I thought we'd fixed \iref so that we can break before it, just with a high penalty? :(

Except that, when I tried it, it didn't work in this case.

Perhaps we could add an optional parameter to \iref to specify its \nolinebreak penalty. I'm sad that it doesn't "just work" here; it definitely does work in other places, such as [class.dtor]p14, but IIRC turning down the line break penalty any further globally causes TeX to wrap inline references even in places where it's unnecessary to do so.

From https://tex.stackexchange.com/questions/94217/what-is-the-difference-between-nobreak-and-nolinebreak it looks like we're probably effectively using \penalty 301 currently (that's for article, but I'd guess memoir is probably not very different). Maybe we could binary search to find whatever value works in this case and try changing \iref to use something like that.

The wording that remains here is wrong, isn't it? The reason you can't provide an explicit template argument list to a conversion function is simply because grammar productions for template-id do not include a production for a conversion-function-id followed by a template-argument-list. (The reason for that is presumably the visual ambiguity between template arguments of the conversion function and template arguments of its type, combined with there being no way to write operator T </*T=*/int>, since you can't name T.) But it seems simply wrong to say that conversion member function templates are called without using a function name; they can be.

I don't like these "because" sentences in any case.

Is the change "are called" -> "are named" here intentional? I think it's clear that (for example) the expression A() calls a constructor without using a function name, and more generally, there is no syntax to call a constructor using a function name. However, it's not clear to me that template<typename T> A::A(T t) names the constructor without using a function name -- in that context, A::A names the constructor, and the constructor is a function, so arguably A::A is a function name.

[temp.arg.explicit] p2 lists the circumstances under which an explicit template argument list can be supplied for a function template. "A function is called" is just one of five options. I agree that "named" is bad term here, though, because we use "naming the constructor" when we want to weasel-word around the fact that constructors don't have names to start with. p1 talks about "function template name", and the note wants to highlight that a strict reading of "name" is required here (and thus constructors can't have template argument lists). Oh, and [class.qual] p2 has a lot of words to avoid mentioning the phrase "function name" for constructors, in particular for your declaration example.

I'll add a cross-reference.

"conversion member function template" is an awkward phrase, as it splits the term "conversion function". The "member" here seems redundant, since every conversion function [template] is already a member function [template]. So how about replacing that with just "conversion function template"?

It was there before...

Instead of "constructor member function templates", I think "constructor templates" would be clearer and easier to understand.

Again, pre-existing junk.