env(ENV), **os.environ} # process env wins over .env
token = env.get("TWITCH_ACCESS_TOKEN", "").strip()
nick = env.get("TWITCH_NICK", "").strip()
channel = env.get("TWITCH_CHANNEL", nick).strip()
mm_hook = env.get("MM_HOOK", "").strip()
missing = [k for k, v in (("TWITCH_ACCESS_TOKEN", token), ("TWITCH_NICK", nick)) if not v]
if missing:
print(f"[err] missing in {ENV}: {', '.join(missing)}", file=sys.stderr)
sys.exit(1)
bot = TwitchBot(nick=nick, token=token, channel=channel, mpd=Mpd())
tasks = [bot.run()]
if mm_hook:
print(f"[live] watching {STATUS_URL} → mattermost on offline→live", flush=True)
tasks.append(live_watch(mm_hook, channel))
else:
print("[live] MM_HOOK unset — going-live notifications disabled", flush=True)
await asyncio.gather(*tasks)
if __name__ == "__main__":
try:
asyncio.run(main())
except KeyboardInterrupt:
sys.exit(0)