If we redefine \defnx as follows:

\newcommand{\defnx}[2]{%
  \indexdefn{#2}%
  \ifthenelse{\NOT\equal{#1}{#2}}{%
    \indextext{#1|see{#2}}%
  }{}%
  \textit{#1}%
}

then \defnx{trivially copyable class}{class!trivially copyable} and \defn{blah} would both do what you want.

Issue #1033 is related; we did decide that the page numbers go with the decomposed index entry. (That doesn't mean we should not have "see" entries for the other form.)
It is not obvious that the "see" form is always non-decomposed. For example "class, trivially copyable" has the page numbers and "trivially copyable, class" has the "see". There might also be a "trivially copyable, type" with another "see" to "type, trivially copyable" that points to [basic.types] or so.
@jwakely, I believe such an approach would still make the index useful to find "trivially copyable class"; do you agree?

@godbyk, I'm avoiding the \ifthenelse for now. I have
\newcommand{\defnc}[2]{\indextext{#1|see{#2}}\indexdefn{#2}\textit{#1}}
used as \defnc{trivially copyable class}{class!trivially copyable}.
This does not produce an index entry for "trivially copyable class", most likely because of the "|" in the 'see' part of the entry. (The "see" seems to need a comma-separated thing.)
Any obvious ideas I'm missing, otherwise I'll redesign the macro so that we must use it as \defnc{trivially copyable}{class}?

@jwakely, I believe such an approach would still make the index useful to find "trivially copyable class"; do you agree?

Yes. As long as I can go to T in the index and find "trivially copyable class" in some form or another I think that's fine. Currently it's not there at all under T.

I tried using additional packages so I could replace the "|" with ", " in the index entry, but didn't get it working. Breaking it into two arguments seems like a better solution.

In #2080, there is a specific change to create a \defnc for compound defined terms with uniform indexing. We should form an opinion whether that's the way to go, and then apply it everywhere.

@jensmaurer Why are we avoiding \ifthenelse?

The definition of \defnc in #2080 seems rather narrow in that it only works for x y → y, x transforms. Wouldn't a more general function (like what my \defnx macro above provides) be better? Using \defnc also means that authors and editors have to learn yet another macro.

@godbyk: The point is not the \ifthenelse. I'm not seeing that your \defnx macro above converts an exclamation mark to a comma, as required to form a proper "see".

@jensmaurer Ah, sorry. My brain was still in xindy mode when I wrote it. (xindy requires you use x!y instead of x, y for cross-references.)

Try replacing the existing \defnx macro with the following code:

\usepackage{xstring}
\newcommand{\defnx}[2]{%
  \indexdefn{#2}%
  \ifthenelse{\NOT\equal{#1}{#2}}{%
    \StrSubstitute{#2}{!}{, }[\stdtemp]%
    \indextext{#1|see{\stdtemp}}%
  }{}%
  \textit{#1}%
}

It replaces all instances of ! with , in the index entry so the cross-reference should work again. (If we switch to xindy in the future, we'll want to skip the substitution, though.)

@godbyk: Ah. So, we get the correct "see" when using exclamation marks once xindy lands (see #1917). Thanks.
I'd like to point out, though, that the \defnc macro reduces redundancy in the source code, because every word appears only once. (A \defnx repeats the words in a different order.) Although it's awkward that the position of the exclamation mark is indicated by }{ (i.e. starting the second argument).

@jensmaurer Yeah. xindy actually checks to see if the cross-reference target exists and complains if it doesn't. It's found a few errors in our current indexing.

It's true that it reduces the number of words you have to type, but only for that particular construction. If you want to do any other type of cross-reference, you're stuck using \defnx anyway. I think using \defnx in all cases will reduce the mental load on authors: it provides consistency and clarity.

@godbyk Right, but I suspect we'll have a lot of use for \defnc once we actually get going with #1033. Let's have the editorial team discuss this in Rapperswil.

Editorial meeting consensus: \defnc as presented in #2080 is approved and even liked by the Project Editor. Yes, it covers only two-part terms, but that's likely > 95% of the cases, so good enough.

"trivially copyable class" was fixed. There's no point in leaving this open. Please create a new issue for specific bad index entries.