Migrating to v1.0

v1.0 is one of the biggest breaking changes in the library due to a complete redesign.

The amount of changes are so massive and long that for all intents and purposes, it is a completely new library.

Part of the redesign involves making things more easy to use and natural. Things are done on the models instead of requiring a Client instance to do any work.

Python Version Change

In order to make development easier and also to allow for our dependencies to upgrade to allow usage of 3.7 or higher, the library had to remove support for Python versions lower than 3.5.3, which essentially means that support for Python 3.4 is dropped.

Major Model Changes

Below are major model changes that have happened in v1.0

Snowflakes are int

Before v1.0, all snowflakes (the id attribute) were strings. This has been changed to int.

Quick example:

# before
ch = client.get_channel('84319995256905728')
if message.author.id == '80528701850124288':
    ...

# after
ch = client.get_channel(84319995256905728)
if message.author.id == 80528701850124288:
    ...

This change allows for fewer errors when using the Copy ID feature in the official client since you no longer have to wrap it in quotes and allows for optimisation opportunities by allowing ETF to be used instead of JSON internally.

Server is now Guild

The official API documentation calls the “Server” concept a “Guild” instead. In order to be more consistent with the API documentation when necessary, the model has been renamed to Guild and all instances referring to it has been changed as well.

A list of changes is as follows:

Before

After

Message.server

Message.guild

Channel.server

GuildChannel.guild

Client.servers

Client.guilds

Client.get_server

Client.get_guild()

Emoji.server

Emoji.guild

Role.server

Role.guild

Invite.server

Invite.guild

Member.server

Member.guild

Permissions.manage_server

Permissions.manage_guild

VoiceClient.server

VoiceClient.guild

Client.create_server

Client.create_guild()

Models are Stateful

As mentioned earlier, a lot of functionality was moved out of Client and put into their respective model.

A list of these changes is enumerated below.

Before

After

Client.add_reaction

Message.add_reaction()

Client.add_roles

Member.add_roles()

Client.ban

Member.ban() or Guild.ban()

Client.change_nickname

Member.edit()

Client.clear_reactions

Message.clear_reactions()

Client.create_channel

Guild.create_text_channel() and Guild.create_voice_channel()

Client.create_custom_emoji

Guild.create_custom_emoji()

Client.create_invite

abc.GuildChannel.create_invite()

Client.create_role

Guild.create_role()

Client.delete_channel

abc.GuildChannel.delete()

Client.delete_channel_permissions

abc.GuildChannel.set_permissions() with overwrite set to None

Client.delete_custom_emoji

Emoji.delete()

Client.delete_invite

Invite.delete() or Client.delete_invite()

Client.delete_message

Message.delete()

Client.delete_messages

TextChannel.delete_messages()

Client.delete_role

Role.delete()

Client.delete_server

Guild.delete()

Client.edit_channel

TextChannel.edit() or VoiceChannel.edit()

Client.edit_channel_permissions

abc.GuildChannel.set_permissions()

Client.edit_custom_emoji

Emoji.edit()

Client.edit_message

Message.edit()

Client.edit_profile

ClientUser.edit() (you get this from Client.user)

Client.edit_role

Role.edit()

Client.edit_server

Guild.edit()

Client.estimate_pruned_members

Guild.estimate_pruned_members()

Client.get_all_emojis

Client.emojis

Client.get_bans

Guild.bans()

Client.get_invite

Client.fetch_invite()

Client.get_message

abc.Messageable.fetch_message()

Client.get_reaction_users

Reaction.users()

Client.get_user_info

Client.fetch_user()

Client.invites_from

abc.GuildChannel.invites() or Guild.invites()

Client.join_voice_channel

VoiceChannel.connect() (see Voice Changes)

Client.kick

Guild.kick() or Member.kick()

Client.leave_server

Guild.leave()

Client.logs_from

abc.Messageable.history() (see Asynchronous Iterators)

Client.move_channel

TextChannel.edit() or VoiceChannel.edit()

Client.move_member

Member.edit()

Client.move_role

Role.edit()

Client.pin_message

Message.pin()

Client.pins_from

abc.Messageable.pins()

Client.prune_members

Guild.prune_members()

Client.purge_from

TextChannel.purge()

Client.remove_reaction

Message.remove_reaction()

Client.remove_roles

Member.remove_roles()

Client.replace_roles

Member.edit()

Client.send_file

abc.Messageable.send() (see Sending Messages)

Client.send_message

abc.Messageable.send() (see Sending Messages)

Client.send_typing

abc.Messageable.trigger_typing() (use abc.Messageable.typing())

Client.server_voice_state

Member.edit()

Client.start_private_message

User.create_dm()

Client.unban

Guild.unban() or Member.unban()

Client.unpin_message

Message.unpin()

Client.wait_for_message

Client.wait_for() (see Waiting For Events)

Client.wait_for_reaction

Client.wait_for() (see Waiting For Events)

Client.wait_until_login

Removed

Client.wait_until_ready

No change

Property Changes

In order to be a bit more consistent, certain things that were properties were changed to methods instead.

The following are now methods instead of properties (requires parentheses):

Dict Value Change

Prior to v1.0 some aggregating properties that retrieved models would return “dict view” objects.

As a consequence, when the dict would change size while you would iterate over it, a RuntimeError would be raised and crash the task. To alleviate this, the “dict view” objects were changed into lists.

The following views were changed to a list:

Voice State Changes

Earlier, in v0.11.0 a VoiceState class was added to refer to voice states along with a Member.voice attribute to refer to it.

However, it was transparent to the user. In an effort to make the library save more memory, the voice state change is now more visible.

The only way to access voice attributes is via the Member.voice attribute. Note that if the member does not have a voice state this attribute can be None.

Quick example:

# before
member.deaf
member.voice.voice_channel

# after
if member.voice: # can be None
    member.voice.deaf
    member.voice.channel

User and Member Type Split

In v1.0 to save memory, User and Member are no longer inherited. Instead, they are “flattened” by having equivalent properties that map out to the functional underlying User. Thus, there is no functional change in how they are used. However this breaks isinstance() checks and thus is something to keep in mind.

These memory savings were accomplished by having a global User cache, and as a positive consequence you can now easily fetch a User by their ID by using the new Client.get_user(). You can also get a list of all User your client can see with Client.users.

Channel Type Split

Prior to v1.0, channels were two different types, Channel and PrivateChannel with a is_private property to help differentiate between them.

In order to save memory the channels have been split into 4 different types:

With this split came the removal of the is_private attribute. You should now use isinstance().

The types are split into two different Abstract Base Classes:

So to check if something is a guild channel you would do:

isinstance(channel, discord.abc.GuildChannel)

And to check if it’s a private channel you would do:

isinstance(channel, discord.abc.PrivateChannel)

Of course, if you’re looking for only a specific type you can pass that too, e.g.

isinstance(channel, discord.TextChannel)

With this type split also came event changes, which are enumerated in Event Changes.

Miscellaneous Model Changes

There were lots of other things added or removed in the models in general.

They will be enumerated here.

Removed

  • Client.login() no longer accepts email and password logins.

    • Use a token and bot=False.

  • Client.get_all_emojis

  • Client.messages

  • Client.wait_for_message and Client.wait_for_reaction are gone.

  • Channel.voice_members

  • Channel.is_private

    • Use isinstance instead with one of the Abstract Base Classes instead.

    • e.g. isinstance(channel, discord.abc.GuildChannel) will check if it isn’t a private channel.

  • Client.accept_invite

    • There is no replacement for this one. This functionality is deprecated API wise.

  • Guild.default_channel / Server.default_channel and Channel.is_default

    • The concept of a default channel was removed from Discord. See #329.

  • Message.edited_timestamp

  • Message.timestamp

  • Colour.to_tuple()

  • Permissions.view_audit_logs

  • Member.game

  • Guild.role_hierarchy / Server.role_hierarchy

    • Use Guild.roles instead. Note that while sorted, it is in the opposite order of what the old Guild.role_hierarchy used to be.

Changed

Added

Sending Messages

One of the changes that were done was the merger of the previous Client.send_message and Client.send_file functionality into a single method, send().

Basically:

# before
await client.send_message(channel, 'Hello')

# after
await channel.send('Hello')

This supports everything that the old send_message supported such as embeds:

e = discord.Embed(title='foo')
await channel.send('Hello', embed=e)

There is a caveat with sending files however, as this functionality was expanded to support multiple file attachments, you must now use a File pseudo-namedtuple to upload a single file.

# before
await client.send_file(channel, 'cool.png', filename='testing.png', content='Hello')

# after
await channel.send('Hello', file=discord.File('cool.png', 'testing.png'))

This change was to facilitate multiple file uploads:

my_files = [
    discord.File('cool.png', 'testing.png'),
    discord.File(some_fp, 'cool_filename.png'),
]

await channel.send('Your images:', files=my_files)

Asynchronous Iterators

Prior to v1.0, certain functions like Client.logs_from would return a different type if done in Python 3.4 or 3.5+.

In v1.0, this change has been reverted and will now return a singular type meeting an abstract concept called AsyncIterator.

This allows you to iterate over it like normal:

async for message in channel.history():
    print(message)

Or turn it into a list:

messages = await channel.history().flatten()
for message in messages:
    print(message)

A handy aspect of returning AsyncIterator is that it allows you to chain functions together such as AsyncIterator.map() or AsyncIterator.filter():

async for m_id in channel.history().filter(lambda m: m.author == client.user).map(lambda m: m.id):
    print(m_id)

The functions passed to AsyncIterator.map() or AsyncIterator.filter() can be either coroutines or regular functions.

You can also get single elements a la discord.utils.find() or discord.utils.get() via AsyncIterator.get() or AsyncIterator.find():

my_last_message = await channel.history().get(author=client.user)

The following return AsyncIterator:

Event Changes

A lot of events have gone through some changes.

Many events with server in the name were changed to use guild instead.

Before:

  • on_server_join

  • on_server_remove

  • on_server_update

  • on_server_role_create

  • on_server_role_delete

  • on_server_role_update

  • on_server_emojis_update

  • on_server_available

  • on_server_unavailable

After:

The on_voice_state_update() event has received an argument change.

Before:

async def on_voice_state_update(before, after)

After:

async def on_voice_state_update(member, before, after)

Instead of two Member objects, the new event takes one Member object and two VoiceState objects.

The on_guild_emojis_update() event has received an argument change.

Before:

async def on_guild_emojis_update(before, after)

After:

async def on_guild_emojis_update(guild, before, after)

The first argument is now the Guild that the emojis were updated from.

The on_member_ban() event has received an argument change as well:

Before:

async def on_member_ban(member)

After:

async def on_member_ban(guild, user)

As part of the change, the event can either receive a User or Member. To help in the cases that have User, the Guild is provided as the first parameter.

The on_channel_ events have received a type level split (see Channel Type Split).

Before:

  • on_channel_delete

  • on_channel_create

  • on_channel_update

After:

The on_guild_channel_ events correspond to abc.GuildChannel being updated (i.e. TextChannel and VoiceChannel) and the on_private_channel_ events correspond to abc.PrivateChannel being updated (i.e. DMChannel and GroupChannel).

Voice Changes

Voice sending has gone through a complete redesign.

In particular:

  • Connection is done through VoiceChannel.connect() instead of Client.join_voice_channel.

  • You no longer create players and operate on them (you no longer store them).

  • You instead request VoiceClient to play an AudioSource via VoiceClient.play().

  • There are different built-in AudioSources.

  • create_ffmpeg_player/create_stream_player/create_ytdl_player have all been removed.

  • Using VoiceClient.play() will not return an AudioPlayer.

    • Instead, it’s “flattened” like User -> Member is.

  • The after parameter now takes a single parameter (the error).

Basically:

Before:

vc = await client.join_voice_channel(channel)
player = vc.create_ffmpeg_player('testing.mp3', after=lambda: print('done'))
player.start()

player.is_playing()
player.pause()
player.resume()
player.stop()
# ...

After:

vc = await channel.connect()
vc.play(discord.FFmpegPCMAudio('testing.mp3'), after=lambda e: print('done', e))
vc.is_playing()
vc.pause()
vc.resume()
vc.stop()
# ...

With the changed AudioSource design, you can now change the source that the VoiceClient is playing at runtime via VoiceClient.source.

For example, you can add a PCMVolumeTransformer to allow changing the volume:

vc.source = discord.PCMVolumeTransformer(vc.source)
vc.source.volume = 0.6

An added benefit of the redesign is that it will be much more resilient towards reconnections:

  • The voice websocket will now automatically re-connect and re-do the handshake when disconnected.

  • The initial connect handshake will now retry up to 5 times so you no longer get as many asyncio.TimeoutError.

  • Audio will now stop and resume when a disconnect is found.

    • This includes changing voice regions etc.

Waiting For Events

Prior to v1.0, the machinery for waiting for an event outside of the event itself was done through two different functions, Client.wait_for_message and Client.wait_for_reaction. One problem with one such approach is that it did not allow you to wait for events outside of the ones provided by the library.

In v1.0 the concept of waiting for another event has been generalised to work with any event as Client.wait_for().

For example, to wait for a message:

# before
msg = await client.wait_for_message(author=message.author, channel=message.channel)

# after
def pred(m):
    return m.author == message.author and m.channel == message.channel

msg = await client.wait_for('message', check=pred)

To facilitate multiple returns, Client.wait_for() returns either a single argument, no arguments, or a tuple of arguments.

For example, to wait for a reaction:

reaction, user = await client.wait_for('reaction_add', check=lambda r, u: u.id == 176995180300206080)

# use user and reaction

Since this function now can return multiple arguments, the timeout parameter will now raise a asyncio.TimeoutError when reached instead of setting the return to None. For example:

def pred(m):
    return m.author == message.author and m.channel == message.channel

try:

    msg = await client.wait_for('message', check=pred, timeout=60.0)
except asyncio.TimeoutError:
    await channel.send('You took too long...')
else:
    await channel.send('You said {0.content}, {0.author}.'.format(msg))

Upgraded Dependencies

Following v1.0 of the library, we’ve updated our requirements to aiohttp v2.0 or higher.

Since this is a backwards incompatible change, it is recommended that you see the changes and the Migration to 2.x pages for details on the breaking changes in aiohttp.

Of the most significant for common users is the removal of helper functions such as:

  • aiohttp.get

  • aiohttp.post

  • aiohttp.delete

  • aiohttp.patch

  • aiohttp.head

  • aiohttp.put

  • aiohttp.request

It is recommended that you create a session instead:

async with aiohttp.ClientSession() as sess:
    async with sess.get('url') as resp:
        # work with resp

Since it is better to not create a session for every request, you should store it in a variable and then call session.close on it when it needs to be disposed.

Sharding

The library has received significant changes on how it handles sharding and now has sharding as a first-class citizen.

If using a Bot account and you want to shard your bot in a single process then you can use the AutoShardedClient.

This class allows you to use sharding without having to launch multiple processes or deal with complicated IPC.

It should be noted that the sharded client does not support user accounts. This is due to the changes in connection logic and state handling.

Usage is as simple as doing:

client = discord.AutoShardedClient()

instead of using Client.

This will launch as many shards as your bot needs using the /gateway/bot endpoint, which allocates about 1000 guilds per shard.

If you want more control over the sharding you can specify shard_count and shard_ids.

# launch 10 shards regardless
client = discord.AutoShardedClient(shard_count=10)

# launch specific shard IDs in this process
client = discord.AutoShardedClient(shard_count=10, shard_ids=(1, 2, 5, 6))

For users of the command extension, there is also AutoShardedBot which behaves similarly.

Connection Improvements

In v1.0, the auto reconnection logic has been powered up significantly.

Client.connect() has gained a new keyword argument, reconnect that defaults to True which controls the reconnect logic. When enabled, the client will automatically reconnect in all instances of your internet going offline or Discord going offline with exponential back-off.

Client.run() and Client.start() gains this keyword argument as well, but for most cases you will not need to specify it unless turning it off.

Command Extension Changes

Due to the Models are Stateful changes, some of the design of the extension module had to undergo some design changes as well.

Context Changes

In v1.0, the Context has received a lot of changes with how it’s retrieved and used.

The biggest change is that pass_context=True no longer exists, Context is always passed. Ergo:

# before
@bot.command()
async def foo():
    await bot.say('Hello')

# after
@bot.command()
async def foo(ctx):
    await ctx.send('Hello')

The reason for this is because Context now meets the requirements of abc.Messageable. This makes it have similar functionality to TextChannel or DMChannel. Using send() will either DM the user in a DM context or send a message in the channel it was in, similar to the old bot.say functionality. The old helpers have been removed in favour of the new abc.Messageable interface. See Removed Helpers for more information.

Since the Context is now passed by default, several shortcuts have been added:

New Shortcuts

  • ctx.author is a shortcut for ctx.message.author.

  • ctx.guild is a shortcut for ctx.message.guild.

  • ctx.channel is a shortcut for ctx.message.channel.

  • ctx.me is a shortcut for ctx.message.guild.me or ctx.bot.user.

  • ctx.voice_client is a shortcut for ctx.message.guild.voice_client.

New Functionality

Subclassing Context

In v1.0, there is now the ability to subclass Context and use it instead of the default provided one.

For example, if you want to add some functionality to the context:

class MyContext(commands.Context):
    @property
    def secret(self):
        return 'my secret here'

Then you can use get_context() inside on_message() with combination with invoke() to use your custom context:

class MyBot(commands.Bot):
    async def on_message(self, message):
        ctx = await self.get_context(message, cls=MyContext)
        await self.invoke(ctx)

Now inside your commands you will have access to your custom context:

@bot.command()
async def secret(ctx):
    await ctx.send(ctx.secret)

Removed Helpers

With the new Context changes, a lot of message sending helpers have been removed.

For a full list of changes, see below:

Before

After

Bot.say

Context.send()

Bot.upload

Context.send()

Bot.whisper

ctx.author.send

Bot.type

Context.typing() or Context.trigger_typing()

Bot.reply

No replacement.

Command Changes

As mentioned earlier, the first command change is that pass_context=True no longer exists, so there is no need to pass this as a parameter.

Another change is the removal of no_pm=True. Instead, use the new guild_only() built-in check.

The commands attribute of Bot and Group have been changed from a dictionary to a set that does not have aliases. To retrieve the previous dictionary behaviour, use all_commands instead.

Command instances have gained new attributes and properties:

  1. signature to get the signature of the command.

  2. usage, an attribute to override the default signature.

  3. root_parent to get the root parent group of a subcommand.

For Group and Bot the following changed:

Check Changes

Prior to v1.0, check()s could only be synchronous. As of v1.0 checks can now be coroutines.

Along with this change, a couple new checks were added.

Event Changes

All command extension events have changed.

Before:

on_command(command, ctx)
on_command_completion(command, ctx)
on_command_error(error, ctx)

After:

on_command(ctx)
on_command_completion(ctx)
on_command_error(ctx, error)

The extraneous command parameter in on_command() and on_command_completion() have been removed. The Command instance was not kept up-to date so it was incorrect. In order to get the up to date Command instance, use the Context.command attribute.

The error handlers, either Command.error() or on_command_error(), have been re-ordered to use the Context as its first parameter to be consistent with other events and commands.

HelpFormatter and Help Command Changes

The HelpFormatter class has been removed. It has been replaced with a HelpCommand class. This class now stores all the command handling and processing of the help command.

The help command is now stored in the Bot.help_command attribute. As an added extension, you can disable the help command completely by assigning the attribute to None or passing it at __init__ as help_command=None.

The new interface allows the help command to be customised through special methods that can be overridden.

Certain subclasses can implement more customisable methods.

The old HelpFormatter was replaced with DefaultHelpCommand, which implements all of the logic of the old help command. The customisable methods can be found in the accompanying documentation.

The library now provides a new more minimalistic HelpCommand implementation that doesn’t take as much space, MinimalHelpCommand. The customisable methods can also be found in the accompanying documentation.

A frequent request was if you could associate a help command with a cog. The new design allows for dynamically changing of cog through binding it to the HelpCommand.cog attribute. After this assignment the help command will pretend to be part of the cog and everything should work as expected. When the cog is unloaded then the help command will be “unbound” from the cog.

For example, to implement a HelpCommand in a cog, the following snippet can be used.

class MyHelpCommand(commands.MinimalHelpCommand):
    def get_command_signature(self, command):
        return '{0.clean_prefix}{1.qualified_name} {1.signature}'.format(self, command)

class MyCog(commands.Cog):
    def __init__(self, bot):
        self._original_help_command = bot.help_command
        bot.help_command = MyHelpCommand()
        bot.help_command.cog = self

    def cog_unload(self):
        self.bot.help_command = self._original_help_command

For more information, check out the relevant documentation.

Cog Changes

Cogs have completely been revamped. They are documented in Cogs as well.

Cogs are now required to have a base class, Cog for future proofing purposes. This comes with special methods to customise some behaviour.

Those that were using listeners, such as on_message inside a cog will now have to explicitly mark them as such using the commands.Cog.listener() decorator.

Along with that, cogs have gained the ability to have custom names through specifying it in the class definition line. More options can be found in the metaclass that facilitates all this, commands.CogMeta.

An example cog with every special method registered and a custom name is as follows:

class MyCog(commands.Cog, name='Example Cog'):
    def cog_unload(self):
        print('cleanup goes here')

    def bot_check(self, ctx):
        print('bot check')
        return True

    def bot_check_once(self, ctx):
        print('bot check once')
        return True

    async def cog_check(self, ctx):
        print('cog local check')
        return await ctx.bot.is_owner(ctx.author)

    async def cog_command_error(self, ctx, error):
        print('Error in {0.command.qualified_name}: {1}'.format(ctx, error))

    async def cog_before_invoke(self, ctx):
        print('cog local before: {0.command.qualified_name}'.format(ctx))

    async def cog_after_invoke(self, ctx):
        print('cog local after: {0.command.qualified_name}'.format(ctx))

    @commands.Cog.listener()
    async def on_message(self, message):
        pass

Before and After Invocation Hooks

Commands have gained new before and after invocation hooks that allow you to do an action before and after a command is run.

They take a single parameter, Context and they must be a coroutine.

They are on a global, per-cog, or per-command basis.

Basically:

# global hooks:

@bot.before_invoke
async def before_any_command(ctx):
    # do something before a command is called
    pass

@bot.after_invoke
async def after_any_command(ctx):
    # do something after a command is called
    pass

The after invocation is hook always called, regardless of an error in the command. This makes it ideal for some error handling or clean up of certain resources such a database connection.

The per-command registration is as follows:

@bot.command()
async def foo(ctx):
    await ctx.send('foo')

@foo.before_invoke
async def before_foo_command(ctx):
    # do something before the foo command is called
    pass

@foo.after_invoke
async def after_foo_command(ctx):
    # do something after the foo command is called
    pass

The special cog method for these is Cog.cog_before_invoke() and Cog.cog_after_invoke(), e.g.:

class MyCog(commands.Cog):
    async def cog_before_invoke(self, ctx):
        ctx.secret_cog_data = 'foo'

    async def cog_after_invoke(self, ctx):
        print('{0.command} is done...'.format(ctx))

    @commands.command()
    async def foo(self, ctx):
        await ctx.send(ctx.secret_cog_data)

To check if a command failed in the after invocation hook, you can use Context.command_failed.

The invocation order is as follows:

  1. Command local before invocation hook

  2. Cog local before invocation hook

  3. Global before invocation hook

  4. The actual command

  5. Command local after invocation hook

  6. Cog local after invocation hook

  7. Global after invocation hook

Converter Changes

Prior to v1.0, a converter was a type hint that could be a callable that could be invoked with a singular argument denoting the argument passed by the user as a string.

This system was eventually expanded to support a Converter system to allow plugging in the Context and do more complicated conversions such as the built-in “discord” converters.

In v1.0 this converter system was revamped to allow instances of Converter derived classes to be passed. For consistency, the convert() method was changed to always be a coroutine and will now take the two arguments as parameters.

Essentially, before:

class MyConverter(commands.Converter):
    def convert(self):
        return self.ctx.message.server.me

After:

class MyConverter(commands.Converter):
    async def convert(self, ctx, argument):
        return ctx.me

The command framework also got a couple new converters: