Re: emacs 21.1 writes invalidly encoded ISO-2022-JP-2 files

[Kenichi Handa]

Bruno Haible <haible@ilog.fr> writes:
> The reason is that the saved file (de.po, attached) is in fact
> invalid. It contains in line 22 a G2 designation for ISO-8859-1 (ESC
> . A) and a use of it (ESC N ;), and later in line 27 a use of G2 (ESC
> N ;) without prior designation on the same line. But the document
> which defines ISO-2022-JP-2, RFC 1554, says:

>    At the beginning of a line, information on G2 designation of the
>    previous line is cleared.  New designation must be given before a
>    character in 96 character sets is used in the line.

Yes.  This is a bug of Emacs.  iso-2022-jp-2 must be defined
with flag `init-bol'.

---- from docstring of make-coding-system ----------------------------
FLAGS specifies more detailed information of the coding system as follows:

  If TYPE is 2 (ISO-2022), FLAGS is a list of these elements:
[...]
    INIT-BOL non-nil means any designation state is assumed to be reset
      to initial at each beginning of line on output.
----------------------------------------------------------------------

I remembered I fixed this bug looooong ago (in Emacs 20.x).
But, there's no ChangeLog entry for that.  Hmmm, perhaps, I
fixed it only locally.  :-(

Anyway, this is the current definitioin of iso-2022-jp-2 (in
lisp/language/japanese.el).

(make-coding-system
 'iso-2022-jp-2 2 ?J
 "ISO 2022 based 7bit encoding for CJK, Latin-1, and Greek 
(MIME:ISO-2022-JP-2)."
 '((ascii japanese-jisx0208-1978 japanese-jisx0208
          latin-jisx0201 japanese-jisx0212
          chinese-gb2312 korean-ksc5601) nil
          (nil latin-iso8859-1 greek-iso8859-7) nil
 short ascii-eol ascii-cntl seven nil single-shift)
 '((safe-charsets ascii japanese-jisx0208-1978 japanese-jisx0208
                  latin-jisx0201 japanese-jisx0212
                  chinese-gb2312 korean-ksc5601
                  latin-iso8859-1 greek-iso8859-7)
   (mime-charset . iso-2022-jp-2)))

We must insert  " nil nil nil init-bol" after "single-shift".

Could you please try it?

To Emacs maintainers, could someone please install this fix.

---
Ken'ichi HANDA
handa@etl.go.jp