Yes, this is do-able, but the cost is then in maintenance burden (this effectively inlines the function at all the call sites). We can work around this in various ways, I was really asking if people agreed it was a bug (which they don't seem to).

I disagree that the docs are faulty. The code used to work that way. There is a good reason for the code to work that way.

The docs (correctly) liken the utf8 flag to the string/integer tag on a scalar.

I think a change like this is similar to, say, changing all perl numbers greater than a certain size to be held as a string representation "for consistency". It would change the performance characteristics, but not the correctness (assuming the relevant routines to perform numeric operations on strings of digits). People who did numerical work would be upset. Especially if the docs said "perl stores numbers in native format for speed".

Thanks to everyone for their time and their comments. It's interesting no-one agrees with me that this is a problem. I'll take that as a sign to let it lie and we'll work around it as best we can.