From 54fd356b39a328c3a89eca81174ed83974359b55 Mon Sep 17 00:00:00 2001
From: Mike Lang <mikelang3000@gmail.com>
Date: Fri, 23 May 2025 18:25:15 +1000
Subject: [PATCH] Add youtubebot

This adds a zulip bot that polls the youtube API for new comment threads, and posts them to Zulip.
Some limitations:
- It doesn't keep any state, so it won't post anything it "missed" while not running.
- It can only find top-level comments, not replies
- For quota reasons, we shouldn't poll more often than every 1 minute (at this rate we consume approx 1 upload worth of quota per day)
- If somehow there are more than 100 comments within 1 minute, it will miss all but the last 100.
---
 docker-compose.jsonnet              |  16 +++++
 zulip_bots/zulip_bots/youtubebot.py | 107 ++++++++++++++++++++++++++++
 2 files changed, 123 insertions(+)
 create mode 100644 zulip_bots/zulip_bots/youtubebot.py

diff --git a/docker-compose.jsonnet b/docker-compose.jsonnet
index 3031ee9..7a1ee87 100644
--- a/docker-compose.jsonnet
+++ b/docker-compose.jsonnet
@@ -37,6 +37,7 @@
     pubbot: false,
     blogbot: false,
     prizebot: false,
+    youtubebot: false,
     bus_analyzer: false,
     graphs: false,
   },
@@ -358,6 +359,13 @@
     state_path:: "./prizebot_state.json",
   },
 
+  youtubebot:: {
+    zulip_email: "youtube-bot@chat.videostrike.team",
+    zulip_api_key: "",
+    google_credentials_file: $.cutter_creds_file,
+    channel_id: "UCz5-PNQxaT4WtB_OMAwD85g", // DesertBusForHope
+  },
+
   // template for donation data urls
   donation_url_template:: "https://example.com/DB{}/DB{}.json",
 
@@ -805,6 +813,14 @@
         volumes: ["%s:%s" % [$.prizebot.state_path, $.prizebot.state]],
       },
 
+    [if $.enabled.youtubebot then "youtubebot"]:
+      bot_service("youtubebot", $.youtubebot + {
+        zulip_url: $.zulip_url,
+        google_credentials_file: "/creds.json",
+      }) + {
+        volumes: ["%s:/creds.json" % $.youtubebot.google_credentials_file],
+      },
+
   },
 
 }
diff --git a/zulip_bots/zulip_bots/youtubebot.py b/zulip_bots/zulip_bots/youtubebot.py
new file mode 100644
index 0000000..25e42e3
--- /dev/null
+++ b/zulip_bots/zulip_bots/youtubebot.py
@@ -0,0 +1,107 @@
+
+import gevent.monkey
+gevent.monkey.patch_all()
+
+import json
+import logging
+import time
+
+from common.googleapis import GoogleAPIClient
+
+from .config import get_config
+from .zulip import Client
+
+def get_comments(google, channel_id):
+	resp = google.request("GET",
+		"https://www.googleapis.com/youtube/v3/commentThreads",
+		params={
+			"part": "snippet",
+			"allThreadsRelatedToChannelId": channel_id,
+			"maxResults": "100",
+			"textFormat": "plainText",
+		}
+	)
+	resp.raise_for_status()
+	items = resp.json()["items"][::-1] # flip direction so we get earliest first
+	if items:
+		earliest = items[0]["snippet"]["topLevelComment"]
+		logging.info(f"Got {len(items)} comment threads, oldest is {earliest['id']} at {earliest['snippet']['publishedAt']}")
+	else:
+		logging.info("Got no comment threads")
+	# We could look at replies, but since we can only check for new replies in the first 100 threads,
+	# we'd rather just never show them than confuse people when they don't show up sometimes.
+	comments = []
+	for thread in items:
+		logging.debug(f"Got thread: {json.dumps(thread)}")
+		comment = thread["snippet"]["topLevelComment"]
+		comment["videoId"] = thread["snippet"]["videoId"]
+		comments.append(comment)
+	return comments
+
+
+def show_comment(zulip, stream, topic, comment):
+	c = comment["snippet"]
+	author = f"[{c['authorDisplayName']}]({c['authorChannelUrl']})"
+	video = f"https://youtu.be/{comment['videoId']}"
+	message = f"{author} commented on {video}:\n```quote\n{c['textDisplay']}\n```"
+	logging.info(f"Sending message to {stream}/{topic}: {message!r}")
+	# Empty stream acts as a dry-run mode
+	if stream:
+		zulip.send_to_stream(stream, topic, message)
+
+
+def main(conf_file, interval=60, one_off=0, stream="bot-spam", topic="Youtube Comments", keep=1000, log="INFO"):
+	"""Config:
+		zulip_url
+		zulip_email
+		zulip_api_key
+		channel_id
+		google_credentials_file:
+			Path to json file containing at least:
+				client_id
+				client_secret
+				refresh_token
+			These creds should be authed as the target account with Youtube Data API read perms
+
+	In one-off=N mode, get the last N comments and show them, then exit.
+	"""
+	logging.basicConfig(level=log)
+
+	config = get_config(conf_file)
+	zulip = Client(config["zulip_url"], config["zulip_email"], config["zulip_api_key"])
+	with open(config["google_credentials_file"]) as f:
+		credentials = json.load(f)
+	google = GoogleAPIClient(credentials["client_id"], credentials["client_secret"], credentials["refresh_token"])
+	channel_id = config["channel_id"]
+
+	if one_off:
+		comments = get_comments(google, channel_id)
+		for comment in comments[-one_off:]:
+			show_comment(zulip, stream, topic, comment)
+		return
+
+	seen = None
+	while True:
+		start = time.monotonic()
+
+		if seen is None:
+			# Get latest messages as of startup, so we know what's new next time
+			seen = [comment["id"] for comment in get_comments(google, channel_id)]
+		else:
+			for comment in get_comments(google, channel_id):
+				if comment["id"] in seen:
+					logging.debug(f"Comment {comment['id']} already seen, skipping")
+					continue
+				show_comment(zulip, stream, topic, comment)
+				seen.append(comment["id"])
+		seen = seen[-keep:]
+
+		remaining = start + interval - time.monotonic()
+		logging.debug(f"Keeping {len(seen)} seen, waiting {remaining:.2f}s")
+		if remaining > 0:
+			time.sleep(remaining)
+
+
+if __name__ == '__main__':
+	import argh
+	argh.dispatch_command(main)