Remake Controller Emulation

[zxzxwu]

I am considering whether we should remake our controller emulation. The current implementation majorly serves our local unit tests, but it seems many users also use it to test their own host stacks, probably even more popular than RootCanal (if not considering its bundle with Android Emulator).

An significant advantage to emulate Bluetooth controllers in Python is - of source - easy to configure. Due to its original requirement to run inside Android, RootCanal is written in C++ and Rust, but C++ and Rust are not as easy as Python - they don't have so many packages ready to use like PIP, or fire-and-forget asyncio, and building on Windows is a big problem. Besides, Bluetooth emulation doesn't need extreme performance usually, so C++ and Rust cannot play their best here. (The same reasons are also valid for BlueZ btvirt as well)

However, I think just refine the current rule-based emulation and make it a Python version of RootCanal or btvirt is not enough. A better emulation should also simulate the broadcast nature of wireless communication. Real controllers and firmwares also send and receive a lot of LL/LMP packets from the air, and properly filter those useful for themselves. I guess this could help simplify the implementation, because we don't need each controllers to monitor all the others, so that lifecycle management will be simpler. Besides, the current emulations don't properly simulate the relationship between intervals and windows.

[barbibulle]

It would be nice to have a pure-python equivalent to RootCanal. But it is a lot of work, if you want to be full-featured (Classic + LE, with ACL, SCO, ISO, multi-phym ranging, ...).
Using RootCanal via its python bindings (thanks to @hchataing rootcanal is easy to install from PyPI at least for macOS-arm and Linux-x86_64) is a good alternative. I would love to see more pre-compiled binaries for rootcanal over time (at least Windows and maybe other Linux targets).

[hchataing]

Currently the binary releases are limited by the available Github action runners.
Supporting Windows build is yet another matter due to the internal use of sockets.
Feel free to create a tracking issue on https://github.com/google/rootcanal/tree/main about supporting additional platforms.

[dlech]

[A] significant advantage to emulate Bluetooth controllers in Python is - of [course] - easy to configure.

Yes, this is the big appeal of Bumble being pure Python.

Being able to customize every aspect of every possible operation is important to be able to emulate real-world Blueototh adapters and peripherals. Real-world devices are not always well-behaved and follow the rules. So to be able to override the default implementation in just a few key places makes it really easy to reproduce the quirks of real-world hardware without actually needing any hardware.

[hchataing]

RootCanal is somewhat customizable with the use of the protobuf configuration file. https://github.com/google/rootcanal/blob/main/proto/rootcanal/configuration.proto
There is a list of quirks added to emulate non standard behaviors obversed in real life controller.
The list is short right now but feel free to propose new additions.