In python 3.8, asyncio.CancelledError is a subclass of BaseException
rather than Exception, so `except Exception:` will not swallow
CancelledError. This change prevents an error in 3.8 from being printed
to the console when the following is run:
```
async with ctx.typing():
pass
```
This has been a massive pain point for me personally due to the poor
design of the Enum class leading to the common use cases used in the
library being significantly slow. Since this Enum is not public facing
in terms of *creation*, I can only implement the APIs that are used
when *accessing* them.
This Enum is a drop-in replacement to the pre-existing enum.Enum class
except it comes with significant speed-ups. Since this is a lot to go
over, I will let the numbers speak for themselves:
In [4]: %timeit enums.try_enum(enums.Status, 'offline')
263 ns ± 34.3 ns per loop (7 runs, 1000000 loops each)
In [5]: %timeit NeoStatus.try_value('offline')
134 ns ± 0.859 ns per loop (7 runs, 10000000 loops each)
In [6]: %timeit enums.Status.offline
116 ns ± 0.378 ns per loop (7 runs, 10000000 loops each)
In [7]: %timeit NeoStatus.offline
31.6 ns ± 0.327 ns per loop (7 runs, 10000000 loops each)
In [8]: %timeit enums.Status.offline.value
382 ns ± 15.2 ns per loop (7 runs, 1000000 loops each)
In [9]: %timeit NeoStatus.offline.value
65.5 ns ± 0.953 ns per loop (7 runs, 10000000 loops each)
In [10]: %timeit str(enums.Status.offline)
630 ns ± 14.8 ns per loop (7 runs, 1000000 loops each)
In [11]: %timeit str(NeoStatus.offline)
253 ns ± 3.53 ns per loop (7 runs, 1000000 loops each)
In [12]: %timeit enums.Status('offline')
697 ns ± 8.42 ns per loop (7 runs, 1000000 loops each)
In [13]: %timeit NeoStatus('offline')
182 ns ± 1.83 ns per loop (7 runs, 10000000 loops each)
Embed edits are no longer special cased in the dispatch code, which
could lead to on_message_edit being called more often than it used to
be called. I am not sure on the general impact on that being removed.
Fixes#2195
Probably not a significant difference but might as well use it here.
The basic idea is to cache the getattr calls instead of repeatedly
doing it (since they're around 105ns on my machine). The dictionary
lookup is about 41ns on my machine.
The next step in speeding up library code some more should be in
the parser bodies themselves but that's a problem to tackle another
day.
This includes raw events (which didn't have any) and a few other
types that were missing them. Upon review some more useful fields were
added to the repr output which would be more useful during debugging.
