bug#45474: Icomplete exhibiting in recursive minibuffer when it shouldn’t

[Stefan Monnier]

>> I'm more and more thinking that we should really bite the bullet and go
>> for a "really buffer-local" setup, such as in the patch below.
>> I'd be surprised if it doesn't introduce backward incompatibilities, but
>> at least it's "The Right Thing" to do.
> I've tested your new patch with tests that Gregory posted
> in an earlier message that cover all cases of nested
> completion and non-completion minibuffers:

AFAICT my proposed patch should work fine with the builtin UI and with
icomplete, yes.  I expect ido-ubiquitous should also work just fine.

The only significant source of incompatibility I can foresee is for code
which uses `minibuffer-completion-*` from code running outside of the
minibuffer (such as in a buffer like *Completion*).

There's another source of incompatibility for functions that currently
operate like `completion-read-default` (i.e. let-bind
`minibuffer-completion-*` and then use that setting in another buffer,
such as a new minibuffer): these may fail to work when used from within
a completing minibuffer.


        Stefan