Incorrect special character sorting

[vonox7]

The JVM, Swift and all other programming languages (see also unicode specification) mentioned in src/realm/unicode.cpp) define the following character sequence as "sorted:
!"#$%&'()*+,-./123:;<=>?ABCXYZ_abcxyz

In contrast, Realm 3.1.2 considers the following as "sorted":
'- !"#$%&()*,./:;?_+<=>123aAbBcCxXyYzZ

Code for reproduction

The realm sorting algorithm has the following bugs:

• The characters ' and - are before the characters !"#
• The numbers (0-9) are after all other special characters like : and =
• The characters ?, + and _ are in the wrong position

The realm sorting documentation has the following missing:

• Realm sort is case-insensitive (in contrast to Java/Swift greaterThan/smallerThan Operator (< / >))

It seems like realm doesn't comply to any standard (unicode, ascii...) and just sorts all special characters by a random specification and order.

[kneth]

@vonox7 The sorting behaviour you mention is the expected one. I have created realm/realm-java#4527 to clarify. Moreover, I will update the documentation so it is clear tha this is the intended behaviour.

[vonox7]

@kneth Thanks for the update. The documentation update explains the intended lowercase/uppercase problem. But the characters like'-!"#:=?+_ ARE in the utf-8 range (0-591). See also https://en.wikipedia.org/wiki/Basic_Latin_(Unicode_block)#Table_of_characters . I think you missed here my point.

[ironage]

#777 is related

[nalexn]

I'm shocked that such divergence from all the existing specs exists for years unresolved...

How am I supposed to sort contacts by name to form a (#) group after a (z), if (z) is always the last, no matter which character I use for grouping? Have to appeal to yet another hack to make Realm work.