This was failing due to lack of systemd devel packages. Disable
systemd user service files like we do on the static-scan build.

Move away from oldish F37, and use "latest". This should allow
updates to more recent Fedora versions without modifying the
yml definition.

Instead of using the templates provided by gitlab, which have
limited configurability, adopt the approach taken at Mutter, where
duplicate push/merge-request pipelines are prevented, but there is
still a manual escape hatch for running a branch pipeline.

For this, the "repo-sanity" phase has moved to a different stage
prior to everything else, and will require manual triggering if
running on a branch that is not the default one, or one of our
release branches.

Since we separated test/deploy for the website, we can allow the
testing itself on other branches than the default one.

Avoid a warning when running it (align_nl_cont not being boolean
anymore), and specify that these macro newline continuations
should be aligned to the rightmost '\'.

This is just noise in the CI task output.

We had a cli_metadata_dir variable unused in main(), besides
being assigned a value.

This might be used without any value assigned.

Much unlikely, but static analysis complains about this being
possible.

This should never be NULL, so this checks fools
static analysis.

This should not be used initialized, but static scan is fooled
by it.

Assert that the given string is non-NULL. Might hint
the static scan build that this is not possible.

Make it clear to the static scan build that this string should never
be NULL.

Don't make the static scan build think that this variable might be
NULL after test output is collected.

The TrackerDBManager object is already automatically managed with
g_autoptr, these error paths might lead to a double free.

CID: #1511357

These variables were already being freed a little bit above in
the code.

CID: #1501117

This did not lead to errors since the first connection created
held a reference on the file. Still not straightforwardly correct,
so move the g_object_unref() call to a point where this variable
is no longer used directly.

CID: #1501125

Some CI fixes/improvements

See merge request https://gitlab.gnome.org/GNOME/tracker/-/merge_requests/657

Fixes for "static-scan" build warnings

See merge request https://gitlab.gnome.org/GNOME/tracker/-/merge_requests/658

This is the de-facto default nowadays, we can always set up and use
the mutex. Also, this might give Coverity a hint that there's no
chances about the flag being on when locking, but off when unlocking.

CIDs: #1584275, #1584279, #1584280, #1584284, #1584285, #1584293

This is now unused.

Assert these returning TRUE where expected, in order to avoid
Coverity from freaking out.

CIDs: #1501101, #1501120, #1501134, #1501137, #1505932, #1505933, #1507949, #1511341, #1518802

Assert for these, to make Coverity happy.

CIDs: #1501110, #1501128, #1584277

This was initially motivated by Coverity not being happy with
us not checking g_unlink() return value. But we don't need/use
settings anywhere in the Tracker repository for a long time, we
can drop this old piece of dconf/settings manipulation.

CID: #1501107

CIDs: #1501129, #1505931

It may not happen that we enter in error paths with a GError
without a message. Stop accounting for that.

CIDs: #1511356, #1511358

As done everywhere else.

CID: #1496386

These cursors can't be NULL after the error checks, don't let
Coverity think that they might be.

CIDs: #1511359, #1511355

This is already done before.

CID: #1530395

The mapped file contents were needlessly duplicated, also port to
g_clear_pointer().

CID: #1501123

In some error paths, the bus_name and object_path variables could be
leaked.

CID: #1584290

CID: #1584296

CIDs: #1515042, #1515043, #1532056, #1584283

Use the return values from json-glib API, like everywhere else.

CID: #1584286

If multiple threads try to iterate the cursor forward, there
might be issues if both try to mark the cursor as finished.
It's unclear this has actual effect, and it is otoh clear that
this is API misuse. Still, protect around this.

CID: #1584289

Fixes for some reported errors in Coverity

See merge request https://gitlab.gnome.org/GNOME/tracker/-/merge_requests/659

Debug logging in functional-tests is controlled by setting
`TRACKER_DEBUG=tests` in the host environment, this wasn't working
due to meson.build clobbering the value.

The ASK query would return a variable name with a load of SQL noise,
now the variable is named "result".

This means the endpoint needs to support POST queries.

The SPARQL 1.1 standard says that this MAY be supported:
<https://www.w3.org/TR/2013/REC-sparql11-protocol-20130321/#query-via-post-direct>

The previous method using GET says the same, it MAY be supported:
<https://www.w3.org/TR/2013/REC-sparql11-protocol-20130321/#query-via-get>

The 500 error code is meant for internal errors, while request errors
should be 400. This is in line with the SPARQL 1.1 Protocol standard:
<https://www.w3.org/TR/2013/REC-sparql11-protocol-20130321/#update-failure>

Exercise various query methods and error cases.

Extend HTTP endpoint and test coverage

See merge request https://gitlab.gnome.org/GNOME/tracker/-/merge_requests/645

This is unused, firstly always unset then creating TrackerDBManager
instances, secondly ineffective since it also needs the
SQLITE_OPEN_SHAREDCACHE flag when creating sqlite3 DB instances, which
we never pass.

Closes: https://gitlab.gnome.org/GNOME/tracker/-/issues/411

core: Drop shared_cache toggle

Closes #411

See merge request https://gitlab.gnome.org/GNOME/tracker/-/merge_requests/660