Updates to CoreBluetooth backend

[bsiever]

This includes several updates to the CoreBluetooth backend. These include:

• Support for scanning w/ filtering (Using BlueZ-style flags)
• Performance improvements (connect will use cached peripheral for faster connect following discovery and/or faster reconnecting)
• Several bug fixes
  • Some errors were causing crashes rather than raising exceptions. (Exceptions have been moved out of delegate callbacks)
• Initial work for supporting multiple clients (still needs further testing)
  • Different approach to disconnect callbacks for multiple peripherals
• Misc. small consistency changed (Ex: importing/using CoreBluetooth constants rather than their values)
• A few of these are a bit hacky and merit revision

There are a lot of changes. I'm happy to answer any questions and won't be hurt if this is rejected. I'm also able to do a PR for my test suite (pretty rough) to help quickly vet the work (requires 1+ microbit).

[5frank]

Nice! I have not had the time to test it yet.
Some thoughts:

Not sure I like adding variables to class instances. e.g

        self[xUUID]._error = None  # Clear error prior to new event

I believe it is a bit confusing and it makes pylint unhappy. Perhaps inherit ayncio.Event instead? This would also make it easy to add a timeout to event.wait() if/when needed and remove some duplicated code. Example (untested):

class _BleakCorebluetoothEvent(asyncio.Event):
    def __init__(self):
        self._error = None
        super().__init__()

    def clear(self):
        self._error = None
        super().clear()

    def set(error: NSError = None): # or pass a Exception?
        self._error = error
        super().set()

    async def wait(self, timeout=None):
        if timeout:
            try:
                await asyncio.wait_for(super().wait(), timeout)
            except asyncio.TimeoutError:
                raise TimeoutError("")
        else:
            await super().wait()

        if self._error:
            raise SomeException(self._error)

[bsiever]

My approach is a hack. I like the idea of a subclass (and better semantics for waits/timeouts). I'll probably try to incorporate your approach in the next few days.

[hbldh]

Whoa. This was a lot of changes in one PR... I would have liked them spread out over several to be able to review them better. Nevertheless, thank you!

I will await you changes and then plan to look into this as soon as possible though.

[bsiever]

Yeah, sorry for the number of changes in one batch. It was one of my motivations for Issue #207 (Unit testing with devices). My development path was pretty haphazard. I'll try to be a little more structured in the future.

I've incorporated a first pass at @5frank's approach, which is much much cleaner. It also provides an opportunity for timeouts on all device events, which could make "hanging" almost impossible. I think this could use further review discussion, but the constructors in initWithPeripheral_ just need to specify the timeout for events. I've chosen to include the new _BleakCorebluetoothEvent class in PeripheralDelegate.py. It may be better as a separate file so it could be used elsewhere.

[f0rodo]

+1 This fixed bleak breaking in Catalina

[dlech]

Hi @bsiever, can we break this down into some smaller pull requests to start getting things merged? I started fixing things myself before I saw this PR. Changes are here if you want to compare notes (still a bit of a work in progress though).

[hbldh]

I have planned on devoting an entire day to bleak during next week, with explicit focus on macOS issues. @dlech Could you integrate your changes on with these ones (if needed)? I could merge this into a separate branch in my repo so that additional changes can be appended. Would that help?

@bsiever Thank you for the work done here. I will try it out first thing on the Bleak-day!

[bsiever]

@hbldh I hope Bleak Day goes well! If you have a micro:bit the unit tests I've made can help quickly vet that additions aren't breaking too many things.

@dlech It's clear that you caught things I missed and I restructured parts and added features you haven't hit. We both made a lot of changes and it's definitely a bit messy to integrate. The "big ones" for me were:

• Moved error handling out of delegates. That was causing un-catchable Python crashes.
• Added support for disconnect via weak references from the cbapp singleton.
• Added support for multiple clients (also using the cbapp's weak references for some updates)
• Used cached CBPeripheral objects for faster connections
• Lots of small, misc. error handling.

I probably won't have time to make many updates for the next week or two, but I should have time to reply to any questions.

[hbldh]

I merged the PR #227 before this one, since it was smaller and contained exacty what I needed for the moment. Regrettably I have to had time to merge this with that yet, so I will leave this one for a 0.7.X release in a few days.

[hbldh]

@bsiever I tried to do a careful merge of your changes into a separate branch (bsiever-develop), but it ended up in a broken state. I do not now how to move forward from there and I think I have to abandon the attempt. Reluctantly, I will close this PR and ask you to update your changes with current develop. Thank you for the effort.

The things I liked in the PR, that wasn't solved in merging of #227:

1. the scanning filter work: If you could integrate your scanning filter work with the current develop and make a PR I would be very grateful. If there are more changes still needed for making your test suite run, please add them in a separate PR and we can address that separately.
2. the reuse of scan results: For the reuse of scan results I have an idea for version 0.8.0 that will collide with that, so for the time being I will go a different route.