Improve history browsing with file completions

[clemera]

Ensure the changes only affect the intended purpose and improve general behaviour:

• When navigating history the candidates should be shown again when returning to the start
• The history position should only reset when it was defined before
• There should be no refresh for selectrum-insert-current-candidate in general only when browsing through history

[clemera]

@minad Do you think the current behaviour for file history browsing is sufficient? Or should there be a help message that you have to press TAB to refresh for the current history item?

[minad]

Do you think the current behaviour for file history browsing is sufficient? Or should there be a help message that you have to press TAB to refresh for the current history item?

I am not sure I understand exactly what you mean but I never had problems with file browsing 😆

[clemera]

I mean when you press M-p the candidate list does not refresh, you have to press TAB to do that.

[clemera]

This avoid issues with tramp and unintended hitting of slow file systems when going through the items using M-p.

[minad]

I mean when you press M-p the candidate list does not refresh, you have to press TAB to do that.

Yes, I've seen that. Why is that? It is not a big problem but I always felt that this is some kind of bug, but not a problematic one.

[minad]

Okay, I see - because of TRAMP. I would just keep it as is then. No help message please. Usually I am annoyed by too many messages. I even have the following somewhere in my config:

;; completely suppress some standard messages
(defconst config--suppressed-messages
  '("Mark set" "Mark activated" "Mark deactivated"
    "Mark saved where search started"
    "Back to top level"
    "Beginning of buffer" "End of buffer"))
(defconst config--suppressed-errors
  '(beginning-of-buffer end-of-buffer))
(advice-add #'message :around
            (lambda (fun fmt &rest args)
              (if (and (null args) (member fmt config--suppressed-messages))
                  fmt
                (apply fun fmt args))))
(advice-add #'command-error-default-function :around
              (lambda (fun data &rest args)
                (unless (memq (car data) config--suppressed-errors)
                  (apply fun data args))))

[clemera]

I see 😆, okay no message then but maybe indicating it in the prompt like for describe-face (we do that for completing-read-multiple in general)?

[clemera]

But I would also prefer to avoid introducing ugly handling code because of that. Maybe i's obvious enough that you have to use TAB to refresh.

[minad]

Hmm, I would also not add it to the prompt in particular if it needs special casing. It is really not important. In contrast, in the case of completing-read-multiple I find the message helpful and good, since otherwise it is not obvious that you have a completing-read-multiple. But maybe the message could be slightly more concise?

add more using TAB
TAB to add more
TAB for more
+TAB

[clemera]

That is a good idea, I think I will change it to TAB to add more, that reads better, too I think

[clemera]

Oh, no that is not possible, because sometimes you get a more elaborate message like add more using TAB and crm-separator

[clemera]

Or the word order would just be different then I guess that would be ok.

[minad]

"..and crm-separator", that's an overly verbose message. I did't even understand it before looking at the source. And it is not helpful since it does not show you what the crm separator is. I would probably just show the keybinding since I guess this is what people are going to use. And it is also clear that something special is going on since a keybinding is shown. But it might not be super beginner friendly.

[minad]

You could also change it to , or TAB for more and the character to show for the crm would have to be configured somehow in this crm alist.

[clemera]

If the regexp used is known we already do show the character instead of "crm-separator", that is only used if someone uses a not so common regexp not found in our alist.

[minad]

Ah, I see. I didn't look that closely ;)

[clemera]

I think generally I prefer that history commands don't automatically trigger updates, it find it distracting when using M-p. I already worked on making this the general behaviour but the history behaviour for async commands would need to stay the old way, at least I couldn't see how it make it work. I'm still undecided, UI wise I would prefer it but not sure if its worth it.

[clemera]

An alternative could be to also update for file history by default but check for tramp paths and when there is one don't update and show a message. Either this or what I have written above.

[clemera]

I went with the latter, see #370. Now only tramp paths are not refreshed and a message is shown in that case, I hope the message won't annoy you to much it only appears when using M-p and there is a tramp path.