A defined collation is active for all server character sets, but they are not applied equally.
You can change the MULTINATIONAL collation for all 256 characters in LATIN, and this definition holds true for those characters whether they occur in GRAPHIC, KANJISJIS, or UNICODE. Note that KANJISJIS and UNICODE have many more characters than LATIN and that the “extra” characters continue to follow the default multinational collation rules.
The collation is two level, which, in the following case specific example, allows you to specify a collation that always produces the following ordering:
- aa
- Aa
- ab
- Ab
A single-level, case blind collation might give that answer sometimes, but could theoretically also produce several other orderings, such as the following, on consecutive queries:
- Aa
- aa
- ab
- Ab