twitterhal 0.5.4

A MegaHAL bot for Twitter

TwitterHAL

A MegaHAL Twitter bot in Python.

This project is in alpha, and should NOT be considered stable in any way.

Live examples (in Swedish): @bibel3000, @trendhal3000

Prerequisites

• Python >= 3.6
• My fork of Chris Jones' Python MegaHAL
• bear/python-twitter
• carpedm20/emoji
• python-Levenshtein
• OPTIONAL: detectlanguage
• OPTIONAL: redis-py

But all those should be installed automatically by pip or setup.py.

Use pip install twitterhal[detectlanguage] to install detectlanguage, pip install twitterhal[redis] to install redis-py.

Usage

Command line

$ twitterhal
usage: twitterhal [-s SETTINGS_MODULE] [-d] [-m] [-f] [-t]
                  [-r | --chat | --stats | --print-config | --post-random | --version]

optional arguments:
  -s SETTINGS_MODULE, --settings SETTINGS_MODULE
                        Python path to settings module. If omitted, we try
                        looking for it in the 'TWITTERHAL_SETTINGS_MODULE'
                        environment variable.
  -d, --debug           More verbose logging output
  -m, --include-mentions
                        Include all mentions in replies (rather than just the
                        handle we're replying to)
  -f, --force           Try and force stuff, even if TwitterHAL doesn't want
                        to
  -t, --test            Test mode; doesn't actually post anything
  -r, --run             Run the bot!
  --chat                Chat with the bot
  --stats               Display some stats
  --print-config        Print current parsed config
  --post-random         Post a new random tweet
  --version             Show program's version number and exit

twitterhal --run will post random tweets at random_post_times (see below), as well as answering all incoming mentions, all while trying its best not to exceed the Twitter API rate limits.

As a library

from twitterhal import TwitterHAL
with TwitterHAL(screen_name="twitterhal") as hal:
    for mention in hal.get_new_mentions():
        hal.generate_reply(mention)
    hal.generate_random()
    hal.post_from_queue()

Configuration

Settings are read from a Python module specified in the TWITTERHAL_SETTINGS_MODULE environment variable, or whatever module you supply to the command-line utility via the [-s | --settings] parameter.

Some example settings:

SCREEN_NAME = "my_k3wl_twitter_user"
RANDOM_POST_TIMES = [datetime.time(8), datetime.time(16), datetime.time(22)]
INCLUDE_MENTIONS = True
DETECTLANGUAGE_API_KEY = ""
DATABASE = {
    "class": "path.to.DatabaseClass",
    "options": {},
    "test_options": {},
}
BANNED_USERS = ["my_other_twitterhal_bot"]
RUNNER_SLEEP_SECONDS = 5
POST_STATUS_LIMIT = 300
POST_STATUS_LIMIT_RESET_FREQUENCY = 3 * 60 * 60

TWITTER_API = {
    "consumer_key": "foo",
    "consumer_secret": "bar",
    "access_token_key": "boo",
    "access_token_secret": "far",
    "timeout": 40,
    "tweet_mode": "extended",
}

MEGAHAL_API = {
    "max_length": twitter.api.CHARACTER_LIMIT,
    "brainfile": "twitterhal-brain",
    "order": megahal.DEFAULT_ORDER,
    "timeout": megahal.DEFAULT_TIMEOUT,
    "banwords": ["MOST", "COMMON", "WORDS"],
}

BANNED_USERS: List of Twitter usernames (handles), without leading "@". We will never respond to, or mention, these users. Useful if you, for example, run two bots and don't want them to get stuck in an eternal loop responding to each other. (Perhaps, someday, I will figure out a clever way to detect such loops automatically.)

DATABASE: A dict of info about the database backend. Must at least contain the key class, which must be the path of a class inheriting from models.BaseDatabase. Included are models.ShelveDatabase and models.RedisDatabase. The options key contains kwargs to be sent to that database class' __init__() method. When TwitterHAL is run with the --test option, the options will be extended with the contents of the test_options dict.

INCLUDE_MENTIONS: if True, TwitterHAL will include all mentions in its replies. That is, not only the @handle of the user who wrote to it, but also every user they mentioned in their tweet. Perhaps you should use this carefully. Anyway, the default is False.

MEGAHAL contains keyword arguments for megahal.Megahal. Consult that module for more info.

MEGAHAL_API["banwords"]: you may want to set this if your bot will not be speaking English. Pro tip: search for a list of the ~300 most commonly used words in your language, and use those.

POST_STATUS_LIMIT and POST_STATUS_LIMIT_RESET_FREQUENCY: For some reason, Twitter's API doesn't provide info about the current ratio limits for posting tweets (and retweets), so I had to implement that check myself to my best ability. The numbers are taken from here.

RANDOM_POST_TIMES: TwitterHAL will post a randomly generated tweet on those points of (local) time every day. Default: 8:00, 16:00, and 22:00 (that is 8 AM, 4 PM and 10 PM, for those of you stuck in antiquity).

RUNNER_SLEEP_SECONDS: The interval with which runtime.runner starts its loop tasks. See below.

TWITTER_API contains keyword arguments for twitter.Api. Read more about it here.

Extending

Persistent storage

You may extend TwitterHAL's database by subclassing TwitterHAL and adding models.DatabaseItem definitions to its init_db() method. Maybe you want to feed the MegaHAL brain by regularily fetching top tweets for trending topics, and need to keep track of those? I know I do.

By default, the database (which is a subtype of models.BaseDatabase) will contain:

• posted_tweets (models.TweetList): List of posted Tweets
• mentions (models.TweetList): List of tweets that mention us, and whether they have been answered

Language detection

Tweets are internally stored in models.TweetList, which contains the method only_in_language(). This will filter out all tweets that are probably in the chosen language, with the help of the Language Detection API. Just pip install detectlanguage, get yourself an API key and feed it to detectlanguage.configuration.api_key (or set it in your settings; see above), and you're all set.

Twitter API calls

If you extend TwitterHAL with new methods that call the Twitter API, it's recommended you also check TwitterHAL's can_do_request(url), where url is something like /statuses/mentions_timeline (consult this page for full list), to see whether this call should be made at this time.

Runtime

The "daemon" (not really a daemon) twitterhal.runtime.runner, invoked by twitterhal --run, does these things:

1. Starts workers, which will run continuously in separate threads
2. With an interval of settings.RUNNER_SLEEP_SECONDS seconds (default: 5), runs loop tasks, each in a new thread
3. On exit, runs post loop tasks

Workers are registered by TwitterHAL.register_workers() through runner.register_worker(), and should be callables that loop until interrupted by a signal (see GracefulKiller section below). If they accept the boolean keyword argument restart, they will be executed with restart=True in case they exited prematurely and had to be restarted by the runner.

Loop tasks, unlike workers, should be finite in time. They are registered by TwitterHAL.register_loop_tasks() through runner.register_loop_task(), are run max once per loop, and can be any callable. If runner.register_loop_task() is called with the integer argument sleep (seconds), GracefulKiller.sleep() (see below) will be called at the end of every execution of this task, and the runner will be prohibited from starting new executions of the task until this one has finished. (If you don't want it to sleep at the end, but still want to block the task from being run multiple times concurrently, send sleep=0.)

Post loop tasks are registrered by TwitterHAL.register_post_loop_tasks() through runner.register_post_loop_task(), and can be any callable. They are called after the loop has been interrupted, and are not run in separate threads. Useful for various clean-up actions. By default, there are none.

GracefulKiller

gracefulkiller.killer is an object that listens for SIGINT and SIGTERM signals, whereupon its kill_now attribute is set to True. It also has a sleep() method, that mimics time.sleep() but aborts max 1 second after one of the aforementioned signals has been caught. sleep() returns True if SIGALRM was caught sometime during the sleeping, which could be used for pinging. Feel free to use this in your workers, loop tasks, etc.

Example:

from twitterhal.gracefulkiller import killer

def hello_world_worker():
    while not killer.kill_now:
        print("Hello, world!")
        ping = killer.sleep(10)
        if ping:
            print("Pong!")

Q & A

Why doesn't TwitterHAL see all my mentions?

Twitter has a setting called "quality filter", which is said to "filter lower-quality content from your notifications", and is turned on by default. You can go to your bot's notifications settings and uncheck the "Quality filter" checkbox (at least, that's how you did it 2020-04-24). This should solve it.