If this is a nntp group, what actually encodes it is the news server. For instance, news.newsfan.net uses gb2312 (or possibly gbk). Gnus reads it through the net and uses it as-is internally. Only when displaying it for a user, the newsgroup name is decoded according to `gnus-group-name-charset-method-alist' or `gnus-group-name-charset-group-alist'. Gnus does it for groups based on the other back ends, too. But please note that Gnus trunk supports non-ASCII newsgroup names[1] for only nntp, nnml (including nnagent), and nnrss back ends. In those cases, encoding of newsgroup names is done by Gnus by itself. When reading such an encoded newsgroup name from the ~/.newsrc.eld file, Gnus uses `read' and `eval' (gnus-start.el:2391). Once both Emacs trunk and Emacs Unicode-2 did read it as a multibyte string, and it didn't match the one that was in the active data. That is why I added `mm-string-as-unibyte'. Though it seems to be unnecessary nowadays, it behaves as no-op for a unibyte string, doesn't it? > It's also because Gnus pre-encodes the names when they're read > from the keyboard in gnus-read-move-group-name (lisp/gnus/gnus-sum.el:11785). Because a non-ASCII group name should be an encoded unibyte string for the internal use. But there should be no non-ASCII name in the nnimap groups and pre-encoding doesn't affect those newsgroup names (am I wrong?). > I see 3 problems here: > 1 - The use of mm-string-as-unibyte (I consider any use of > string-as-unibyte to be wrong, unless it is accompagnied by a comment > that explains why it is right). > 2 - Inconsistent encoding: gnus-sum.el apparently uses utf-8 (at least > that's what (gnus-group-name-charset to-method to-newsgroup) returned > in my tests, tho maybe it's because of my locale), whereas > gnus-start.el uses emacs-mule (implicitly, via mm-string-as-unibyte). > 3 - imap.el tries to re-encode in utf7 a folder names that have already > been encoded (with emacs-mule or utf-8). I'm sorry I'm ignorant in IMAP and don't know what encoding in utf-7 is for. But re-encoding of ASCII newsgroup names makes no difference, doesn't it? Although I'm not capable in improving nnimap.el, it might have to decode encoded newsgroup names according to `gnus-group-name-charset-method-alist' or `gnus-group-name-charset-group-alist' before re-encoding in utf-7. [1] (info "(gnus)Non-ASCII Group Names")