After the joy of bootstrapping my language, and the subsequent tedious analysis of performance problems, I am programming again!
I failed to produce any interest in my language, which is somewhat expected. Programmers want completed tools, and other language writers are more interested in their own projects, so, I guess all interest will be suspended until I create a performing front-end to gcc or llvm. Which may never happen. It is questionable whether the simple rewrite system employed can be exploited to generate fast binaries. The DOT notation is a straight intermediate between a SECD machine and a TRS/GRS. Possibly, instead of inheriting the best of both worlds, it inherits the worst of both worlds performance wise.
But, ah well, it beats playing Sokoban. Changes I am now concentrating on are getting literal support (including strings) right before, possibly, changing to a ref-counting model.
Refcounting support isn't hard, but implementing it will probably mean wasting a few weeks on debugging. Still unsure about it. I was worried of running out of scratch space during a rewrite, now I get the feeling supporting strings may mean I need less scratch space. Simple examples like [F s -> cons 'x' (cons 'y' (cons 'z' s)) ] actually need more scratch space than [F s -> append "xyz" s] in case the text s is small.
Now doing a bit boring work, the amount of literals just mean a lot more cases on the AST. It would be nicer if I could somehow factor literal support out with direct general support in the language, but it seems that would imply writing a lot of code.