>> CI AI indexes are much bigger because of ICU implementation of index key derivation.

Compareing those icu depending collations, CI AI is smallest.

http://demo.icu-project.org/icu-bin/collation.html

Input
-------
ç
C
c

Sort case 1 with sort keys: strength=default (equivalent of collate unicode)

Output
-------
c
    2D , 05 , 05 .
C
    2D , 05 , DC .
ç
    2D , 45 A0 , 06 .


Sort case 2 with sort keys: strength=primary (equivalent of collate unicode_ci_ai)

Output
-------
ç
    2D .
C
    2D .
c
    2D .


Sort case 3 with sort keys: strength=secondary (equivalent of collate unicode_ci)

Output
-------
C
    2D , 05 .
c
    2D , 05 .
ç
    2D , 45 A0 .



Hiro