[dcl.init.general] p20 contradicts the wording of parenthesized aggregate-initialization

[poyaoc97]

[dcl.init.general]p20 says:

If the initializer is a parenthesized expression-list, the expressions are evaluated in the order specified for function calls ([expr.call]).

However, that's not intended for parenthesized aggregate-initialization introduced by P0960R3, of which the evaluation order went back-and-forth but finally settled on in-order, left-to-right evaluation.

-If the initializer is a parenthesized \grammarterm{expression-list},
+Unless otherwise specified,
 the expressions are evaluated in the order
-specified for function calls\iref{expr.call}.
+specified for function calls\iref{expr.call}
+if the initializer is a parenthesized \grammarterm{expression-list}.

Would this be a suitable change?

[eisenwave]

P1. There is a normative wording contradiction for T x(a, b) where T is aggregate between:

• [dcl.init.general] p20, which states that the evaluation order is that of function calls, and
• [dcl.init.general] p16.6.2.2, which adds a "sequenced before" left-to-right ordering.

There is no "otherwise" in those paragraphs that would prevent a contradiction.

I think that the design and wording intent is clear, and a simple "unless otherwise specified" would fix it, similar to what you've suggested.

[jensmaurer]

We also have the array case in p16.5.

I think I would prefer to say in p20 "If the destination type is not an aggregate and the initializer is a parenthesized expression-list, ..."

"Unless otherwise specified" always makes me go hunting, which is tedious.

@eisenwave , can you get the pull request ready for this afternoon's CWG session?

[eisenwave]

@jensmaurer I would do it right now, but the repo is private, my fork is detached from it, and I don't have write permissions, so I cannot make branches directly in here.

I could probably do it immediately after plenary if the repo is made public again then.

[tkoeppe]

(I'm also holding back on responding on a few issues, because I would like to tag the reporters, too, which I believe isn't currently possible.)