[Development] What's the status of a moved-from object?

[Mutz, Marc]

On 2019-05-21 13:03, Konstantin Ritt wrote:
> вт, 21 мая 2019 г., 11:32 Mutz, Marc via Development
> <development at qt-project.org>:
> 
>> And while the partially-formed
>> state can be extended to non-pimpled classes easily:
>> 
>> class QRect {
>> int x, y, w, h;
>> public:
>> QRect() = default;
>> };
>> QRect r; // partially-formed
>> r.x();   // compilers _already_ warn about this
>> QRect r = {}; // zero-initialized
>> 
>> That
>> should be modelled by optional<QRect>, not by QRect itself.
> 
> Whilst the statement feels reasonable, this will require tons of API
> changes and double checks on the user side:

There are no double-checks. If the old code didn't check the rect before 
using it, it was a bug.

> optional<QRect> Item::childrenRect() const
> {
>     if (hasChildren()) {
>         QRect r = {};
>         for (auto *child : children())
>             r.unite(child->boundingRect().value_or({});

   for (auto *child : children())
       if (auto rect = child->boundingRect())
           r.unite(*rect);

>         return r;
>     }
>     return nullopt;
> }
> 
> QRect r = item->boundingRect().value_or({});
> if (!r.isEmpty())
>     ~~~

if (auto r = item->boundingRect())
     use(*r);

> Note that I'm ok with that, but should we enforce such a huge efforts
> all over Qt API just for making the default-constructible QRect a
> no-op?

This is not proposed for Qt 6. Not by me, at least. But it's the logical 
extension and once Qt can depend on C++17, iow: std::optional, it's the 
correct way forward, IMNSHO.

Thanks,
Marc