GitLab

Bastien Nocera authored Aug 21, 2020 and Carlos Garnacho committed Aug 23, 2020

Relegate the tracker2 plugin build and testing to a separate target
which we'll remove in the future.

Otherwise this is racy, and prone to failures in CI or other slow
circumstances.

Query further data from GrlMedia in events, and filter out those GrlMedia
that don't resolve properly (eg. other resources we get notifications about,
but don't are a nie:InformationElement with a nie:isStoredAs relation).

This avoids emission of GrlMedia that are actually irrelevant to grilo.

Fixes: #73

The ID is unique enough to trigger media queries, and we are already
prepared for that, so just allow that here.

We are essentially interested in data from audio/video/image graphs.
Everything is foreign to this plugin.

It underwent some substantial changes, despite being based on the
2.x code.

Adapt to the changes in tracker-miners about this metadata.

This may be useful if using domain ontologies for the tracker-miner
services (e.g. a flatpak app without tracker3 in the host), in this
case the miner service would be based on the app DBus name, thus
queries and notification subscriptions need to change.

Even though both can be compiled and installed separately, the usual
approach to grl_registry_load_all_plugins() will cause GType clashes
as both plugins will be attempted to be initialized.

Even if we could make that work, the benefits are theoretical since
other dependencies (eg. gtk) have explicit dependencies on one or
another Tracker version, so it's more a distributor-level choice
really...

Use the tracker 3.0 testutils helpers to run a tracker sandbox, and put
some sample files on it (borrowed from chromaprint and tracker-miners
tests) so we can test features in the grilo plugin.

This should be added explicitly, just in case the connection is local
(and thus does not listen to events from tracker-miner-fs bus name by
default).

GNOME Music strongly relies on this.

Use G_*_PRIVATE when declaring the type.

Do things straight from the query, so we get a proper media type
for the given row.

This is no longer used.

A number of fields in the struct is now unused, can be removed.

The utils code to create sparql queries is no unused, this can
be removed.

The data stored by miner-fs is no longer considered readwrite for
all. These updates used to trigger the tracker writeback service
which would rewrite file metadata and make everything match again.
Make this use the writeback service directly, the miner-fs data
will then be updated indirectly.

We now look up variable names here, so update this. Also drop the
other (no longer necessary) mappings.

Set it afterwards, instead of relying on a second for the same
metadata key. This is the only place where this is needed and
can be easily handled in other ways.

tracker:id() is an internal detail, it sounds better not to expose
it in public API.

Shuffle things so we don't miss errors there.

In tracker3 these 2 classes are no longer attributed to the very same
resource object. Those are 2 different entities, so this must be
accounted for. Prefer to stay wherever possible in the InformationElement
domain, and resolve properly the DataObject from the InformationElement
otherwise.

Those are not used anymore.

We drop the support for full queries, the query is extended so that
subselects work.

Carlos Garnacho authored May 22, 2020 and Jean Felder committed Aug 21, 2020

This is essentially a GRL_TRACKER_QUERY_MEDIA_FROM_URI with no
additional keys or filters.

Carlos Garnacho authored May 23, 2020 and Jean Felder committed Aug 21, 2020

We match by GrlTypeFilter, so it's easy to discard those that don't apply.

Carlos Garnacho authored May 23, 2020 and Jean Felder committed Aug 21, 2020

So it can be used to generate variable names in SPARQL queries, and
parameter bindings for filters.

Carlos Garnacho authored Aug 20, 2020 and Jean Felder committed Aug 21, 2020

If a leak happened, this shouldn't be warned on random app's stderr.

Carlos Garnacho authored May 22, 2020 and Jean Felder committed Aug 21, 2020

This only unsets the private connection, which is rather wrong.
Let it live so it can be used during the object lifetime.

Carlos Garnacho authored May 22, 2020 and Jean Felder committed Aug 21, 2020

For a large part, this work duplicates tracker's internal handling
(eg. updates are already guaranteed to be executed sequentially), or
even hinders it (eg. select queries are serialized even though they
can run in parallel).

This will also get a bit in the middle, as it makes sense to
minimize the time between a query happening and a
TrackerSparqlStatement being executed for that TrackerSparqlStatement
to be ready for reuse ASAP.

Drop this queue and let Tracker handle things itself.

Carlos Garnacho authored May 22, 2020 and Jean Felder committed Aug 21, 2020

The operation is passed as data to the cursor asynchronous method,
so it's already implicitly attached to it.

Carlos Garnacho authored May 22, 2020 and Jean Felder committed Aug 21, 2020

The cursor will already error out if the cancellable was cancelled.
Check for that error and avoid checking the cancellable directly.