kcgi VS nim-on-samd21

You can use KCGI to use C or C++ as a FastCGI backend.

I like the premise of embedded generating inline c. Kinda like php for c, rather than html

I'm not sure of custom BNF per generating command. I notice they're mostly order independent, which makes it a little easier to learn, but what's the advantage over c syntax? Is it roughly that pointers etc are impossible. Maybe a c-esque syntax would be more learnable

How does the existing syntax compose, i.e. can the user run away with writing tcl-style scripts

If you don't get a lot of feedback, I would guess it is inpart due to custom syntax being an entry barrier to understanding

I don't know, I think html is easy, so people want power with php. On the other hand c is powerful and hard, by the time I'm using that, does Vely's custom syntax give me more than a set of convenient c library calls... such as https://kristaps.bsd.lv/kcgi/

In any case, I like the apis, and I think the premise has legs

It's not that hard, I use kcgi.

The C library used (https://github.com/kristapsdz/kcgi) is portable and working on linux as well. Putting this behind nginx as fastcgi seems very well doable.

C++ has decent industry acceptance in embedded nowadays, or at least that has been my impression.

After C++, rust is likely the most popular, quite a lot of effort has been put into running rust on embedded, see eg https://github.com/rust-embedded. However, once again to my understanding, industry acceptance is still highly marginal.

After that, there's a bunch of toy-ish efforts to run other languages. Zig, nim, python and javascript variants, etc. Usually anything that has C ABI compatibility should be possible to get up and running (without writing a compiler backend from scratch). I've had fun with some toy projects using nim for ARM cortex-M targets (https://github.com/EmbeddedNim/svd2nim, https://github.com/auxym/nim-on-samd21, https://github.com/EmbeddedNim/picostdlib).

Using Nim (and eg svd2nim to generate the equivalent of CMSIS headers for register access in pure nim), it would be entirely possible to write even the low level stuff (SPI drivers and whatnot) in 100% nim, with the same performance as C and better safety (better static type system and compile-time checks, etc). Runtime (eg overflow) checks and garbage collection are available (at the cost of some performance) but optional. See eg. a pretty basic higher-level API for GPIO access, that provides native performance, since the abstraction is implemented as macros (compile-time abstraction): https://github.com/auxym/nim-on-samd21/blob/master/src/port....

To be clear: Ratel isn't my project, just something I'm following due to interest.

In the interest of shameless self promotion :), my own experimentations are :

https://github.com/EmbeddedNim/svd2nim

https://github.com/auxym/nim-on-samd21

And I've used and contributed to picostdlib (https://github.com/beef331/picostdlib), the rp2040 support library.

All just as a hobby, but it's interesting to learn that some companies are actually looking into Nim for firmware! Embedded seems like such a slow moving industry. I believe the author of Nesper and Nephyr also developed them for professional work.