bug#20316: 24.5; `string-lessp' doesn't respect value of LC_COLLATE

[Alexis <flexibeast@gmail.com>]

Eli Zaretskii <eliz@gnu.org> writes:

> Emacs is a multi-lingual editor, and it isn't clear how to apply 
> locale-specific settings, including collation, to text that can 
> potentially include many scripts.  E.g., a locale could specify 
> a codeset that doesn't cover characters outside of a particular 
> script, which will produce undefined results if you try using 
> system sorting routines with characters outside of that single 
> script.

Okay, fair point.

> Also, using locale-specific sorting would produce different 
> results not only in different locales, but also on different 
> platforms in the same locale, because the implementation of 
> locale collation order differs from platform to platform and 
> from one C library to another.

Huh, okay.

> Please also note that locale-sensitive collation order could 
> mean more than just order of characters.  E.g., it could specify 
> that punctuation characters or differences in accents should be 
> ignored.

*nod*

> So by default, Emacs sorts disregarding locale-specific 
> ordering, basically using the Unicode codepoints of the 
> characters to order them.

This makes sense given what you've said above, but can this still 
be referred to as 'lexicographic' ordering? To me, 'lexicographic 
ordering' is ordering as per a dictionary for the relevant 
language, not by codepoint for an arbitrary encoding. Is this 
wrong?

> You could use the external 'sort' utility in the meantime, if it 
> supports locale-dependent ordering.  Once again: the results 
> will be not 100% deterministic, even for the same locale, so 
> your users should "caveat emptor".

*nod* Thanks, i'll pass that on. (And note it for the future.)

> May I ask what does that package of yours do that it needs 
> locale-dependent sorting?

The package itself doesn't do it, but the results it produces 
might subsequently require such sorting. The package is 
`org-vcard':

    https://github.com/flexibeast/org-vcard

which allows one to import vCards to contacts in Org-based formats 
(and export contacts from Org-based formats to vCards).

One of the package's users had imported a set of contacts, then 
expected to be able to sort those contacts according to Croatian 
rules, using `org-sort' (from `org.el'). However, to quote the 
user, this resulted in the contacts being sorted:

    according to the English alphabet rules where the contact 
    entries which start with Croatian characters (Č,Ć,Đ,Š,Ž) are 
    at the end of the list, iow. after 'Z' entries, although it 
    should go like this:

    A,B,C,Č,Ć,D,Dž,Đ,..S,Š,..Z,Ž 


Alexis.