My benchmark for how serious a C replacement is, is when it successfully moves to not embedded and people actually use it.
So far Rust has the smallest of inroads to embedded.
I hope Zig makes it, but I’ve been down this road before.
So far Rust has the smallest of inroads to embedded.
I hope Zig makes it, but I’ve been down this road before.
I currently use Nim for our firmware at work, because at the end of the day —compileOnly means it’s just C. It’s been excellent for ESP-IDF/FreeRTOS, but I’d love to see Zig tackle it as well.
One lovely side effect of using Nim is our “business logic” code is remarkably simple in the firmware, as I got PPPoS working nicely with our Cat-M1/LTE NB-IoT modem, so making network requests is literally a single line of standard-library Nim. Quite lovely!
The less we have to write C/C++ the better, and I’d love to see Zig and Rust do a great job in embedded-land!
Getting to use Nim for your day job sounds like a dream! I'm particularly keen on Nim's UFCS
One of my favourite parts about UFCS is how it can turn C libraries that I've had to bind into nice clean looking interfaces!
esp_err_t sw_enableRx(SwSerial *self, bool State)
proc sw_enableRx*(self: ptr SwSerial, State: bool): esp_err_t {.importc: "$1", header: "<SwSerial.h>".}
var port = sw_open(params, etc)
port.enableRx(true)Nim almost feels like I'm writing a dynamic scripting language, except that I've worked with a typed python code base and it sucked because typing always felt taped on. With generics in Nim I feel like I'm duck typing, but my ducks are compile time checked!
I'm keen on experimenting with protocol oriented programming and as far as I understand, UFCS provides this really neatly to the language. I really like the idea of being able to define my own types and have them seamlessly work with functionality of other libs (and vice-versa) without needing to resort to a language feature like traits.
One piece of advice I can give, is functional programming (which I am a huge proponent of), as defined by a tonne of chained/composed operations on sequences
.map.filter.etc.etc()
You can still use a lot of FP ideas though, but one thing I always remind myself is at the end of the day Nim _is_ C, C is the output, so working within that idea helps a lot in terms of keeping things optimised with little effort on my behalf!
> The standard library can be used unmodified on freestanding targets thanks to explicit allocation, comptime is extremely useful for things like precalculated lookup tables, and the unfinished C backend (part of the stage 2 compiler) means we’ll be able to target exotic architectures that LLVM can’t.