I'm generally in agreement with everything you state. I do think, however, that with the appropriate amount of engineering, one could build an incremental JIT c

I'm generally in agreement with everything you state. I do think, however, that with the appropriate amount of engineering, one could build an incremental JIT compiler that sped up common base idioms meaningfully while executing the whacky stuff using the old perl VM. (Ie. more like psycho, as opposited to a full reimplementation such as PyPy.) Yes, there's a while slew of things you can't do in the case - the old guts and data structures would permeate it. It would be a bit of a horrible maintenance burden. And it's generally not worth it (anymore). There's simply better options available. I played around with something like that, wow, the better part of a decade ago. The approach was to go from the OP tree back to something that had a few of the execution specifics undone such that it resembled an AST a bit more. Then find sections that could be compiled by the partial implementation and convert those. Tracing facilities were there just at a function level. We'd settled on optional type annotations using a keyword plugin that would populate a structure attached to the CV (function struct) with that information as it went through regular perl compilation. Of course, this falls short of any sort of interesting complier work. Anyway, my point is that I think it'd be doable for a sufficiently motivated and skilled (and funded) team of folks, but there's very little economic incentive to do so.