Arduino Joins Zephyr Project(zephyrproject.org) |
Arduino Joins Zephyr Project(zephyrproject.org) |
Set https://docs.zephyrproject.org/latest/build/kconfig/tips.htm... and the issue: https://github.com/zephyrproject-rtos/zephyr/issues/52575. And don't even get me started on trying anything to group options together.
Since Kconfig is used to configure the kernel, why aren't the kernel devs experiencing similar issues with their configurations?
Multi core SoCs with different archs and memory protection schemes all sharing a bus. This complexity can be wrangled with device tree and Zephyr readily and has good company with other large silicon vendors being contributors.
I for one am looking forward to seeing more contributions and companies joining.
The upside is that you no longer have to essentially write an entire OS if you just want to make a light blink, making it way easier to get started with a complicated project. The downside is that all those layers of abstraction make it really hard to figure out what is actually going on, and they can quickly get in the way when you are trying to do something which isn't a 1:1 match to the high-level model used.
Zephyr does have some obvious downsides: it uses a custom build tool (West) to in turn use KConfig to select various options for CMake which then uses Ninja or Make to build the actual source. That means the build process is complex, hard to understand fully, and difficult to troubleshoot. The included libraries and drivers are usually good, but as with any code it's always harder to understand code you didn't write (or wrote a long time ago), and tend to be a higher level of abstraction than you might need (and thus a bit less efficient in some cases than you'd want).
Arduino's base APIs with the Arduino "IDE" are similarly opaque. Zephyr will make quite a few things easier, since it'll come with device tree files for the various Arduino boards.
If you're doing particularly simple projects, Zephyr is overkill. If you've got multiple interacting components, an RTOS becomes extremely helpful.
Hopefully that makes sense, I'm new to all of this.
Bluetooth is an absolute nightmare if you don't understand the majority of what's going on (which in and of itself takes... a lot of time). There's a bunch of logic going on and most of it is handled in callbacks that you will never see, except the dreaded timeout/failure to handle print at the end of the main loop. Zephyr will ease a lot of those painpoints, with the understanding that you're ignoring a lot of the machinery humming under your feet.
Things that stood out to me:
0. If this is your first embedded project / you're actually new to all of this, get ready to take a beating. "Abandon all hope, ble who enter here."
1. Do yourself a huge favor and get two* dev kits. Nordic provides walkthroughs on getting setup with two flavors of code (zephyr, or low-level drivers). Each has a tutorial for uart forwarding/handling. Expanding on that tutorial will take a lot of futzing around, or actually learning what the machinery is doing under the hood. Learning the stack did not come naturally / I found it very difficult. Why two? Two lets the bluetooth abstraction handle itself / you don't have to deal with it right away when you're getting started.
https://www.nordicsemi.com/Products/Development-hardware/nRF...
2. If / when you want to attach your bluetooth device to something more useful (e.g. a computer or mobile device), do yourself a second huge favor and develop using linux + a laptop. I tried to do development on windows + WSL and there were many, many hangups with the hardware handoffs from PyBlues to the local bluetooth drivers. Maybe it's gotten better, but I doubt it. My other alternatives were driving direct from Windows bluetooth libraries (behemoths that would take time to setup / understand), or develop for mobile (which also will take time to setup / understand). Neither was an enjoyable experience.
so in summary:
Is Zephyr overkill? - Absolutely. But the path is well trodden.
Regardless, Zephyr is an RTOS which provides OS functionality like scheduling, interrupt handling, semaphores, etc.
It is most likely overkill for what you are attempting to do. You should instead look at the vendor provided SDK for the nrf52 chip and program it bare metal. The SDK is most likely just libraries/drivers and does not come with an RTOS.
Would something like circuitpython not be easier to work with?
Maybe I'm just doing it wrong with just trying to add the dependencies I want. It might work better if I used the menuconfig GUI. To me though it seems if I decide I want to enable 'CONFIG_NET_TCP' on my project, I shouldn't also have to specify 'CONFIG_NETWORKING'.
As for why it works better in the kernel, that's a good question that I don't know the answer to as it's been years since I tried to build a kernel with a minimum configuration.