Extract the OS event filter implementation

[brainplot]

Description

The XcbEventFilter and WinEventFilter classes have been extracted into a separate file and a single class.
The new file has then been added to the build chain.

The new class, called OSEventFilter, implements the inherited nativeEventFilter() method and a preprocessor condition picks the right test for the if statement.

Installing a native event filter has now become this:

#if defined(Q_OS_WIN) || (defined(Q_OS_UNIX) && !defined(Q_OS_MACOS))
    installNativeEventFilter(new OSEventFilter());
#endif

which is oblivious of the actual implementation. The new structure should make it easier to add new event filters for new platforms: you just add a new preprocessor case in the nativeEventFilter() implementation.

Motivation and context

The XcbEventFilter and WinEventFilter classes had basically the same implementation, except for a single line: the condition in the if statement within the nativeEventFilter() method.
This made for a lot of code duplication. In addition to that, the release build generated the following warning:

'XcbEventFilter' has no out-of-line virtual method definitions; its vtable will be emitted in every translation unit.

In order to get rid of the warning, a base class's virtual method must have an out-of-line implementation in one of the derived classes. This will make the compiler place the vtable in that translation unit.
Moreover, Application.cpp felt a bit cluttered in my humble opinion: to reach the constructor implementation I had to scroll quite a few lines down.

How has this been tested?

I performed a dev build and a release build; and ran all the tests.
Tests n.30, n.31 and n.32 failed but they failed before I made any change as well.

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]

Now these are the kind of PRs I wanna see! Great work. Can we follow this architecture for the global hotkey detection as well?

[brainplot]

@droidmonkey thank you!

I think I've not yet read through the code that handles the global hotkey detection; but I can look into that and see if I can spot something that can be improved!

[droidmonkey]

Please rebase on develop, not sure what happened to the CI.

[brainplot]

@droidmonkey sure! I'm gonna do that.

I have a little question though. Does installNativeEventFilter() take care of deleting the memory when we do this?

installNativeEventFilter(new OSEventFilter());

I looked up on the documentation and I haven't found anything related to this. Maybe I'm missing something but if the method doesn't, I think we're introducing a leak. I coded it that way because it's how it was previously coded. We could use a QScopedPointer in the Application class.

[droidmonkey]

The docs don't say the application owns the filter object after calling the function so I agree with a scoped pointer.

[brainplot]

@droidmonkey I added the scoped pointer. However, I think there's something weird going on.
I defined a test destructor like this in the OSEventFilter class:

OSEventFilter::~OSEventFilter()
{
    qWarning("Destructor was invoked.\n");
}

and the message Destructor was invoked was not being printed upon exit.

I spent some time trying to figure out why but I didn't succeed 🤔

[droidmonkey]

I wouldn't worry about it since it lives the whole length of the application. Does ASAN report issues?

[brainplot]

This is the application output:

=================================================================
==21784==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 1536 byte(s) in 3 object(s) allocated from:
    #0 0x7fa8f8b87491 in __interceptor_realloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cc:105
    #1 0x7fa8f02da3f1  (/usr/lib/libfontconfig.so.1+0x213f1)

Direct leak of 136 byte(s) in 3 object(s) allocated from:
    #0 0x7fa8f8b88f19 in operator new[](unsigned long) /build/gcc/src/gcc/libsanitizer/asan/asan_new_delete.cc:93
    #1 0x7fa8f6a2b398 in qstrdup(char const*) (/usr/lib/libQt5Core.so.5+0xc6398)

Indirect leak of 2336 byte(s) in 73 object(s) allocated from:
    #0 0x7fa8f8b87231 in __interceptor_calloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cc:95
    #1 0x7fa8f02d9ee5  (/usr/lib/libfontconfig.so.1+0x20ee5)

Indirect leak of 518 byte(s) in 33 object(s) allocated from:
    #0 0x7fa8f8acff01 in __interceptor_strdup /build/gcc/src/gcc/libsanitizer/asan/asan_interceptors.cc:405
    #1 0x7fa8f02d9c25 in FcValueSave (/usr/lib/libfontconfig.so.1+0x20c25)

Indirect leak of 144 byte(s) in 3 object(s) allocated from:
    #0 0x7fa8f8b87019 in __interceptor_malloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cc:86
    #1 0x7fa8f02d3eae in FcLangSetCreate (/usr/lib/libfontconfig.so.1+0x1aeae)

SUMMARY: AddressSanitizer: 4670 byte(s) leaked in 115 allocation(s).
15:47:53: /home/gianluca/Projects/build-keepassxc-Desktop-Debug/src/keepassxc exited with code 1

which is what I've always got since I first started checking out the code.

[phoerious]

None of the Qt plugin destructors is called reliably.