Issue: The collation algorithm changes slightly between different versions of php's intl library. The first letters' algorithm we use to display the first letters on category pages is sensitive to these changes. Specifically the first-letters-root.ser file goes with a specific version of the icu library used with php intl extension
Affect: Some of the category headers will be incorrect. Things may switch back and forth between different headers. For example "Aa" was sorted under a U+214D instead of the letter A in MatmaRex's test wiki.
What needs to be done:
*We need to get the version of php intl's extension in use. This will probably have to be done by grepping for ICU version in phpinfo()'s output
*We need to have versions of first-letters-root.ser that correspond to different versions of icu library. These are fairly easy to make (I believe). Simply need to feed maintinance/languages/generateCollationData.php different data files corresponding to different versions of unicode. (Need to double check that icu library is using the unicode files we think it is. I have a vague memory of it changing to some file from the CLDR stuff a couple versions ago)
*Some of the docs about which data files to download for generateCollationData.php need to be updated. Right now its telling people to download the latest version of one file, and the version corresponding to unicode 6.0 for another.