[expr.unary.noexcept] Replace informative wording

[Eisenwave]

Context and Problem with the Current Wording

I was briefly under the impression that a + b can throw for two ints a and b, because signed integer overflow is UB, and UB can result in throwing exceptions, and maybe the noexcept operator needs to consider that.

Paragraph 1 in its current form says:

determines whether the evaluation of its operand [...] can throw an exception

An unsuspecting reader like me might assume that this includes exceptions thrown as the result of UB. However, this isn't actually possible, and this wording is purely informative. Paragraph 3 contains the actual normative wording, explaining that the noexcept operator checks whether an expression is potentially-throwing.

Suggested Solution

This PR moves the actual normative wording from p3 up into p1 to avoid any future confusion. The informative wording is removed.

[jensmaurer]

I think there is some room for improvement in this subclause, but talking about "the operand" in detail before we've even introduced the grammar seems unhelpful.

My suggestion:

• grammar first
• then talk about the operand
• then talk about the result
• then have the note about constant expression

[Eisenwave]

I've implemented most of what you've suggested, but I don't understand what you mean by "note about constant expression". Neither the old wording nor the new wording talks about constant expressions.

Wouldn't adding a note be a topic for another PR?

[jensmaurer]

"[Note 1 : A noexcept-expression is an integral constant expression (7.7). — end note]"

[Eisenwave]

Ah yes, I wasn't looking at the second paragraph. I've moved it up now.

[Eisenwave]

updated version

[jensmaurer]

Sorry, but that's not exactly what I had in mind. Since the change is a little larger, I've created CWG2792 to have CWG review the change. See there for my suggestion; feel free to continue to discuss here.