For collate_byname<char>, the collate data section begins with:
$collate_narrow
For collate_byname<wchar_t>, the collate data section begins with:
$collate_wide
The syntax for both the narrow and wide data sections is the same. The data consists of a single string that has a syntax very similar to Java's RuleBasedCollator class. This syntax is designed to provide a level three sorting key consistent with the sorting algorithm specified by the Unicode collation algorithm.