My package Vcomplete implements this functionality, however I stopped developing it since I reached the conclusion that this isn’t a very useful feature (to me) and all other features of the package were implemented in some form in Emacs. It still works though, and will use the base Emacs implementation for those other features if they’re available.

- Using after-change-functions to implement this can cause weird issues in packages like Tramp and embark which modify the minibuffer during completion.

- If there are a lot of completions, the completions buffer can take a while to update, and you need to make sure this process blocks user input as little as possible.

- I recall having issues with the completion buffer popping up randomly when typing after completing in a buffer, but I don’t remember why it happened and how I solved it.

- There is also the package aggressive-completion which does something similar.

(Sent from a phone, sorry for any weird formatting)

On 13 Oct 2023, at 8:31, sbaugh@catern.com wrote:

It would be nice if there was a built-in customization which caused
*Completions* to update as you type, as long as that buffer is visible.
I imagine such a request has been made before, so what is the obstacle
to adding it?

I would like to figure out a solution which everyone is happy with, and
then I would be happy to implement it. It seems to me that it's just a
matter of, after each keystroke, triggering (for minibuffer completion)
minibuffer-completion-help or (for buffer completion) some new function
to populate *Completions* with all-completions output. This could (I
guess) be done with after-change-functions, although maybe others have a
better idea.

(Btw, if we had this behavior, it also seems like it would help with
another long-time request: asynchronous completion support. A
programmed completion table could internally do something asynchronous
and stateful, and accumulate results over time, and return more and more
results each time Emacs calls all-completions/try-completion. If Emacs
automatically called all-completions with such a programmed completion
table, the resulting behavior would be an acceptable approximation of
asynchronous completion, without having to complicate the programmed
completion API.)