active_support inflector (4.1.0.beta1) raises ConverterNotFoundError in jruby

[jcoyne]

Same as https://jira.codehaus.org/browse/JRUBY-7194

> Encoding::Converter.new(Encoding::UTF_8, Encoding::UTF_8_MAC)
Encoding::ConverterNotFoundError: code converter not found (UTF-8 to UTF8-MAC)
    from org/jruby/RubyConverter.java:162:in `initialize'
    from org/jruby/RubyConverter.java:135:in `initialize'

It was added in this commit:
rails/rails@738dbc0#commitcomment-5248924

[jcoyne]

It appears this transcoding is defined here: https://github.com/ruby/ruby/blob/trunk/enc/trans/utf8_mac-tbl.rb

[headius]

In order to support UTF_8_MAC we'll need to port the whole transcoding subsystem. Currently we're using Java's Charset logic to transcode, and it does not support UTF_8_MAC.

My understanding of UTF_8_MAC is that it prefers to use combining characters rather than single codepoints, so UTF_8 to UTF_8_MAC and back is not likely to round-trip in all cases.

I would suggest that instead of this hack, Rails should use some version of the pure-Ruby String#scrub I implemented (and I think @yorickpeterse improved) from this issue: rubinius/rubinius#2912

Note that this version does not successfully handle all bad characters on JRuby due to incompatibilities in the Charset-based transcoding pipeline (#1459), but for strings with malformed input or no errors, it will work fine and not have the error above.

I will mark this as a bug for JRuby 9k, since by then we should have a proper port of MRI's transcoding logic.

/cc @burke

[headius]

UTF-8 MAC encoding is now in 9k.