Make sure --enable-online-update does not leave behind UpdateCheckThread

...which is known to cause test processes to crash during exit.  (Also see
5cd04405c6b2d1ee46294ce0696d89d2edc97d16 "Disable automatic onFirstVisibleTask
UpdateCheck during CppunitTests", so this commit now covers tests like
JunitTests and UITests that start a full soffice process.)

The relevant code is already rather complex and incomprehensible; it handles the
automatic update check triggered by the onFirstVisibleTask UpdateCheckJob,
manually triggered update checks, and downloading of new versions.  To avoid
regressions in all those various scenarios, I tried to add very targeted code
that should only wait for a still running UpdateCheckThread triggered by the
automatic onFirstVisibleTask UpdateCheckJob---which is the scenario that caused
issues during `make check`.  (For a manually triggered update check or download
it is probably both less likely that related threads would still run during
exit, as the user would likely wait for a response before closing LibreOffice,
and less severe if threads would still run during exit, as any resulting crashes
would likely be "harmless" and might not even be noticed by the user.)  However,
I note that this targeted fix does increase the complexity and inscrutability of
this code even further...

The new code does not actually try to join the UpdateCheckJob---see
UpdateCheckThread::join referencing
<https://bz.apache.org/ooo/show_bug.cgi?id=73893> "webdav ucp does not implement
XCommandProcessor::abort".  Rather, it waits for UpdateCheckJob::run to finish
(after unblocking at least any m_aCondition.wait calls).  This, however, can
take a while if the update check's HTTP request is in progress.

To be able to use a std::condition_variable (with sane semantics compared to
osl::ConditionVariable) in UpdateCheck::waitForUpdateCheckFinished, the
corresponding UpdateCheck::m_aMutex had to be changed from a (recursive)
osl::Mutex to a std::recursive_mutex one (sticking to a recursive one just to be
on the safe side, whether or not the code actually makes use of that).

Change-Id: I2726acd33b884b807d840cfee2786fcf45815ad7
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/103013
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>

0 files changed, 0 insertions, 0 deletions