I use   mule XEmacs 21.5-b28 "fuki" (+CVS-20071205) configured for `i686-pc-linux'.
to edit large number of polish texts encoded in iso-8859-2.

init.el: (I've found this somewhere in the list)
(set-language-environment "Latin-2")
(setq latin-unity-preapproved-coding-system-list '(iso-8859-2))
(latin-unity-install)


locale : LANG=pl_PL.UTF-8

In most cases xemacs recognizes coding system correctly but sometimes
coding system for saving buffer is set to
iso-8859-1 :
Coding system for saving this buffer:
  Latin 1 -- iso-8859-1-unix
Default coding system (for new files):
  Latin 2 -- iso-8859-2
Coding system for keyboard input:
  Latin 2 -- iso-8859-2
Coding system for terminal output:
  Latin 2 -- iso-8859-2

I can even I get :
Coding system for saving this buffer:
  UTF8 -- utf-8-unix
Default coding system (for new files):
  Latin 2 -- iso-8859-2
Coding system for keyboard input:
  Latin 2 -- iso-8859-2
Coding system for terminal output:
  Latin 2 -- iso-8859-2


I think the files are properly encoded ( `iconv -f iso-8859-2 -t utf8` does not complain).
In fact some of them were prepared in xemacs in Latin2 environment.
(usually edit in Latin-2 env -> save -> close -> open again ->  Latin-1)

I redused the problem to a very small (couple of letters) documents and got strange results:


1. if a document contains exactly one small polish letter (there are 9 of them) then coding system is always Latin-1

2. if there are just 2 polish letters then coding system is Latin-2 unless these letters are separated by any string i.e.
for example: it is ok for "wziąć"  but not for  "wzią ć"

3. I could not automaticaly get Latin-2 coding system for documents with exactly 3 polish letters - did't check all posibilites.

4. I could't see any rule. in more complicated cases 

Is this a bug or my xemacs is not configured properly?
Could you please help me or at least sugest where I can get help?


thanks in advance
Krzysztof