Workaround for list_muses failing when backend='gatt' on linux

[hubertjb]

This PR introduces a workaround to use list_muses with gatt on Linux.

Some context: When using backend='gatt' on Linux, pygatt relies on the command line tool hcitool to scan for BLE devices. hcitool is however deprecated, and seems to fail on Bluetooth 5 devices. This workaround roughly replicates the functionality of pygatt.backends.gatttool.gatttool.GATTToolBackend.scan() using the more modern bluetoothctl tool.

I had to rely on pexpect to run bluetoothctl scan on as subprocess.run seems to be returning before any device can be found when used in a jupyter notebook environment, which doesn't make for a very elegant approach. Also, I suppose this might be of interest to the pygatt folks, but wanted to make a quick fix available for the eeg-notebooks sprint at BrainHack Ontario.

Deprecation of hcitool: https://git.kernel.org/pub/scm/bluetooth/bluez.git/commit/?id=b1eb2c4cd057624312e0412f6c4be000f7fc3617

@m9h @JohnGriffiths

[ErikBjare]

Some background on the hcitool issue, and a workaround for Arch Linux: https://github.com/alexandrebarachant/muse-lsl/issues/119#issuecomment-701289365

[hubertjb]

Awesome, thanks for trying it out @ErikBjare! I think sudo is still required because the code first tries scanning with pygatt (which asks for sudo), and only if that fails will it use bluetoothctl. Not sure if that holds for muselsl stream though.

[xloem]

I've merged this into https://github.com/xloem/muse-lsl .

[jdpigeon]

Fantastic. Thanks for this, @hubertjb.

I'm going to test this myself and make a small follow-up PR that makes this the standard way to search for muses on Linux and avoid the pygatt failure step.

[jdpigeon]

I'm running into a little issue with this change where the bluetoothctl subprocess will not complete until I type quit into the terminal to end the interactive session it inits.

I'm not too familiar with bluetoothctrl. Is there a way to make it return directly?

[xloem]

This change could also be good as a patch to the https://github.com/peplin/pygatt project itself; then more people will review, benefit from, and maintain it.

[xloem]

Note: to make it terminate on its own pass --timeout: bluetoothctl --timeout 5 scan on scans for only 5 seconds. Passing that would be a good decision, but note @jdpigeon it works fine for me, each launch of the subprocess is passing individual commands and not opening an interactive session.

[hubertjb]

@jdpigeon interesting, I haven't had that problem (as seems to be the case for @xloem too). Can you try with the timeout as @xloem suggests?

[hubertjb]

@xloem yes good point, I was thinking this workaround could be useful in the pygatt project itself. I'll look into submitting a PR there too.

[xloem]

Just to clarify and restate, I didn't have the issue jdpigeon had, it sounds like they have some version of bluetoothctl where an interactive shell is launched when a command is passed?

[jdpigeon]

I just tried it out on another more up to date linux machine and bluetoothctl devices successfully prints to stdout.