@guy038
Thank you so much for all your testing!
As I think you’ve seen, I decided to release a nominally “stable” version that has only a few small changes from the fourth experimental version and with (I think) up-to-date documentation.
I will be taking a closer look at the equivalence classes, and your tests and references will be very helpful. I also need to sort out the situation with digraphs and ligatures. (Unicode distinguishes between the two, and I might be forgetting a third thing — all cases where what we normally think of as multiple characters act as one, sometimes visually, sometimes for collation even when they are visually separate.) Particularly vexing is that this should affect character ranges depending on the active locale; e.g., in Slovak, “ch” sorts between “h” and “i”; in Spanish, it sorts between “c” and “d” and in English, it isn’t a digraph at all, just two letters. So [[.ch.]-i] should match “f” in Spanish but not in Slovak… and in English it makes no sense at all, because ch isn’t a digraph in English.
This sort of thing is why I decided to table the whole notion until I can rest and regroup.