Q: sort keys to string? about twitter-cldr-rb

[jrochkind]

okay, after studying UTS#10 (UCA) for a few hours, I have determined that the sort keys are 16-bit ints, not 8-bit ints.

I'm still confused over endian-ness issues, with regard to how to turn the sort keys into a string having the desired collation properties. Researching/thinking on it.

[camertron]

I believe collation elements can be expressed in hex, so it stands to reason the sort keys can be as well. Remember that sort keys don't have anything to do with character representation, so encoding them as text won't be that useful (and will likely generate some pretty funky characters all over the Unicode spectrum). Have you considered storing the sort key in hex? Two hex strings can be compared easily, and if that doesn't work, you could always consider converting them back into a byte array:

[39, 61, 51, 1].map { |i| i.to_s(16).rjust(2, "0") }.join  # => 273d3301

And back again:

"273d3301".chars.each_slice(2).map { |s| s.join.to_i(16) }  # => [39, 61, 51, 1]

[jrochkind]

Thanks! Hmm, if converted to hex like that, are the output strings still byte-order comparable with desired collation? If so, that might make a lot of sense. I just need a single binary value that can be stored in an external system, and compared for proper collation. Thanks for your tips!

[camertron]

Unless I'm missing something, you should be able to compare two hex-encoded sort keys. The hex digits 0-9 and a-z are conveniently in ascending ASCII order. Here's an easier way to convert (with pack this time):

[39, 61, 51, 1].pack("c*").unpack("H*")  # => ["273d3301"]

[jrochkind]

thanks you for your advice!