I left an important part out of my report: I am using fido-vertical-mode.
So to reproduce:

emacs -Q
M-x fido-vertical-mode
M-x <consta> <-- any search term to narrow down the potential completions,
in this case 12 items remain matching
narrow the whole emacs window so the search results have to "break" because
of long lines
<down> <down> ...
The highlighted active line remains visible until the last items, than the
active line becomes invisible

I hope it's more clear now.

Am Mi., 22. Feb. 2023 um 13:37 Uhr schrieb Eli Zaretskii <eliz@gnu.org>:

> > From: Johann Höchtl <johann.hoechtl@gmail.com>
> > Date: Wed, 22 Feb 2023 07:59:04 +0100
> >
> > I experience the following annoying behavior: If the text in the
> minibuffer get's longer than the display width
> > and lines are therefore continued on the next line, the minibuffer
> scrolling no longer works.
> >
> > What I mean by that is that it "logically" works as when I press <down>
> or <up> the indicator correctly
> > displays the number of the item I am supposed to choose when pressing
> <RET> yet I can't visually see
> > what I would select.
> >
> > First I thought it was a marginalia issue but that's not the case. With
> marginalia it only shows much more
> > easily as marginalia adds text to minibuffer entries thus making lines
> longer. So this is a thing I can easily
> > reproduce when making the whole Emacs window narrow enough to trigger
> continuation lines in the
> > minibuffer.
> >
> > Seems to be an issue with the highlight line logic and scrolling?
>
> Thank you for your report.
>
> To help investigate and eventually fix the issue, please provide a
> reproducible recipe, preferably starting from "emacs -Q" (if
> additional packages are needed, include their loading and activation
> in the recipe).  This will make sure we see and investigate the same
> issue that you are experiencing, and will prevent misunderstandings.
>
> TIA
>