Joe Corneli <jcorneli@math.utexas.edu> writes:
> Adapted from w3m-filter.el:
>
> (while (re-search-forward "&#\\([0-9]+\\);" nil t)
>   (setq ucs (string-to-number (match-string 1)))
>   (delete-region (match-beginning 0) (match-end 0))
>   (insert-char ucs 1))
>
> This would appear to work if the characters themselves were recognized...
>
> But when I run this expression on a buffer containing the string
> "&#29572;&#22872;" what I get is an error, like this:

Is that really what w3m does?  I'm not sure how the above could possibly
work in any normal version of Emacs -- the argument to `insert-char' is
an Emacs characater, not a unicode code-point.  So, you need to
translate from the unicode code-point to the Emacs character encoding.

One method might be to translate the unicode code-point into a utf-16
string (should be trivial I guess), and then use `decode-coding-string'
to translate that into Emacs' internal encoding; e.g.:


 (while (re-search-forward "&#\\([0-9]+\\);" nil t)
   (let* ((ucs (string-to-number (match-string 1)))
          (ucs-string (string (logand ucs #xFF) (logand (ash ucs -8) #xFF)))
          (decoded-string (decode-coding-string ucs-string 'mule-utf-16le)))
     (delete-region (match-beginning 0) (match-end 0))
     (insert decoded-string)))


For me, this does the right thing on your example, and on the text of
that wikipedia page:

   The fictional character Xuanzang ($B8<Ty(B, WG:  Hs.AN|an-tsang), a central
   character of the classic Chinese novel Journey to the West ...


It probably will only work well in recent CVS versions of Emacs
that have `utf-translate-cjk-mode' turned on by default though. [*]

-Miles


[*] In the current CVS Emacs, there seems to be a function that does
    this translation directly too, `utf-lookup-subst-table-for-decode'
    but given the odd name, it's probably not intended for general
    use...

-- 
Love is a snowmobile racing across the tundra.  Suddenly it flips over,
pinning you underneath.  At night the ice weasels come.  --Nietzsche