But none of this is running yet, so I won't speculate any further... I better spend my time implementing it. The paper will be available through ftp in July.
Well, I will just have to wait 'till July!
The reason I started thinking about multiple compilers is that I am looking at the inmos Transputer and how to implement Self ( or as close as I can get to Self ) on it. Seeing how closely the transputer's machine language resembles the bytecodes, I want to see if I can skip the bytecode phase and generate bytecode-like transputer code as a first step. That would run very slowly, but if a better compiler ( optimizer ? ) would rewrite the most frequently called methods we should have a reasonable performance.
I see that a whole series of compilers might be useful, but you would have to gather some statistics to decide when, and what methods, to recompile.
I hadn't thought of background compilation, but it is a fantastic idea! That would go a long way towards making the system more responsive in interactive applications.