I know the language is being kept minimal on purpose, but some of the limitations are surprising.
Bitwise operators are the most obvious. If one of your fixed point numbers is just an i64 behind the scenes, then is should support bitwise ops, which seem really useful on these constrained chips. Even just XOR would allow for simple but efficient transmission encryption.
For strings really anything else would be useful, as they seem pretty brittle to work with right now. Even just an operator to return string length would make life easier. If indexing is too much, then even adding a POP operator (like the string "--" operator but evaluates to the removed character) would make some kind of string parsing feasible.
Speaking of indexing, a version of the memory chip that allowed indexed instead of named fields (`n=0 x=:mem[n]` instead of `x=:mem0`), would save some tedium. I've already seen a proof-of-concept for turning normal YOLOL chips into indexed memory. It just takes 2-3 extra cycles to get or set a field. Directly indexing the memory chips wouldn't so much give extra functionality (besides helping with race conditions), but it would make it easier to work with.
Bitwise operators are the most obvious. If one of your fixed point numbers is just an i64 behind the scenes, then is should support bitwise ops, which seem really useful on these constrained chips. Even just XOR would allow for simple but efficient transmission encryption.
For strings really anything else would be useful, as they seem pretty brittle to work with right now. Even just an operator to return string length would make life easier. If indexing is too much, then even adding a POP operator (like the string "--" operator but evaluates to the removed character) would make some kind of string parsing feasible.
Speaking of indexing, a version of the memory chip that allowed indexed instead of named fields (`n=0 x=:mem[n]` instead of `x=:mem0`), would save some tedium. I've already seen a proof-of-concept for turning normal YOLOL chips into indexed memory. It just takes 2-3 extra cycles to get or set a field. Directly indexing the memory chips wouldn't so much give extra functionality (besides helping with race conditions), but it would make it easier to work with.
Last edited:
