• splitRef here to not allocate strings that are not used anyway (since they are split again later on)
• you are using a C++11 for loop on a Qt container, and this will detach it -- you need to store the container as variable:

const auto parts = data.splitRef(...);
for (... : parts)

• QLatin1Char is better than QStringLiteral here (since you have a single character as separator)

ditto, QStringLiteral -> QLatin1Char

no space between the enum and the colon (applies to the other lines too)

this is not needed, the destructor of QTemporaryDir will do it by default

move this variable directly in the code block where it is used

coding style: they go in the same line

better have this as very last operation done, so early returns will not have this notification in m_notifications

most probably check whether starting the process succeeded, and if not call finish(notification) and return earlier

IMHO it is better to check whether the process exited correctly, and its return status was a success, finish'ing the notification in case of failure

considering here the notification is being closed and thus checking what the tool does has almost no effect on the flow, then there is no need to manually create a QProcess object:

• collect the arguments (like you already do)
• use the static QProcess::startDetached(program, args)

splitRef

is interesting, everyone learns in a code review.
Wouldn't a qAsConst also prevent detaching?

I've added more inline docs for now. @pino if you could guide me on how to update the docs on the website, that'd be great! 😃

what do you need XmlPatterns for?

this is indented too much

if you could guide me on how to update the docs on the website, that'd be great!

which website?

instead of getting the real QString of str and then get a substring of it, first get the substring(ref) and then get the real QString of it

QStringRef has toInt()

if the notification is not found, this will be an uninitialized pointer; TBH if the search for the notification with the specified id fails, then it should be better to return earlier, as it means the notification is unknown

please document what is the issue, so in the future the workaround can be dropped

they are indented too much

the logic here is swapped: if waitForStarted() returns false, that means the process did not start successfully; also, after finish() you must return earlier (do not forget to delete the process), otherwise the rest of the code does things as if the process run fine

the removal ought to be done no matter the exit status of the process: the process exited, so image for it is no more needed

no need to use it here, you already have the notification parameter...

deleteLater should also close the socket.
Also don't use a pointer after you deleted it, even with a deleteLater.

I can't put my finger on it I can only tell that https://github.com/KDE/snoretoast/blob/a53a6b733624ada78ef0f852407cad4be1c00b16/examples/qt/main.cpp#L58 produces a corrupted string.
Runs fine with msvc2017.
I tried to create a minimal example https://invent.kde.org/snippets/253 but this one works fine..

I doubt this compiles...You can't initialise variables inside the class

Have you heard of this new c++ standard? C++11?

https://api.kde.org/frameworks/knotifications/html/index.html

well, I just found out the patch had broken functionality that I fixed just after putting here an else return! 😆

I'm removing this code chunk because we already check for successful show of notif in the following connect.

I have fixed the issue now.

find_package calls are usually in the toplevel CMakeLists.txt

where is the win32_defaults.notifyrc file?

Can someone please confirm if this file is correctly used?

pls fix the duplication