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

[Drew Adams]

> We should think hard before adding yet another argument.  I agree that
> the current design is problematic.  Basically, I think that
> `minibuffer-completion-table` should be set buffer-locally in the
> minibuffer instead of being "set" by dynamic scoping.

Apologies for not following this thread, and so
possibly misunderstanding what you mean there.

But my code depends on `minibuffer-completion-table'
being "set" by dynamic scoping.  I update/change the
table dynamically during the same minibuffer reading.

I do this in several ways, for different purposes
(multiple, very different, purposes).

I also change `minibuffer-completion-predicate',
similarly.