It seems that ambiguous-wide behaviour (i.e. double-width property for 
characters in the East Asian Ambiguous width category) for CJK locales 
with UTF-8 encoding is inconsistent with Linux locale definitions.
The attached patch changes that. Characters like ─ ü æ are no longer 
wide in the following locales:
ja_JP.utf8
ko_KR.utf8
zh_*.utf8
but only in ja, ko, zh locales with legacy encoding. Explicit modifiers 
@cjkwide and @cjknarrow are not affected.
Thomas