[Development] qMoveToConst helper for rvalue references to movable Qt containers?

[Elvis Stansvik]

Den sön 21 okt. 2018 kl 17:50 skrev Giuseppe D'Angelo
<giuseppe.dangelo at kdab.com>:
>
> Hello,
>
> Il 21/10/18 16:15, Elvis Stansvik ha scritto:
> > I couldn't find a way to contact them.
>
> The best shot would be the std-discussion mailing list, I think.
>
> > In order to try out the unsafe usage you suggested in your other mail,
> > and also another unsafe usage pointed out in an SO question
> > (https://stackoverflow.com/questions/39051460/why-does-as-const-forbid-rvalue-arguments/39051612#39051612),
> > I made the following test program.
> >
> > The output when running is:
> >
> > [estan at newton move-to-const-test]$ ./move-to-const-test
> > without moveToConst:
> > FooPrivate constr from vector
> > Foo constr with arg 0x7fffdb627200
> > Foo begin 0x7fffdb627200
> > Foo end 0x7fffdb627200
> > Foo destr 0x7fffdb627200
> > FooPrivate destr
> >
> > with moveToConst:
> > FooPrivate constr from vector
> > Foo constr with arg 0x7fffdb627208
> > Foo move constr 0x7fffdb627210
> > Foo destr 0x7fffdb627208
> > Foo begin const 0x7fffdb627210
> > Foo end const 0x7fffdb627210
> > Foo destr 0x7fffdb627210
> > FooPrivate destr
> > [estan at newton move-to-const-test]$
> >
> > Which just shows it's working as intended.
>
> However the third test should be a without moveToConst, but storing in a
> temporary, i.e. the current best practice. It should output exactly like
> the first one, but of course call const begin/end.

For completely other reasons, I came across "Range-based for
statements with initializer" today:

    http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2017/p0614r1.html

With that, the Qt best practice could become

for (const auto list = getQList(); const auto &v : list) {
    ...
}

Which may or may not be pretty, but avoids the extra line of code and
keeps the scope clean.

Elvis

>
> And note the extra move/destruciton in the second example.
>
>
> > The two unsafe usages are commented out, because they wouldn't compile (good!).
>
>
> Whops, you're absolutely right. My bad. With your proposed implementation:
>
> > template <typename T>
> > const T moveToConst(T &&t)
> > {
> >     return std::move(t);
> > }
>
> This won't compile when called on a non-const lvalue (the return type
> will be a lvalue reference to const, which won't bind to a non-const
> rvalue). I stand corrected.
>
> Which actually makes me think of yet another possibility of misuse, that
> is, applying moveToConst to a function returning a const object. That
> will compile, using a copy...
>
> > const QVector<Obj> getVector();
> > for (auto &obj : moveToConst(getVector()) ~~~
>
>
> Long story short, I think it's good if we stay away from this in Qt.
> Clazy warns you if you "do it wrong", and being a Qt-specific problem,
> we better not offer too many ways that could have unpleasant drawbacks.
>
>
> My 2 c,
> --
> Giuseppe D'Angelo | giuseppe.dangelo at kdab.com | Senior Software Engineer
> KDAB (France) S.A.S., a KDAB Group company
> Tel. France +33 (0)4 90 84 08 53, http://www.kdab.com
> KDAB - The Qt, C++ and OpenGL Experts
>