The reason why Rust and other languages have bounds checking that approaches C’s speed is less to do with the compiler being able to prove away bounds checks and more to do with those languages making the bounds easy to carry around.
Bounds checks themselves aren’t that expensive, if you already have the length to check against in a register or some other convenient spot.
And - if the bounds are convenient to find and known to be immutable then it does make it easier for the compiler to eliminate those checks using the same techniques it would use to eliminate all kinds of redundancies. C has many redundancies that we don’t have to think about because they’re straightforward for the compiler to reason about.
Like many things in Rust vs C/C++ comparisons, it’s about defaults: Rust has dedicated syntax and everyone uses it, in C and C++ it’s more manual and not everyone does.
But C/C++ pointers don’t. You can add bounds checking to all C/C++ pointers and there are many projects that do that (CCured, SoftBound, CHERI, Fil-C, CheckedC, -fbounds-checked) but they all come at cost (new hardware, language changes, or reduced perf).
I'm really excited to see something like Xr0 https://xr0.dev/learn get some traction. Feels like the best of both worlds. You a low-level language with optional safety guarantees. Especially for legacy code bases that shouldn't be replaced, Xr0 would be gold. I love Rust, and it does so much more than just memory safety, however there is a place for C if the memory safety can be fixed.
Feels like most people reacting to this think it's about C vs Rust. And part of it is, but the bigger conclusion is: If you're starting a new programming project, you probably shouldn't be using a systems language at all. Go, Java, C#, Swift, Python, or Ruby are probably better choices for most projects.
Go in particular has a lot of the advantages of C/C++ and Rust--fully-compiled binary, easy access to system calls, high performance--without the overhead of needing to constantly think about memory management.
I work for a huge company which acquired two startups in the same domain. One was using exclusively Go and the other Rust. They merged them together to create a unified product. To this day, the codebase is still using these two languages.
The rust part of the product is significantly better, the code is much saner and the team velocity is also much faster.
I've been meaning to turn this into a proper article, but Rust has some really nice advantages once you get comfortable with it, even in programs where performance/embedding isn't important:
"Academia stopped teaching C++, moving to Java and then to Python."
There are a lot more practical and lucrative skills for a young programmer to focus on than C or C++. Many will be able to have good careers without going that low level.
What impact will that have on the economics of legacy C/C++ code bases in the 15-30 year time frame?
Zig is probably not so much popular as go or rust. But it seems like a great middle ground. It interops with existing c code. Even has it own compiler for c. Bun project is one of the major projects using it.
19 comments
[ 3.0 ms ] story [ 63.2 ms ] threadFor comparison "C++ shared_ptr is only safe with tree-shaped structures" would be getting-at-the-truth-but-kinda-weirdly-worded in some similar ways.
Bounds checks themselves aren’t that expensive, if you already have the length to check against in a register or some other convenient spot.
And - if the bounds are convenient to find and known to be immutable then it does make it easier for the compiler to eliminate those checks using the same techniques it would use to eliminate all kinds of redundancies. C has many redundancies that we don’t have to think about because they’re straightforward for the compiler to reason about.
But C/C++ pointers don’t. You can add bounds checking to all C/C++ pointers and there are many projects that do that (CCured, SoftBound, CHERI, Fil-C, CheckedC, -fbounds-checked) but they all come at cost (new hardware, language changes, or reduced perf).
Go in particular has a lot of the advantages of C/C++ and Rust--fully-compiled binary, easy access to system calls, high performance--without the overhead of needing to constantly think about memory management.
The rust part of the product is significantly better, the code is much saner and the team velocity is also much faster.
- enums
- const-ness without immutable types
- no null
- no iterator invalidation
- no surprising captures in closures/lambdas
- no "spooky action at a distance"
- Cargo
There are a lot more practical and lucrative skills for a young programmer to focus on than C or C++. Many will be able to have good careers without going that low level.
What impact will that have on the economics of legacy C/C++ code bases in the 15-30 year time frame?