I actually have a toy language that uses emojis for keywords, allows me to reduce tokenizing to basically "split by whitespace, split into runs of characters of

I actually have a toy language that uses emojis for keywords, allows me to reduce tokenizing to basically "split by whitespace, split into runs of characters of the same class (every character is its own class except for [A-Za-z0-9_-] which make up a single class), then post-process tokens for finer distinctions". Strings are somewhat more painful, but using \q instead of \" to escape the double quotes helps.