4 ms·

I love new languages (unlike a lot of HN, I'm excited about Rust AND Go), but please remember that C/C++ compilers have been around, sometimes for more than 30

by joaodlf 10y ago

I love new languages (unlike a lot of HN, I'm excited about Rust AND Go), but please remember that C/C++ compilers have been around, sometimes for more than 30 years. Do not think new compilers will just waltz in and suddenly topple all this previous work - That will be a long march.

adamnemecek 10y ago
> unlike a lot of HN, I'm excited about Rust AND Go If there is one thing there isn't enough of on hn, it's discussions about Go
kibwen 10y ago
Rust uses LLVM, which is a backend developed with C and C++ in mind that has been under development for a bit over a decade now. As long as you basically pretend to be C++ at a low level (which Rust does), then the two remaining factors to consider are optimizations that happen in the front end and language-level semantics that enable language-dependent optimizations. (We should also consider that LLVM isn't as venerable as GCC, but it seems that (for C and C++ code) there's no longer any clearly superior choice between the two wrt general-purpose code generation (would be happy to hear arguments to the contrary)).
Manishearth 10y ago
Alias analysis has been the holy grail of compiler development for a while. In C++ it's a global, imperfect, and expensive analysis. In Rust you get it for free. C++ can't do the same thing Rust does because it would be a breaking change to have finer-grained distinctions between the types of references. This is a matter of language semantics, not compiler implementation. rustc isn't doing anything new here, LLVM already knows how to make this optimization and is making it.
- tmccrmck 10y ago
  What specifically makes you think alias analysis is the holy grail of compiler optimization?
  - Manishearth 10y ago
    I spent some time studying the subject a few years ago. Most of the current efforts I saw were about improving alias analysis. Could have been a skewed view of the thing, but the researchers I talked to seemed to feel like it was. Some compiler devs I've talked to do too. In general it does unlock a lot of opportunities when it comes to optimization.

by joaodlf 10y ago

adamnemecek 10y ago

> unlike a lot of HN, I'm excited about Rust AND Go If there is one thing there isn't enough of on hn, it's discussions about Go

kibwen 10y ago

Rust uses LLVM, which is a backend developed with C and C++ in mind that has been under development for a bit over a decade now. As long as you basically pretend to be C++ at a low level (which Rust does), then the two remaining factors to consider are optimizations that happen in the front end and language-level semantics that enable language-dependent optimizations. (We should also consider that LLVM isn't as venerable as GCC, but it seems that (for C and C++ code) there's no longer any clearly superior choice between the two wrt general-purpose code generation (would be happy to hear arguments to the contrary)).

Manishearth 10y ago

Alias analysis has been the holy grail of compiler development for a while. In C++ it's a global, imperfect, and expensive analysis. In Rust you get it for free. C++ can't do the same thing Rust does because it would be a breaking change to have finer-grained distinctions between the types of references. This is a matter of language semantics, not compiler implementation. rustc isn't doing anything new here, LLVM already knows how to make this optimization and is making it.

tmccrmck 10y ago
What specifically makes you think alias analysis is the holy grail of compiler optimization?
- Manishearth 10y ago
  I spent some time studying the subject a few years ago. Most of the current efforts I saw were about improving alias analysis. Could have been a skewed view of the thing, but the researchers I talked to seemed to feel like it was. Some compiler devs I've talked to do too. In general it does unlock a lot of opportunities when it comes to optimization.