Eli Zaretskii <eliz@gnu.org> writes:

> I'm not sure I understand: if we put aside the fundamental opposition
> to using read-from-minibuffer, what problems will be left if we
> temporarily enable recursive-minibuffers while prompting for the
> register?

The only complaint now (once my two missing patches will be applied) is
that we have to use C-d to save a key sequence register e.g. for saving
C-a we have to use C-q C-a.  For jumping to such a register we have to
insert in minibuffer C-q C-a or just use the navigation keys (C-n/p).

>
>> It's up to you maintainers to decide, I think.  Following your request,
>> I've proposed a patch that reverts Thierry's changes, and implements the
>> parts I find useful in a clean and backward compatible way.
>
> Thierry said your patch was incomplete.

It is, the use of read-key is a limitation in the features we provide,
the Eshel patch, as already said have no filtering, no possibility to
configure possible new future register commands, no navigation, no
defaults  etc...
Also it reintroduce a bad implementation of the usage of read-key which
lead to things like bug#27634 (even if this one was fixed, incorrectly IMHO). 

> And I wonder why we need to completely revert his changes.

Reverting those changes would be a regression, I have now a patch that
make register behave as before:

Here the docstring of register-use-preview:

    Maybe show register preview.

    When set to `t' show a preview buffer with navigation and highlighting.
    When nil show a basic preview buffer and exit minibuffer
    immediately after insertion in minibuffer.
    When set to 'never behave as above but with no preview buffer at
    all.

IOW (setq register-use-preview nil) will behave as before.

-- 
Thierry