How to determine encoding for file?

How to determine encoding for file?

[kj]

I've downloaded a large file that is supposed to contain a mixture
of Japanese and English (it's basically a learner's dictionary).
The English is displayed correctly, but not so for the Japanese.

I've tried setting the buffer's coding system to utf-8,
japanese-shift-jis, japanese-shift-jis-mac, japanese-shift-jis-dos
(just guessing).  None worked.

In fact, I'm not even sure that any of these changes of the coding
system achieved *anything*, since the buffer's appearance remained
unchanged throughout all this mucking around.  I used the command
set-buffer-file-coding-system to do this.  Should I need to do
anything besides re-setting the coding system to see a change in
how the file is displayed?

More importantly, is there a better way to determine a file's
correct coding system besides trial and error?

TIA!

~K

Re: How to determine encoding for file?

[Pascal J. Bourguignon]

kj <no.email@please.post> writes:

> I've downloaded a large file that is supposed to contain a mixture
> of Japanese and English (it's basically a learner's dictionary).
> The English is displayed correctly, but not so for the Japanese.
>
> I've tried setting the buffer's coding system to utf-8,
> japanese-shift-jis, japanese-shift-jis-mac, japanese-shift-jis-dos
> (just guessing).  None worked.
>
> In fact, I'm not even sure that any of these changes of the coding
> system achieved *anything*, since the buffer's appearance remained
> unchanged throughout all this mucking around.  I used the command
> set-buffer-file-coding-system to do this.  Should I need to do
> anything besides re-setting the coding system to see a change in
> how the file is displayed?
>
> More importantly, is there a better way to determine a file's
> correct coding system besides trial and error?

No, there is no better way.

From the sequences of bytes you can find inside the files, you may
eliminate some encodings (eg. some byte sequences are invalid in
UTF-8), but there are several encodings that use all the byte values
and where any sequence of byte is valid, so you cannot choose between
these encoding without knowing the meaning of the file.

For example, the following file:

$ od -t x1 -c /tmp/a.txt
0000000  50  72  69  63  65  20  6f  66  20  74  68  65  20  69  74  65
          P   r   i   c   e       o   f       t   h   e       i   t   e
0000020  6d  3a  20  31  34  20  a4  0a
          m   :       1   4     244  \n
0000030


The encoding cannot be determined without knowing what item is
referenced, and what price ranges are probable for this item.  It
could contain:

Price of the item: 14 €

or:

Price of the item: 14 ¤

(It could also contain another character, but it would be less probable).


-- 
__Pascal Bourguignon__                     http://www.informatimago.com/

Re: How to determine encoding for file?

[tomas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Sun, Jan 24, 2010 at 10:59:46PM +0100, Pascal J. Bourguignon wrote:
> kj <no.email@please.post> writes:
> 
> > I've downloaded a large file that is supposed to contain a mixture
> > of Japanese and English (it's basically a learner's dictionary).
> > The English is displayed correctly, but not so for the Japanese.
> >
> > I've tried setting the buffer's coding system to utf-8,
> > japanese-shift-jis, japanese-shift-jis-mac, japanese-shift-jis-dos
> > (just guessing).  None worked.
> >
> > In fact, I'm not even sure that any of these changes of the coding
> > system achieved *anything*, since the buffer's appearance remained
> > unchanged throughout all this mucking around.  I used the command
> > set-buffer-file-coding-system to do this.

This won't do the trick (see below for what will do). This function just
says: "forget you loaded this file as shift-JIS. From now on it will be
UTF-8" (for example). So it doesn't change anything, but when you save
the file, it will be transformed to the new coding system (if possible).

> >                                            Should I need to do
> > anything besides re-setting the coding system to see a change in
> > how the file is displayed?

You'll have to use `revert-buffer-with-coding-system' (by default mapped
to the key seqence C-x RET r). This will reload the file under
assumption of the new coding system.

> > More importantly, is there a better way to determine a file's
> > correct coding system besides trial and error?

Pascal answered this part better than I could :-)

There will be always lots of byte sequences valid under several coding
systems (but meaning different things). The methods out there to get a
grip on the problem are heuristic, partly based on statistical
properties of the text. If you want to have some fun understanding the
kind of problems involved, have a look at [1]. For an implementation in
Emacs  Lisp, see Unicad [2]

- --------
[1] <http://www.mozilla.org/projects/intl/UniversalCharsetDetection.html>
[2] <http://www.emacswiki.org/emacs-en/Unicad>

Regards

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFLXTLXBcgs9XrR2kYRAnR5AJ9Jowgc9pPrCaW0lRe1Tv7xFGya+QCfRXJ8
mLTW2GBvke8OYbVdWiVcrcU=
=gJuQ
-----END PGP SIGNATURE-----

Re: How to determine encoding for file?

[kj]

Pascal, Tomas, many thanks for your comments and suggestions!  I
eventually resorted to a brute-force solution: I created many copies
of a fragment of the file, with a different -*- coding: ??? -*-
line at the top, and opened all these copies in Emacs.  Then I
scanned all of them using *Buffer List*.  Dumb but effective.

~K

Re: How to determine encoding for file?

[Kevin Rodgers]

kj wrote:
> I've downloaded a large file that is supposed to contain a mixture
> of Japanese and English (it's basically a learner's dictionary).
> The English is displayed correctly, but not so for the Japanese.
> 
> I've tried setting the buffer's coding system to utf-8,
> japanese-shift-jis, japanese-shift-jis-mac, japanese-shift-jis-dos
> (just guessing).  None worked.
> 
> In fact, I'm not even sure that any of these changes of the coding
> system achieved *anything*, since the buffer's appearance remained
> unchanged throughout all this mucking around.  I used the command
> set-buffer-file-coding-system to do this.  Should I need to do
> anything besides re-setting the coding system to see a change in
> how the file is displayed?

Use `C-x RET r' (aka M-x revert-buffer-with-coding-system).

> More importantly, is there a better way to determine a file's
> correct coding system besides trial and error?

-- 
Kevin Rodgers
Denver, Colorado, USA

Re: How to determine encoding for file?

[Thien-Thi Nguyen]

() kj <no.email@please.post>
() Mon, 25 Jan 2010 14:55:06 +0000 (UTC)

   Dumb but effective.

Out of curiosity, how many rounds did it take?
What was the encoding found, in the end?

If such a method is unavoidable, perhaps Emacs should have:

(defun interactive-search-encoding (&optional candidates)
  "Search `coding-system-priority-list' interactively.
For each encoding, do `revert-buffer-with-coding-system', and
query the user if the result is acceptable.  Stop looping if so.
Optional (prefix) arg CANDIDATES is a list of encodings (symbols)
to try.  Return non-nil if an acceptable encoding is found."
  (interactive
   (list (let ((raw (read-string "Try (space-separated) encodings: ")))
           (when (stringp raw)
             (mapcar 'intern (split-string raw))))))
  (unless candidates (setq candidates (coding-system-priority-list)))
  (let ((revert-without-query '(".*")))
    (find-if (lambda (coding)
               (revert-buffer-with-coding-system coding)
               (when (y-or-n-p (format "%s acceptable? " coding))
                 (message "buffer-file-coding-system now %s"
                          buffer-file-coding-system)
                 t))
             candidates)))

Emacs does have `select-safe-coding-system', but it is not interactive.

thi