Lars Magne Ingebrigtsen wrote:
>> Yes, I think that should work, but I think using
>> `fill-find-break-point-function-table' instead of "[^\000-\377]" and
>> stuff will probably give you better results when mixing, say, Japanese
>> and Russian.

I see.  It's much smarter.

> I've now tweaked the filling algorithm slightly, and it seems to do the
> right thing in the Han Han group (with Chinese text), but I haven't
> tested it in any groups with mixed European/Japanese text.  What's a
> good test group?

Try gwene groups of which the group names contain ".jp.".  For
instance: gwene.jp.gr.gentoo.gentoojp-news

If you don't think very troublesome, I'd like to recommend
nnshimbun.el.  It makes html articles from contents obtained
from web sites.  You don't have to alter `mm-text-html-renderer'.
My recommendation is:

M-x gnus-group-make-shimbun-group RET asahi RET rss RET

Cf. (info "(emacs-w3m)Gnus"), (info "(emacs-w3m)Nnshimbun")

> Also, what font(s) should I install on Debian to get a display that has
> the traditional kanji-are-twice-as-wide-as-non-kanji buffer?

I believe there should be such fonts, if you've installed all
the fonts Debian distributes. ;-)  I use Fedora and have all the
fonts installed.  My favorites are:

-*-fixed-medium-r-normal-*-16-*-*-*-*-*-iso8859-1
-*-fixed-medium-r-normal-*-16-*-*-*-*-*-jisx0208.1983-0

BTW, the present shr.el code deletes CJK characters that are at
the end of lines, inserts useless SPC between wide characters,
and doesn't seem to do kinsoku.  So, I tried improving them.  A
patch follows.  I think it is near completion.  WDYT?

Though I suspect there are wrong assignments for some Chinese
characters in the kinsoku configuration.  I'll ask Handa-san
later.