Caterpillar

The current Caterpillar prototype is focused around the question, if a debugger is a practical solution for getting the language to a useful state relatively quickly. A debugger needs breakpoints, and right now those are implemented by always checking between instructions, if there's a breakpoint at the current one.

Which is fine, for now. But it seems like the kind of thing that the language won't be able to afford in the long-term, for performance reasons. So today, I had this idea: If the language is interactive anyway, meaning I can deploy code into the running program, then why don't I just deploy new code that implements the breakpoint?

Then breakpoints would only cause overhead right where you need them. Which would still leave the overhead of the interactive runtime, of course, but since that is the central premise of the language, I'll have to make that work somehow.