This list may be NULL, so protect against that happening. Fixes
a number of runtime warnings on files that trigger the situation.

Related: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/233

This is sadly something that may happen with btrfs subvolumes, so
will be seen frequently in certain environments, or might cause
the test suite to fail for no (unforeseen) reason. Convert this
errno into a g_info() what will not be logged by default.

Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/212

tracker-extract: Protect against NULL GstTagList

See merge request GNOME/tracker-miners!412

libtracker-miner: Avoid warning on EXDEV from fanotify_mark()

Closes #212

See merge request GNOME/tracker-miners!413

Instead of doing a specific run to get testing coverage, collect the
information of all tests being done. Since some of these will check for
different build options, it's not possible to do that from a single run.

This way we will get accurate information for all code that is being
actually exercised by the tests.

In consequence, using eatmydata to run the tests is no longer ok,
since there is the possibility that it also eats the coverage reports
that are generated during test runs.

Also, since running tracker-extract in tests with seccomp enabled
prevents coverage information from being written for the metadata
extractors, this also means we now disable the seccomp jail on all
tests (as opposed to just the coverage target).

It is no longer used, so we can stop fetching it.

This stage runs the tests with seccomp enabled, thus still helping
us catch disallowed syscalls that might sneak in dependencies as
shipped by distros.

Traditionally, we've only rebuilt images on an as-needed basis, this
however isn't most useful if we want to make CI able to track seccomp
breaks in recent updates from distros.

In order to help there, add the CI handling so we can trigger rebuild
of testing containers on a scheduled job set up through the gitlab
UI. This will help us catching newly introduced syscalls more
proactively than now.

Instead of having to browse for it, expose the artifact in
the gitlab MR UI.

ci: Collect coverage information from all tests

See merge request GNOME/tracker-miners!411

"Increase coverage with this weird trick", quit the main loop when
SIGTERM is received, so that the writeback process can exit cleanly,
and its coverage information ends up written.

Should fix the 0% coverage seen in src/tracker-writeback altogether,
since we do have functional tests that exercise this code.

tracker-writeback: Exit cleanly on SIGTERM

See merge request GNOME/tracker-miners!414

They cause builds to fail on Debian with

"Disallowed syscall "get_mempolicy" caught in sandbox"

Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/234

Allow {get,set}_mempolicy syscalls

Closes #234

See merge request GNOME/tracker-miners!415

gst_discoverer_stream_info_get_tags() now returns a
`const GstTagList *`.

No functional changes.

tracker-extract-gstreamer.c: Fix 'const' warning

See merge request GNOME/tracker-miners!416

We already did let extractor modules provide a shutdown function, but
this was never called. Add the missing code so that these shutdown
functions are called during finalization.

Given the limitations imposed by seccomp, extractor modules cannot
(and should not) use the DBus connection or a readwrite connection
where direct updates are possible.

However these may want to inspect data existing in the filesystem
miner database, provide this API so that extractor modules can obtain
a readonly connection.

As the seccomp jail works, extractor modules may have a hard time poking
at DBus things, so we cannot obtain a dbus connection here. We can however
open files readonly, so open the database this way.

This will allow to use the miner database to resolve miscellaneous queries
about files surrounding the file being extracted (e.g. cue sheets).

We used to iterate all files in the same folder than the audio file to
look up CUE sheet files, but this may be overly expensive on very
flat folder hierarchies where the audio file shares folder with other
thousands of elements.

Since we have this information indexed already, we can use a readonly
connection to query for these CUE sheet files. It is very likely that
these files are just indexed as nfo:FileDataObject, so we have to look
up the file extension to determine cue files, since mimetype detection
relies on the file extension too, the check is equivalent.

And anyways, keep a fast-path lookup for CUE files that have the same
filename but just different extension. Partly to keep tests happy,
partly to also avoid possibly iterating over thousands of CUE files.

Closes: https://gitlab.gnome.org/GNOME/tracker/-/issues/378

It seems The gst_discoverer_info_get_stream_info() API call can
return us a GstDiscovererContainerInfo or a plain
GstDiscovererStreamInfo. Find out which is it, and use the right
API to get its tags.

Fixes the following warning seen with some files extracted by
GStreamer:

tracker-extract-3:25304): GStreamer-PBUtils-CRITICAL **: 13:44:06.792: gst_discoverer_container_info_get_tags: assertion 'GST_IS_DISCOVERER_CONTAINER_INFO (info)' failed

Related: https://gitlab.gnome.org/GNOME/tracker/-/issues/378

This does currently have initialization issues in rawhide (through
recursive gst_init calls), but also it does not provide anything that
will be useful in metadata extraction.

Simply avoid it entirely.

This is needed by SQLite on some circumstances, but these mostly
apply to databases being opened with other users. This is something
that happens on CI, but is not expected to happen in real circumstances.

Anyhow, SQLite does not check for fchown return value, so just error
out softly if that happens.

Look up related CUE files through a SPARQL query

Closes tracker#378

See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/417

This is preferred over epoll_create(), and is used in GLib 2.74 and
later due to https://gitlab.gnome.org/GNOME/glib/-/merge_requests/2868

This reverts commit 8603d673.

The master branch is not protected to allow a specific release process,
which means i can accidentally push to master instead of opening a PR.

This is preferred over epoll_create(), and is used in GLib 2.74 and
later due to https://gitlab.gnome.org/GNOME/glib/-/merge_requests/2868

https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/239

seccomp: Allow epoll_create1()

See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/419

Export the gcovr report using the cobertura format, that enables
Gitlab to show coverage information inline in the diff view.

ci: Enable coverage integration in MR diff view

See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/420

This is not being freed on TrackerMinerFs finalization.

If we fail to create a monitor with errors other than EOVERFLOW,
we would leak the data we just created to attempt to create the
FANotify directory monitor.

Fix issues reported by valgrind

See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/421

Fix flatpak command

See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/418

This internal library depends on generated headers from
libtracker-miners-common, add a dependency on it so meson/ninja
get the order right.

Closes: https://gitlab.gnome.org/GNOME/tracker-miners/-/issues/246

libtracker-miner: Add missing libtracker-miners-common dep

Closes #246

See merge request https://gitlab.gnome.org/GNOME/tracker-miners/-/merge_requests/422