Interesting. I have no idea how he will implement TML over TML without some kind of parser but I guess I will find out. I mean I suppose he could do it but it seems like it would be quite difficult. What I can see from the current Github code direction is you can definitely implement the rules, and you can for sure handle the logic, and I can sort of see how you can do stuff like deductive reasoning and model checking etc.

I admit the self definition part seems to be conceptually the more difficult part. From what I understand the Futamura projection component is what ultimately will allow TML to be implemented in TML. From what you say it seems he will have to reverse engineer the C++ code base and do it again in the TML language which I'm hoping will be simple else we will not be able to even read it.

I guess I can look forward to the Futamura projection which is the next key component I expect Ohad to implement as it's the logical next step. For instance PyPy seems to be an example of what Ohad might be trying to do next where Python is using the Python language to rewrite a better version of itself. The benefit seems to be "automatic compiler generation and optimization". This might explain how TML could be generated from TML itself to be far superior and better optimized than done in C++. Strategically speaking this makes a lot of sense but it also means we might not see or know the benefits until later.

Reference

1. https://en.wikipedia.org/wiki/PyPy
2. https://arxiv.org/abs/1611.09906