And if it doesn't, then there's certainly a reason to optimize.
This is spot on. When you hear NetBSD developers arguing about proposals (or commits, ex post facto, like here), there is often mention of "users". But these are mostly fictional.
I think it's safe to say most NetBSD users are comfortable working with C.
I reckon it's not much different with OpenBSD: most changes they make are for the benefit of the developers themselves, to suit personal preferences, than for the benefit of other "users". (And that's perfectly acceptable, I think.) Of course, these changes can benefit users beyond simply those who make them and those who approve them, but those users are not the ones driving the changes.
I think this is true of much open source software. Just for the sake of example, a few years ago the Gnome devs removed some printer config options because they decided the "users" didn't need so many options. It's not that Gnome doesn't have users, but the developers seemed to have some notion of a type of user that didn't really exist.
I think more often than not, fictional users are used as justification for the addition of some nice-to-have feature, but of course I'm just speculating.
If you really want to get into OS programming, I recommend you start with "Computer Architecture" by Hennessy & Patterson along with a course in C programming.
Well, it might not be such a bad thing to write your stuff in Lua and get the lights blinking as a way to work your way up to C and mainline device driver development. If the device drivers interest the novice programmer, let's encourage that. Its not like there aren't a ton of safeguards between the novice and getting their driver in the mainline.
I always wondered if a lot of components like this could be either sped up, or moved out of the kernel, using a zero-copy messaging system based on splice() / vmsplice() in the Linux kernel (which works by page table remapping as far as I understand). I guess the prime candidate would be device drivers.
I was looking into embedding lua in a kernel module in Linux, just to hack around with stuff.
How did you strip out the stuff that means it won't compile for kernel space?
> This is partly driven by "we need device drivers and no one understands C anymore"
I don't buy this argument. Who are the people
* Using NetBSD
* having cutting-edge hardware for which drivers don't exist
* deciding to write device drivers
* that have device specs/docs needed to write them
* that don't know/want to use C
* that would do it if only they could write it in a scripting language?
Do these people exist? I assert that the far more common case is that the device specs, documentation, BIOS, etc. aren't accessible. If that's true, they couldn't write the driver regardless of the language, C is hardly the limiting factor here