Comment # 8 on bug 64140 from Patrick Ohly

(In reply to comment #7)
> The locale was:
...
> LC_COLLATE="POSIX"
...
> It seems solved the problem with:
...
> LC_COLLATE="ja_JP.utf8"

POSIX probably didn't enable UTF-8 support, thus breaking the 1月 when making
each byte lower-case according to an ASCII character mapping. Any *.utf8 locale
should be fine for searching. Picking the right one for the local country
becomes more important for sorting.