enable HiDPI support with Qt5

[Be-ing]

This will automatically use the system scaling settings as documented at http://doc.qt.io/qt-5/highdpi.html

[Be-ing]

Could someone test this on Windows and macOS? @JosepMaJAZ, @esbrandt?

[JosepMaJAZ]

AppVeyor build is still with QT4.

I have QT 5.7 (as downloaded from QTs site when it was the current release) with which I made some mixxx build tests quite some time ago and I could try how it works out now.
If I don't get too many problems to build it, I should be able to report about it today.

[rryan]

This attribute must be set before Q(Gui)Application is constructed.
https://doc.qt.io/qt-5/qt.html
Looks like this should go in main.cpp before creating MixxxApplication.

[Be-ing]

Good catch, thanks! Fixed (and force pushed to replace old commit).

[rryan]

LGTM

[rryan]

macOS CI looks broken, but unrelated:

[CXX] src/library/autodj/autodjprocessor.cpp
In file included from src/library/analysisfeature.cpp:7:
In file included from src/library/library.h:18:
In file included from src/library/analysisfeature.h:16:
src/library/dlganalysis.h:64:18: error: field has incomplete type 'QButtonGroup'
    QButtonGroup m_songsButtonGroup;
                 ^
/usr/local/Cellar/qt/5.11.0/Frameworks/QtWidgets.framework/Headers/qabstractbutton.h:53:7: note: forward declaration of 'QButtonGroup'

[daschuer]

Unfortunately this is a regression for Ubuntu Trusty and Xenial, which have a Qt version < Qt 5.6

[rryan]

Unfortunately this is a regression for Ubuntu Trusty and Xenial, which have a Qt version < Qt 5.6

This change only affects >=5.6 so how can it affect those?

[daschuer]

Right, this PR itself is not a regression. Due to moving to QT5 we have lost the HIDPI hacks of QT4, this PR beings HIDPI back, but only for QT 5.6 builds and up. So there is a gap that have still no HIDPI.

This can be fixed by either backport the QT 5.6 changes to detect the OS scaling or show the scaling slider in preferences and let the user choose the QT 5 scaling (not the QT4 Hack of cause)

[rryan]

This can be fixed by either backport the QT 5.6 changes to detect the OS scaling or show the scaling slider in preferences and let the user choose the QT 5 scaling (not the QT4 Hack of cause)

Or do nothing :).

Since there is an LTS available that fixes the problem and users can always stick with MIxxx 2.1 it seems pretty low severity.

[daschuer]

Yes of cause. But It will be surprising for Xenial users, to loose HIDPI after update to 2.2.

[uklotzde]

The auto-scaling does not work well on my T460p on the built-in display with a resolution of 2560x1440. The Mixxx UI should select a scaling factor of 1x instead of 2x to be usable.

We need an option to customize the scaling in the preferences. Until then I have to remove this line from my custom Qt5 build.

[Be-ing]

You can use Qt's standard environment variables if the autodetection doesn't work well for your screen. I have not seen any other Qt apps provide a GUI for adjusting this. I do not think Mixxx should be an exception. Qt's autodetection works great on my 3840 x 2160 14 inch diagonal screen.

[daschuer]

We already have a GUI to reanable the individual scaling. Let's bring it back.
If Uwe want's it, there are probably many users out there with the same demand but without the knowledge to set environment variables.
There is also the regression for legacy qt5 user without auto adjusting.

[uklotzde]

Confirmed: The auto scaling can be overruled by setting QT_AUTO_SCREEN_SCALE_FACTOR=0.

But this workaround is cumbersome for the casual user. We should at least offer an option to disable auto scaling in the DisplayInterface preferences.

Those intermediate resolutions between FullHD and 4K do not work well with the default 2x upscaling. Same issue with GNOME where I use a Scaling Factor of 1,0 even on a 14" screen.

[Be-ing]

I do not think it is an application's place to offer a UI for adjusting this system setting.

[uklotzde]

I disagree. We are now unconditionally setting Qt::AA_EnableHighDpiScaling which works for the majority of uses cases, but fails for some as outlined. Conversely we could also remove this invocation entirely and tell the user to simply set QT_AUTO_SCREEN_SCALE_FACTOR=1 if they really need auto scaling? No, not really ;)

Programmatically enabling auto scaling by default is ok, as long as we offer an option that allows to disable it (i.e. not enable it explicitly). Otherwise I expect support requests by users that are faced with an unusable UI. We cannot expect them to read the Qt developer documentation.

[Be-ing]

I presume if a user is having a problem with Mixxx's scaling, every Qt application is affected too. So why should any one application be responsible for working around it?

[uklotzde]

Not every Qt application. If we explicitly invoke QApplication::setAttribute(Qt::AA_EnableHighDpiScaling) because we assume that this is a sensible default we should also offer the user to opt-out easily if needed and skip setting this attribute.