Seems reasonable to me. I was just wishing for such a thing for exactly the
reason you mention (keeping track of zeroes in large integer literals).
AFAICT you've done a pretty good job laying out the possible error
conditions. I think your option 1 makes sense—the error-prone-ness is the
sort of
This PhEP describes the extension of Pharo numeric literals to accepts (and
ignore) underscore characters (`_` ASCII 95).
Many languages (including Python https://peps.python.org/pep-0515/ , Java
https://docs.oracle.com/javase/7/docs/technotes/guides/language/underscores-literals.html
or Ruby)