GitLab

This reverts commit 2a334d21.

We used to need a lock to be held before a discovery was scheduled and
at least until after the parsing was complete.
This is not the case anymore since the discoverer thread only emits a
discovery completed event upon full completion, rather than each time a
specific entry point gets discovered so we don't need to update various
states from multiple callback in order to wait for all folders to be
discovered and parsing to be completed after that.
This removes a potential lock inversion and simplifies the tests a bit

Hopefully this will help with #362

Fix #362

To clarify in some member functions that the lock is stored in the
instance, and not released at the end of the scope

This makes the member safe to read later on without any mutex since we
never write again to that variable

Which allows the parser to get that information without the FsHolder
going through the media library instance, locking it to get the parser
instance, and invoking the callback manually
Fix #362

refs #362

This allows to break more of the dependency on the media library
instance when dealing with device appearing/disappearing

refs #362

Which allows us to remove some of the forwarding calls in MediaLibrary

This is part of a refactoring aiming at fixing #362
The fsHolder object is initialized during construction and doesn't
require locking from the MediaLibrary instance, and will manage its
internal locking as needed.
The end goal is to have the FsDiscoverer class using the FsHolder
directly without going through the media library

This is only called from the constructor. Clarify this and don't probe
the existing list since it will always be empty

The verbosity is always forwared as is to the logger, storing it has no
purpose since it's never read

We are starting to have too many things protected by the same lock.
Marking these 2 members as const means we can safely assume they do not
need to be protected

Not doing so means we are losing the user's input when stopping, or when
replacing the libvlc instance.

Don't force a parser recreation when replacing libvlc instance since the
instance isn't cached anymore. The next run will simply use the updated
instance
refs #361

This can greatly simplify replacing an external instance by removing the
need to manually stop all parsers before recreating them, for a rather
limited run-time cost

Stop the background tasks as the main locked operation, use the unlocked
helpers for restarting after the instance has been replaced
Should fix #361

VLCInstance has its own mutex to protect the instance, so we don't need
to hold the media library mutex while probing the instance class
refs #361

refs #361

Fix #323

Fix #359

Fix #300

This is not critical but if we chain multiple migrations we might end up
recreating the backup table while it still exists

Since some thumbnails might be in a broken state, let's regenerate
everything

This now monitors insertion to the cleanup table and batches them.
This allows us to filter what kind of thumbnails we want to cleanup (ie.
only the owned ones) and stop relying on potentially invalid assumptions
when computing the thumbnail path.
refs #350

refs #350

refs #350