[RFC] type safe trees

[Richard Kenner]

    Are you seriously saying that by the time you've got a cross toolchain,
    target libraries and target os running testsuites on a simulator that
    the complexity of pressing the self-hosting button for a compiler written
    in C++ would be harder than fixing a few bugs that you'd have to fix
    *anyway*?

Yes, I am.

And for a simple reason: the more complex the task, the harder it is to get
done.  The more you have to get working, the harder it is to do so.

    Heck, at that point you've got a cross toolchain that should produce
    the exact same binaries as your self-hosted compiler -- that's a
    simple way of detecting if there's a bug in your self-hosted system.

Theoretically.  And perhaps even in practice sometimes.  But Murphy's Law is
a powerful player in any complex task and can only be ignored at the peril of
the person doing the ignoring.

Doing the first native bootstrap is a complex and difficult task no matterk
*when* you do it.  Sure, the later in the process you can push it, the easier
it'll be.  But that doesn't make it *easy*!