4 ms·

Notes on a Smaller Rust (2019)

3 points by uncircle 1y ago

nickpsecurity 1y ago
So, maybe add a few features to Zig and we'll have a Rust alternative for many use-cases? For borrow checker, I still think a combo of static analysis and concurrent GC's could help. Pair a static analyzer, like Infer, with your code that doesn't borrow check. If it says it's safe, you're good to go. If part of it is safe, other parts get ref-counted or a concurrent GC by default until proven safe. Users get benefits of all safe code in Rust. They can write their own code as fast as with a GC'd language with smaller-than-usual, performance costs. Anything proven safe by static analysis goes faster without the GC. What do you all think?
- steveklabnik 1y ago
  The problem with this strategy is similar to what happens with escape analysis: you make a small change, and suddenly the performance aspects can change drastically. The errors are useful design pressure on a system, not something you actually want to just codegen in a drastically different way to fix.
  - nickpsecurity 1y ago
    That's a good point on performance. It's very nonlinear and unpredictable. Might still be OK if assessed on a case by case basis. Also, any system language should have unsafe for when performance is highest goal.
- potato-peeler 1y ago
  > So, maybe add a few features to Zig and we'll have a Rust alternative for many use-cases? Always wondered why can’t there be a new compiler for C itself which addresses its shortcomings? Cyclone tried to do that and it was successful. Rust itself took the same ideas from it.
  - nickpsecurity 1y ago
    I'm interested in that. Even a Cyclone update with modern, type systems. There was a paper that added meta-programming to C partly for program analysis-like purposes. Everyone ignored it when I shared it. Later, I found out one or more of them built Coverity. Another did a C/C++-compatible LISP that output C. If nothing else, I'd like to see metaprogramming, strong typing with optional dynamic (or type inference), type/linker-safe modules, Rust-like safety, optional GC (concurrent) for productivity, something for concurrency, amenable to static analysis, and outputs normal C (or C++). Much of this might be done with a small set of features. Note: Modula-3, Myrddin, and Zig are all worth looking at for smallish alternatives to C and C++.

3 points by uncircle 1y ago

So, maybe add a few features to Zig and we'll have a Rust alternative for many use-cases? For borrow checker, I still think a combo of static analysis and concurrent GC's could help. Pair a static analyzer, like Infer, with your code that doesn't borrow check. If it says it's safe, you're good to go. If part of it is safe, other parts get ref-counted or a concurrent GC by default until proven safe. Users get benefits of all safe code in Rust. They can write their own code as fast as with a GC'd language with smaller-than-usual, performance costs. Anything proven safe by static analysis goes faster without the GC. What do you all think?

steveklabnik 1y ago
The problem with this strategy is similar to what happens with escape analysis: you make a small change, and suddenly the performance aspects can change drastically. The errors are useful design pressure on a system, not something you actually want to just codegen in a drastically different way to fix.
- nickpsecurity 1y ago
  That's a good point on performance. It's very nonlinear and unpredictable. Might still be OK if assessed on a case by case basis. Also, any system language should have unsafe for when performance is highest goal.
potato-peeler 1y ago
> So, maybe add a few features to Zig and we'll have a Rust alternative for many use-cases? Always wondered why can’t there be a new compiler for C itself which addresses its shortcomings? Cyclone tried to do that and it was successful. Rust itself took the same ideas from it.
- nickpsecurity 1y ago
  I'm interested in that. Even a Cyclone update with modern, type systems. There was a paper that added meta-programming to C partly for program analysis-like purposes. Everyone ignored it when I shared it. Later, I found out one or more of them built Coverity. Another did a C/C++-compatible LISP that output C. If nothing else, I'd like to see metaprogramming, strong typing with optional dynamic (or type inference), type/linker-safe modules, Rust-like safety, optional GC (concurrent) for productivity, something for concurrency, amenable to static analysis, and outputs normal C (or C++). Much of this might be done with a small set of features. Note: Modula-3, Myrddin, and Zig are all worth looking at for smallish alternatives to C and C++.