> Performance is like money: it's easy to squander and hard to acquire. In this case, there were several decisions to not care about performance right now in o

> Performance is like money: it's easy to squander and hard to acquire. In this case, there were several decisions to not care about performance right now in order to emphasize correctness and shipping faster, while making sure there are no technical obstacles to making compilation faster in the future. The main time sink in Rust compilation is that all the abstractions in the Rust code get compiled into the initial bytecode representation passed to the LLVM side, and they are only reduced there, instead of cutting down on the hierarchies on the Rust side. This costs performance in several places -- the creation of all the bytecode, the copying of it, and then LLVM parsing it all in. The upside of doing it this way is that it makes the Rust-specific compiler much simpler and easier to implement, and that the optimizations that remove the towers of abstraction on the LLVM side are extremely well tested. As Rust matures, optimizations that reduce the complexity of the created initial LLVM bytecode can be, and probably will be done on the Rust side.