bug#39564:

[Eli Zaretskii]

> From: Matt Kramer <mccleetus@gmail.com>
> Date: Sun, 1 Mar 2020 23:20:23 -0800
> Cc: 39564@debbugs.gnu.org
> 
> Followup: The regression in Ivy appears to be fixed when
> set-message-function is bound to nil at the top of the misbehaving
> function.

That is indeed the simplest solution, but it is not the best one.  It
would be better for Ivy to provide its own set-message-function which
plays by the new Emacs 27 rules, i.e. presents the message text in a
way that doesn't completely obscure the original prompt.

> In general, it seems like, given the new defaults defined in
> http://git.savannah.gnu.org/cgit/emacs.git/commit/?id=485b423e8f0df2711a850be7f254665f64ab0bdb,
> it will be necessary to make a similar change to any existing function
> that, say, calls read-key under the assumption that the prompt will
> take over the mini-window. (At least when the prompt contains multiple
> lines). I guess that's the fundamental issue here. The new behavior
> may be a nice improvement, but it's unclear how much code there is out
> there that relies on the old behavior.

Relying on the old behavior was always not a future-proof assumption,
so I see no way around the problem except fixing the code which makes
such assumptions, sorry.