replace-match converts 8-bit chars into multibyte

[Katsumi Yamaoka]

Hi,

I found a strange behavior that replace-match converts 8-bit
characters into multibyte string if the first argument contains
the "\\N" form and it runs in a multibyte buffer.  Here is a
test case:

(let ((default-enable-multibyte-characters t)
      from to)
  (with-temp-buffer
    (insert 128)
    (setq from (buffer-string))
    (goto-char (point-min))
    (looking-at "\\(.+\\)")
    (replace-match "\\1")
    (setq to (buffer-string)))
  (list 'from (append from nil) 'to (append to nil)))
 => (from (128) to (158 2208))

Such a conversion does not occur in Emacs 21.3.  It is odd even
if the buffer's multibyteness is turned on, isn't it?  It does
not occur when fiddling with a string as follows:

(let* ((string (string-as-multibyte (char-to-string 128)))
       (from (copy-sequence string))
       to)
  (string-match "\\(.+\\)" string)
  (setq to (replace-match "\\1" nil nil string))
  (list 'from (append from nil) 'to (append to nil)))
 => (from (128) to (128))

This problem actually occurred, while processing a Japanese mail
message to be sent encoded by shift_jis (i.e., it contained
8-bit characters) in a SEMI MUA.  Although most Japanese mails
are encoded into 7-bit characters, there are some characters
which cannot be expressed with 7-bit.  Moreover, I think it is
natural to encode a Japanese message in a multibyte buffer.

Let me say repeatedly that there is no problem in Emacs 21.3.

Regards,
-- 
Katsumi Yamaoka <address@hidden>