No, I didn't mess it up, I demonstrated exactly what I wanted to demonstrate: that the same character has two different possible internal representations in perl (notice they compare eq for perl). And that these two representations give a different result for substr() under use bytes.

You can for example use Dump() from Devel::Peek to see that internally they are different of course. But code shouldn't depend on how the string happens to be encoded internally if it can be avoided.

Your example however leaves both $a and $b with the same internal representation (non-utf8), so of course they print the same. It also isn't related to my point anymore. Notice that the "but $b is internally in UTF8" isn't actually true for your code.

update after reading pg's reply

I wasn't trying to "disprove" use bytes, it obviously does what it is supposed to do. I was however trying to show that it makes the result depend on the internal representation. Our disagreement is about first if that's a good idea and second if that's what the OP wanted. You obviously think he wants the n-th byte of the internal representation of the string, while I assumed that if he's talking about unicode (which I'm still not sure of), he'd want the n-th byte of the UTF8 representation of the logical string.