An alternative TTY layer

Well, indeed. That's why a tiny reimplementation for those small systems only is preferable to uglifying the existing one in an attempt to make all the abstraction layers optional. Throwing out the TTY layer wholesale and replacing it with something "properly architected" (without any evidence that what we have now is poorly architected for its intended use case, which does not include 1MiB RAM IoT boxes), as the original commenter suggested, is quite different.

(A better API without all the horror shows of the existing one around job control, signal handling, ioctls etc would be nice too, but it sounds like minitty just throws most of that out since none of it is needed for its use case -- and for any other use case, the better API would have to coexist with the horrible existing one that everything uses in any case, and because that API is so ancient and widely used there really could never be a sunset period for it -- like it or not, we're stuck with SIOC* forever.)