[ranges] Remove unnecessary copy-list-initialization LWG3796 #5911
Conversation
| @@ -3411,7 +3411,7 @@ | |||
| // \ref{range.repeat.iterator}, class \tcode{repeat_view::\exposid{iterator}} | |||
| struct @\exposidnc{iterator}@; // \expos | |||
|
|
|||
| @\exposidnc{movable-box}@<W> @\exposid{value_}@ = W(); // \expos, see \ref{range.move.wrap} | |||
| @\exposidnc{movable-box}@<W> @\exposid{value_}@; // \expos, see \ref{range.move.wrap} | |||
There was a problem hiding this comment.
This isn't copy-list-initialization, there's no {} here - it's plain "copy-initialization". The proposed change isn't editorial, it's normative since it causes the default constructor to default-initialize . These comments apply to line 13304 as well.value_ instead of value-initializing it
There was a problem hiding this comment.
Thanks for the correction. I'm going to submit an LWG since it actually caused some issues.
There was a problem hiding this comment.
it causes the default constructor to default-initialize
value_instead of value-initializing it
I believe default-initialization and value-initialization are equivalent for movable-box. Its default ctor, whenever provided, always value-initializes the contained value.
However, the propose change seems non-editorial to me. The current wording does require move construction in these default ctors (as specified via optional's corresponding ctor), which is probably wrong, and perhaps an LWG issue is needed to correct them.
There was a problem hiding this comment.
@CaseyCarter , "it causes the default constructor to default-initialize value_ instead of value-initializing it."
This is not quite accurate. We're initializing a movable-box<W> with a value-initialized value of type W, which may or may not be equivalent to value-initializing the movable-box<W> itself (this issue claims it isn't equivalent, and actually harmful).
|
Please leave a comment with the LWG issue number once you have it. |
|
jensmaurer
changed the title
|
Can we close this editorial issue now that this is tracked elsewhere? |
The two should be equivalent. This is consistent with the rest of
<ranges>.