autotests: Fix autotests to pass under gcc ubsan and leak sanitizer.
The GCC undefined behavior sanitizer (ubsan) flags our deliberate usage of undefined behavior within the KJob autotests, and the leak sanitizer flags some memory leaks from the use of KAutoSaveFile.
It would be good to be able to consistently run the autotests under the use of compiler sanitizers to catch bugs in kcoreaddons libraries, but this relies on the code still passing the test suite in this case.
It appears to me that the use of undefined behavior in KJobTest is as a
simple sanity check that KJob::isFinished is consistent with the fact
that we just called a slot connected to KJob::finished(). But
KJob::isFinished() is a protected method and in the context of a
KJob::finished() signal emitted from KJob::~KJob() there can be no
subclasses to interrogate KJob::isFinished() anyways.
I try to go about the sanity check instead by counting specific number
of finished calls from each individual job, ensuring that no job ever
emits finished more than once, that the corresponding slotResult
call (on success) comes after one (and only one) finished signal for
the job, and that the number of KJob test that emit finished but do
not emit result is exactly equal to the number of expected failed jobs
It might be simpler just to use qobject_cast or dynamic_cast and
only perform the sanity check when we know the finished call is
happening from a subclass.
For KAutoSaveFile, it's just a matter of remembering to call delete on the list of returned KAutoSaveFile* objects. This is worth fixing just because autotests are sometimes treated as documentation on proper usage, but I also make sure to clarify in the API documentation (the code examples in the API docs show correct usage but the per-method documentation does not make clear that the application owns the returned pointers).