RE: Any objection to adding completing-read-function?

[Drew Adams]

> There's been such requests in the past, which I've usually resisted.
> But I guess it's OK to do such a thing.

Good.  Let's please do it.

> a variable completing-read-function should not allow
> a nil value, instead its default value should be
> `completing-read-default' which is the current completing-read, so you
> can always funcall completing-read-function without checking if
> it's nil.

Then that should have been the approach for `read-file-name-function' as well.
These two should be kept similar (purpose, behavior), IMO.

There are already places (both in vanilla Emacs code and in 3rd-party libraries)
that bind `read-file-name-function' to nil to get the default behavior.

So we have a choice of asymmetry between `read-file-name' and `completing-read'
or breaking existing code and backward compatibility.  Or just doing what Leo
did: use nil for the default behavior.

I agree with what you say about not having to check for nil.  But that's where
we are at present.