Rust's incremental compilation does this. Unfortunately, it doesn't help with linking, but it speeds up compilation by reusing everything that didn't change.
I could be wrong but I think Rust’s incremental compilation is still pretty coarse. LLVM is where this would need to be implemented and that’s a tall ask (i.e. skipping expensive computation when the AST is the same or close enough to a previously computed result). And Rust’s incremental compilation doesn’t apply to LTO. There’s only so much you can do from outside the compiler toolchain.
