> but also parts of rustc are implemented in a dynamic library ("librustc_driver")

Nit: 100% of rustc is found within librustc_driver-*.so (I mention it in https://news.ycombinator.com/item?id=32329062 - which also touches on stable ABI).

`rustc` just calls into the .so without any custom configuration: https://github.com/rust-lang/rust/blob/e141246cbbce2a6001f31...

(but rustdoc/clippy/miri all have additional configuration, passes, etc. specific to them)

Also, if you look at the `rustc` executable's size:

    $ du -bh ~/.rustup/toolchains/nightly-*/bin/rustc
    17K     /home/eddy/.rustup/toolchains/nightly-2018-01-01-x86_64-unknown-linux-gnu/bin/rustc
    2.9M    /home/eddy/.rustup/toolchains/nightly-2021-12-02-x86_64-unknown-linux-gnu/bin/rustc
    2.9M    /home/eddy/.rustup/toolchains/nightly-2022-01-13-x86_64-unknown-linux-gnu/bin/rustc
    2.7M    /home/eddy/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/bin/rustc

17K is what it should be, I'm not sure where the extra 2.7M are coming from - my current suspicion is that it's entirely statically linked jemalloc, which I guess we didn't want to impose on all users of rustc_driver, but it might be worth doing it anyway.