Fix tidy_bytes for JRuby

[jcoyne]

The previous implementation was broken because JRuby (1.7.10) doesn't
have a code converter for UTF-8 to UTF8-MAC.

[jcoyne]

@norman @burke can I get your input on this?

[jcoyne]

@headius I wouldn't mind your review either.

[burke]

I wouldn't be surprised if this turns out to be measurably slower. I chose UTF-8 Mac because it's so nearly identical that I figured it should be pretty trivial to transcode, whereas UTF-16 is a completely different encoding.

I don't have any numbers to back this hunch up, however, and even if it is slower, it may not make much of a difference in real cases.

It should, in theory, work. I can't think of any case where a character would be representable in UTF-8 but not UTF-16 (or vice versa).

Comment at the top of the method is in need of minor rewording too.

[jcoyne]

I don't think it should be any slower. It will have a larger memory footprint, but seems better to get this working on JRuby than to save a few bytes.

[tenderlove]

Should we merge this?

[norman]

Has anyone actually benchmarked it? We've all chipped away at this method
over the years to make it fast, I'd hate to take a step backwards. In the
event it's slower on MRI we could just set the transitional encoding
conditionally based on the Ruby platform. I would do it myself right now
but I'm traveling.

Sent from my phone

[jcoyne]

Here's a benchmark run on ruby 2.1.0p0. I don't show any appreciable difference.

https://gist.github.com/jcoyne/8905114

[norman]

Awesome. Let's merge it!

[arunagw]

This will require a rebase with master as well

[jcoyne]

@arunagw rebased.

[burke]

👍 from me too.