The current plan is to write the core of the language in postFORTH itself, and use a Python script to compile it. When the language has reached a certain stage of development, it will then be able to compile itself.
Some ideas are presented here, in no particular order.
Some of the goals: Easy to understand quickly how the system is built and how it runs: transparency. Well commented, self-documenting. Very small core, easy to port. Easy to disassemble/decompile. Good as a student language and for serious applications. Use C operators where possible, to attract C programmers into the Forth fold; Forth operators are no more and no less inscrutable as those of C, so why have two completely different sets of obscure gobbledygook?
Put a back-pointer in front of each block of code to its symbol table entry. That way, even if a word is redefined, its old embodiments can be located. Maybe a version number for each word can be stored there too. I was thinking of storing the elf_hash of each word before the code block too but that would be wasted space; it should be in only one location. Calculating the hash is not an expensive operation, and this might be unnecessary anyway.