In qlocale_data_p.h there are diverse string tables referenced by the QLocaleData that describes locales; various fields in these structures are recorded by an index and length of a substring of the table, at which to find the string data. At present the strings for the string tables are generated during output of the locale_data table (the string tables are output later) by: when about to output a string, search for it in the string table's string; if present as a substring, return index and length; otherwise append and return index and offset.

If we made a first pass, before generating locale_data, over the strings, collecting them up into a set, we could then compile each of those sets into a minimal string that has every string in its set as a sub-string. This would result in smaller string tables (and thus indices into them).