Commented by: @mrotteveel

The collation does actually sort the other characters, but it seems to apply a different sort (if you insert the same values in different order multiple times they are actually grouped together when sorted).

de_de isn't the only collation with this behavior. It also occurs for (only checked for ISO8859_1 collations):
da_da, du_nl, fi_fi, fr_fr, fr_ca, de_de, is_is, it_it, no_no, sv_sv, en_uk, pt_pt

The following sort as expected (based on this small sample):
es_es, en_us, pt_br, es_es_ci_ai, fr_fr_ci_ai

Commented by: @asfernandes

Volker, this is as designed, and you can change this customizing the collation with:

create collation de_de2 for iso8859_1 from de_de 'SPECIALS-FIRST=1';

And then use de_de2 or whatever the name you use. The order will be:

'+'0 %;'-'0 %
'+'0.000;'-'0.000
0
0.00
0.00 %
abc
def
NNN
yyy

Commented by: @mrotteveel

I am curious to the logic that doesn't sort 0 next to 0.00

Commented by: Volker Rehn (vr2_s18)

Thanks a lot, Adriano and Mark. That helps. Now -

1. What's the point of ignoring parts of the charset when sorting? This is rather counter-intuitive.
2. Why is this behavoiur inconsistent even across standard collations?
3. Why not make complete sorting the default and instead have an attribute 'IGNORE-SPECIALS'?
4. and finally, I'm curious too how to sort 0 away from 0.00 ;-)

Regards Volker