Stefan Kangas <stefankangas@gmail.com> writes:

> On closer thought, I think this should be a defcustom.
>
> See `lua-ts-inferior-history' and `sql-input-ring-file-name' for
> example.
>
> I also think nil might as well be a valid value which disables history
> altogether (maybe users don't want that for some reason, let's say
> privacy or whatever).
>
> And the defcustom should be mentioned in NEWS, too.

Thanks, makes sense. Updated in the attached patch!

>>  (defvaralias 'inferior-emacs-lisp-mode-map 'ielm-map)
>>  (defvar-keymap ielm-map
>>    :doc "Keymap for IELM mode."
>> @@ -605,6 +609,11 @@ inferior-emacs-lisp-mode
>>              #'ielm-indirect-setup-hook 'append t)
>>    (setq comint-indirect-setup-function #'emacs-lisp-mode)
>>
>> +  ;; Input history
>> +  (setq-local comint-input-ring-file-name ielm--history-file-name)
>> +  (setq-local kill-buffer-hook #'comint-write-input-ring)
>> +  (comint-read-input-ring t)
>> +
>>    ;; A dummy process to keep comint happy. It will never get any input
>>    (unless (comint-check-proc (current-buffer))
>>      ;; Was cat, but on non-Unix platforms that might not exist, so
>> @@ -631,6 +640,9 @@ inferior-emacs-lisp-mode
>>      (set-marker comint-last-input-start (ielm-pm))
>>      (set-process-filter (get-buffer-process (current-buffer)) 'comint-output-filter)))
>>
>> +(unless noninteractive
>> +  (add-hook 'kill-emacs-hook #'comint-write-input-ring))
>> +
>
> I'm not so sure about this part.  If this is correct, why should it be
> defined here, and not, say, in comint.el?  It seems like a separate,
> unrelated change.  What am I missing?

Usually comint-derived modes use a process sentinel to listen to some
external process and call `comint-write-input-ring' when the process
exists, which works both the buffer is killed or Emacs itself is quit.

For IELM there is no such process to listen to, so I hook into
`kill-buffer-hook' and `kill-emacs-hook' to achieve the same things.