bug#51843: Ispell personal dictionary breaks other dictionaries

[Tor Kringeland <tor.a.s.kringeland@ntnu.no>]

Eli Zaretskii <eliz@gnu.org> writes:

> [Please use Reply All to reply, so that the bug tracker gets a copy.]
>
>> From: Tor Kringeland <tor.a.s.kringeland@ntnu.no>
>> Date: Sun, 14 Nov 2021 16:28:04 +0100
>> 
>> Eli Zaretskii <eliz@gnu.org> writes:
>> 
>> > Isn't this the same problem with the LANG setting in the environment?
>> 
>> I don't think so.  The issues seems to be a disparity between the
>> language codes for the personal dictionary (you can only have a personal
>> dictionary for one language) and the main dictionary.  The header of
>> `/tmp/words.txt' is
>> 
>>   personal_ws-1.1 en 1
>> 
>> where the `en' refers to the language/dictionary and `1' refers to the
>> number of words in your personal dictionary.  If I initiate a personal
>> dictionary with, say, Norwegian, the header looks like
>> 
>>   personal_ws-1.1 nn 1
>> 
>> and I get the opposite problem (that it works for Norwegian but not
>> English).
>
> How do you initiate the personal dictionary?

By setting `ispell-personal-dictionary' to point at a non-existent file.
It then creates the file with the language code for the current
dictionary.

>> I suspect the error comes from `ispell' feeding these two disparate
>> language codes to `aspell': it tries to combine two dictionaries for two
>> different languages and `aspell' complains.  But I'm not certain this is
>> the case, as I don't understand well the code in `ispell.el'.
>
> Does the same problem happen if you invoke Aspell from the shell,
> passing it the Norwegian language and the personal dictionary for
> English?

Yes.  If `/tmp/words.txt' is

  personal_ws-1.1 en 1
  mysterious

running

  echo -n mystiks | aspell -a --lang=nn_NO -p /tmp/words.txt

outputs `Error: Expected language "nn" but got "en".' similar to the
error using `ispell' in Emacs.  Running the following similar command
with `hunspell'

  echo -n mystiks | hunspell -a -d nn_NO -p /tmp/words.txt

gives no complaints and suggests the word is a typo for `mystisk'.

One could argue that the behavior of `hunspell' is not the correct one
here, since we're using dictionaries of two different languages and it
could have yielded a warning at least.  But this has nothing to do with
Emacs.

But given that this is "undefined behavior" for spell checkers one can
use with `ispell', it would be good if Emacs could sanity check the
language code of the personal dictionary with that of the main one and
not include the former if there is a disparity --- or optionally being
able to set personal dictionaries for multiple language codes.