Meta-compiler

This is a descent into the rabbit hole suggested by this excellent (if dated) Meta-Compiler workshop.

Running it

The grammar file has a shebang that invokes meta.py, which has a shebang that invokes python3, which in turn is (normally) interpreted by a dynamic linker.

./meta.g

AEXP - Arithmetic expression grammar / compiler / machine

./aexp/aexp.g > aexp/aexp.asm
# actually runs ./meta.py meta.asm aexp.g

./aexp/case1.aexp > aexp/case1.ae
# actually runs ./meta.py aexp/aexp.asm aexp/case1.aexp

./aexp/case1.ae
# actually runs ./aexp/aexp.py aexp/case1.ae

Testing

There is a round-trip test to check that it really is a meta-compiler, second trip round-trip test to see if it produces a meta-compiler and even a third trip round-trip test to see if a fixed point is reached then. (This happens if changing an opcode emitted in the definition. The first step makes code that emits the new, the second step emits it, as does the third. This makes the third result like the second.)

If it isn't a meta-compiler, but produces one, well, continue with that then.

Hints

To swap in new code

./meta.g > m2.asm && mv m2.asm meta.asm

If you have git-test, configure it like this:

git config test.verify "python3 -m unittest discover tests"