Replace deprecated parts of code

[brainplot]

Performing a dev build against the latest version of Qt fails because of some deprecated members.

Description

As of Qt 5.10, two functions and one class that were used in the code are deprecated.
Since this project supports Qt version 5.2 and higher, I replaced the two functions marked as deprecated with their updated versions and wrapped the new code in a compile-time check which enables it only if the project is being compiled against Qt 5.10 and up.

In the CSV import utility, the QSignalMapper class was used, which is entirely deprecated now. In place of it, lambdas can be used, which are part of the C++11 standard so I don't think that compile-time checks are required here since it's the version that the project is using, looking at the CMake scripts.
I removed the usage of the aforementioned class and used a lambda to achieve the same effect.

Motivation and context

I'm running Archlinux, which uses the latest version of Qt (5.11.2), and I wasn't able to compile the project with the -DWITH_DEV_BUILD flag.

How has this been tested?

Built with -DWITH_ASAN, -DWITH_TESTS and -DWITH_GUI_TESTS; test n.30, n.31 and n.32 failed but they failed before I made any change as well.
I manually tested the parts of code I changed: I auto-typed my password on a few sites; tried to re-import my database previously exported in .csv format and changed some columns around with the comboboxes. Everything worked as expected.

Types of changes

• ✅ Bug fix (non-breaking change which fixes an issue)

Checklist:

• ✅ I have read the CONTRIBUTING document. [REQUIRED]
• ✅ My code follows the code style of this project. [REQUIRED]
• ✅ I have compiled and verified my code with -DWITH_ASAN=ON. [REQUIRED]

[droidmonkey]

It's not wrong, just difficult.

My only issue with this is that our CI will only be testing the original code on Linux (ie, the deprecated code) since we test with Qt 5.2. Windows CI will test the new code.

[droidmonkey]

This is very ugly, almost incomprehensible. I feel like this interaction is just not coded well to begin with due to the use of the QSignalMapper in the first place.

[brainplot]

To be honest I don't love it either, but there's no other way I know of I can tell the compiler to pick

void currentIndexChanged(int)

instead of

void currentIndexChanged(const QString&)

The QOverload<int>::of() syntax is a lot more readable in my opinion but it's only available since Qt 5.7.

Currently, there's no overload of the connect() method that can take a lambda if the SIGNAL() syntax is used. That's why I had to resort to that ugly syntax.

EDIT: if we break up that statement in this way, maybe it's a bit more readable.

#if QT_VERSION >= QT_VERSION_CHECK(5, 7, 0)
    constexpr auto onComboChange =
        QOverload<int>::of(&QComboBox::currentIndexChanged);
#else
    constexpr auto onComboChange =
        static_cast<void(QComboBox::*)(int)>(&QComboBox::currentIndexChanged);
#endif
    connect(combo, onComboChange, [=]{ comboChanged(combo, i); });

It's still ugly, but I think that QOverload<int>::of gives a strong hint about what the purpose of the code is.

[phoerious]

Actually, the AppImage build uses a more recent Qt version (5.10 I believe).

[brainplot]

@phoerious the code is still in my first version though. I haven't pushed my latest revision (i.e. the one where I extract the code that selects the right overload). Is it okay all the same?

[brainplot]

I understand the problem. However, I don't think to have changed the code in a very significant way (apart from the QSignalMapper thing, maybe).
I've only replaced a few method calls with new ones that, by contract, should do the same thing; so theoretically if the old code worked, the new code should work just as well.

But if you feel that this is too risky, feel free to reject this PR! :)

[droidmonkey]

I don't have major heartburn with the changes, see what @phoerious thinks.

[brainplot]

I replaced Q_OS_OSX and Q_OS_MAC with Q_OS_MACOS, as they're now both deprecated according to the Qt documentation. It also made the code base more consistent: only one macro is now used to identify an Apple machine instead of two, which could be a bit confusing.
I recommend switching to my first commit to see the actual reason I opened this PR :)

[droidmonkey]

Good changes!

[phoerious]

Looks good, ship it!