The bug was due to the fact that the way overwrites work is by doing
a loop of all the values and then applying deny first and then allow.
That way the overwrite is defined if role A and role B deny a
permission but role C allows it (and said member has A, B, C roles)
then the resolution should allow it rather than deny it regardless of
the order of the data it is received in.
The library worked with the assumption that whenever the recv call for
the websocket would lead to a closure, the close method would be called
to signal closure and as a result our close method would be called.
This assumption turned out to be false as the websockets library would
instead call an internal function named close_connection instead. So to
solve our problem we need to override this function instead of close.
This seems to be a very common question regarding the docs so might as
well document exactly what this means rather than using cryptic and
confusing wording.
Bug was caused to the shallow copy not copying over the VoiceState
information embedded into the copy. This would mean that when the event
is called, before and after voice state information is essentially
equivalent.
The solution to fix this is to also copy the VoiceState objects.
The way the command cooldown works is using a windowed way of doing it.
That is, if we have a cooldown of 2 commands every 30 seconds then if we
do a single command, we have 30 seconds to do the second command or else
we will get rate limited. This more or less matches the common
expectations on how cooldowns should be.
These cooldowns can be bucketed up to a single dimension of depth for
a per-user, per-guild, or per-channel basis. Of course, a global bucket
is also provided. These cannot be mixed, e.g. no per-channel per-user
cooldowns.
When a command cooldown is triggered, the error handlers will receive a
an exception of type CommandOnCooldown with proper information regarding
the cooldown such as retry_after and the bucket information itself.
This is for exceptions that are inherently based on user errors and not
permission based. e.g. passing incorrect number of arguments, too many
arguments, or an invalid argument. CommandNotFound is not classified
under this since it isn't inherently a user input error in all cases.
Some invalid commands can simply be due to an odd bot prefix. It would
also diminish the usefulness of the new parent class if CommandNotFound
was included.
