I've seen some developers using some_string.bytes.size (or count/length) instead of the constant-time String#bytesize method.

In MRI there is no copy-on-write optimization for string/array conversion, so for large strings, the unnecessary array allocation could result in performance issues

require "objspace"

s = "a" * 1_000_000
initial = ObjectSpace.count_objects_size[:T_ARRAY]
b = s.bytes.count
p (ObjectSpace.count_objects_size[:T_ARRAY] - initial) / (1 << 10)

> 7812

I've opened #474 which will be finished if the idea accepted