Recent searches
Search options
Tactless: "Rust in the kernel is a cancer."
Elegant: "Rust in the kernel is carcinization."
@lispi314 I'm not sure what you mean but let me just have my pun or whatever you call it without bringing up the actual argument... 
Just felt like poking fun at the kernel being written in C when better Libre options were available the entire time.
@lispi314 Anyway IMO the problem is not that the kernel is written in the wrong language. It's that shit device drivers run in the same privilege domain and memory space as the kernel rather than being properly sandboxed.
Sure, something needs to manage the capabilities or references in a memory-safe language that needs to handle hardware, but none of the other parts need to have such access.
With that handled, there isn't any security-based need for multiple address spaces. Heterogenous memory could still be useful for caching and such though, of course.
@lispi314 Yes, memory-safe languages and privsep are two very different philosophies to achieve similat ends, and I vastly prefer privsep because it lets you use any existing code and lets people write in the languages they're most comfortable in, and because it can protect against not just errors representable in your type system but also logic errors.
seL4 has one for C. Memory-safe languages permit for much less awkward interfaces to use them though. They are complementary, not in opposition.
But even Pascal has significant safety improvements over C, as does Modula-2 & Oberon. In some aspects even Fortran (late-ish versions, early ones no) did better. One really didn't have to go far off the path to find better.
I tried to use Pascal and Modula 2 for non-trivial programming in the '80s and they were non-starters. I couldn't even write code that would reliably do I/O on two different compilers on the same computer. It was easier to write cross-platform code in Fortran. That's why the original Software Tools book was revolutionary and Software Tools in Pascal was completely pointless.
It's like the Godel Incompleteness Theorem. A systems programming language is either unsafe or so constrained as to be useless. Fortran is relatively safe because it has no pointers. Swift is close to a Godel-busting language. Rust may be closer. But Pascal and Modula 2 were not in the race.
> But Pascal and Modula 2 were not in the race.
No, I named them more as examples of "even this little is still better than what C does", at least by spec. People probably promptly went and used unsafe options disabling some of their semantics for performance and then went "woe is me" when something went wrong.
Safe languages contemporary to C were kind of few and required expensive non-Free compilers for the most part.
Ada for example /existed/ but couldn't meaningfully be used unless one had military contractor company amounts of money. So it might as well not have been considered to exist before early 2000s.
To avoid that one would otherwise need a hosted architecture more similar to Singularity or Midori's, with the hosting core being in one partly-safe language, and the rest being in a safe language with truly no access to anything else not wrapped & provided by the other as an API.
However, most of the Free alternatives I know unfortunately only came to be a thing a few years after Linux had already gotten started (Ada was only added to the GCC in 2001~), possibly partly as a consequence of it even (poses a bit of a chicken & egg problem; this isn't Ada's case, the project was already ongoing).
The notion that C was possibly one of the few with truly Libre compilers readily available is saddening. What an utter mess.
I don't think it would have been possible to develop a useful portable language with Pascal as a base. There were too many unavoidable restrictions built into the language design. There were Pascal-like compilers that were closer to being usable but things like Turbo Pascal weren't actually Pascal and pretty much abandoned any pretense at safety.
