>>> And it turns the cursor red irreversibly in my config (but not in 
>>> 'emacs -Q').
>> 
>> That's rather strange, color-bell--cursor-background is saved only 
>> once, when visual-bell is called for the first time.  I'll try to 
>> reproduce the issue, but some more detailed information (e.g. with 
>> debug-on-variable-change) would be welcome.
>
> I'll try bisecting when I have the time. In any case, it's just an 
> implementation bug, not a blocker.
>

Does the attached patch fix the problem in your config?  It is probably 
safer to check the cursor color each time color-bell is entered.

>>> Is nobody bothered by having this kind of visual indication while 
>>> 'visible-bell' is nil, though?
>> 
>> It's easy to turn off, as indicated in NEWS: (setq ring-bell-function 
>> nil).
>
> I mean, like, semantically: this new proposal is also visual/visible.
>
> But 'visible-bell' is nil.
>

Yes, it's perhaps a bit unfortunate that "visible-bell" is nil in this 
case, but note that with visible-bell t and ring-bell-function ignore you 
also do not have what you could expect.  The semantics of visible-bell and 
ring-bell-function are a bit unclear, but they cannot be fixed anymore 
without introducing backward incompatible changes.

>> And IMO this is way better than the current situation, in which the 
>> default behavior depends on too many parameters that are outside the 
>> control of Emacs, in which the available options are different 
>> depending on the platform, and in which on certain platforms none of 
>> the available options are good.
>
> My #1 preference would be to make it all behave like (setq visible-bell 
> t) on GNU/Linux does. This way we both get a proven behavior with no 
> significant complaints, as well as consistency across platforms.
>

I understand that you're accustomed to what visible-bell t does on 
GNU/Linux, but frankly, its ugly.  Ask their opinion to non-Emacs users 
about that bell, I'd be surprised if they like it.