Closes: #59

The list of tests is now at the top of each meson.build file, so it's
harder to miss them out by mistake.

Corresponds with tracker!5

We made a big compromise when splitting tracker core from tracker-miners
in that the common code that was needed by both parts would end up
duplicated. It's ugly but it works fine at the moment and allows us to
keep all of the common code private.

I had an issue when trying to embed tracker core into tracker-miners as
a Meson subproject though. Having two targets named tracker-common
caused confusion as duplicate targets aren't allowed, but they are not
quite equivalent so we can't just pick one or the other.

To work around this, I've renamed the copy in this repo to
tracker-miners-common. This only affects the target names, not the
actual function names.

https://bugzilla.gnome.org/show_bug.cgi?id=789725

The miners are being split from the core tracker package.

On both autotools/meson builds, datadir and libdir for
private data has been changed to be separate from the tracker
core. Same goes for the gettext package and other bits.

Additionally, avoid installing the dbus xml descriptions.
That's fairly non-standard and unnecessary with introspection.

There's no reasonable situation why we should change this from
a library. Initialization (and shutdown) was also mostly useless
as it would just set the same settings that it just retrieved.
It's been turned into a sanity check for the relevant places.

First, instead of using add_global_arguments() we now use a variable
to set global compiler args. It turns out that add_global_arguments()
is broken if you ever want to include your project as a subproject of
something else.

Second, dependencies weren't being specified correctly. These are
hopefully now fixed to avoid any more random build failures.

This removes various code paths that we believe to be unused.

  * The userguides miner has been removed altogether.
  * The application miner no longer parses MeeGo-style .desktop files
  * The TRACKER_DISABLE_MEEGOTOUCH_LOCALE environment flag is gone
    (it already did nothing)
  * Character set detection with libmeegotouch is removed, only enca
    or libicu are supported
  * Meego-specific flags are gone from .desktop files
  * Functional tests have lost their vestigal Scratchbox and Aegis support

There are 5 ontologies referencing Maemo, 4 of which are largely unused.
These have been kept around in case anyone is using them outside
Tracker.

https://bugzilla.gnome.org/show_bug.cgi?id=783555

This is useful for people who are not building from Git but still want
to avoid Autotools.

See: https://mail.gnome.org/archives/desktop-devel-list/2017-April/msg00091.html

See <http://mesonbuild.com/> for information about Meson.

Remaining issues:

  * There's no `make dist` equivalent. We currently produce release
    tarballs containing the output files of the Vala compiler. We need
    to think through whether we can stop doing that. Shipping the
    generated .c files does make the Vala preprocessor useless so
    it would be good if we can stop.

  * The Firefox, Thunderbird, Evolution and Nautilus plugins are not
    built

  * https://github.com/mesonbuild/meson/issues/671 -- means we can't
    depend on tracker_common_dep in most places and have to manually
    set link_with, include_directories and sources instead.

  * https://github.com/mesonbuild/meson/issues/1469 -- without this we
    have to install generated Vala headers using a script

  * https://github.com/mesonbuild/meson/issues/1229 -- means adding
    the #include guard to libtracker-sparql/tracker-generated-no-checks.h
    is a pain

  * The test suite has some spurious failures.

Here's a rough speed comparison.

Meson:
        time sh -c 'meson .. --prefix=/opt/tracker-meson -D bash_completion=/opt/tracker-meson/share/bash-completion -Dsystemd_user_services=/opt/tracker-meson/lib/systemd/user -Ddbus_services=/opt/tracker-meson/share/dbus-1/services && ninja-build -j 4 && ninja-build install'

        real  1m8.194s
        user  2m16.962s
        sys   0m20.532s

Autotools:
        time sh -c './configure --prefix=/opt/tracker-autotools --with-bash-completion-dir=/opt/tracker-autotools/share/bash-completion --with-session-bus-services-dir=/opt/tracker-autotools/share/dbus-1/services --disable-nautilus-extension && make -j 4 && make install'

        real  2m37.750s
        user  4m37.214s
        sys   0m54.806s

        Plus 30+ seconds of ./autogen.sh first.

Note that Meson builds may fail if your source tree has generated files
from an Autotools build in there. If you see errors about duplicate
definitions, first try cleaning your source tree (use `git clean -dfx`,
but make sure you commit any work first!!)

It's a bit ugly that GLib string functions take signed chars and
SQLite'3 string functions return unsigned chars. We could disable
that warning instead but maybe it's useful in some other places.

I'm not sure this test is actually useful for anything, maybe we should
just delete it...

The test passes when used with glibc's getline(), but the fallback
implementation we ship will return EINVAL if the size pointer is not set
to zero before the call. The man page suggests that this is valid
behaviour for a getline() implementation, so let's fix the test.

Those are the ones used now.

For a long time, all the Tracker extractors have manually constructed a
SPARQL update command using TrackerSparqlBuilder to represent their
output. This commit changes all of them to use the TrackerResource
class instead, which makes the code a lot more concise and readable.

This introduces some API breaks in the internal libtracker-extract
library. This has been a private library since Tracker 0.16 or earlier,
so it's fine.

If the extractors only output SPARQL then they are only useful to
people who are using a SPARQL store. Now we can output a serialization
format like Turtle as well. This will hopefully make the extract modules
useful outside of Tracker itself.

I've tried to preserve the behaviour of the extractors as much as
possible, but there are two things that are now handled differently:

  * nao:Tag resources are given a fixed URI based on the tag label, such
    as <urn:tag:My_Tag>. Previously they were inserted as blank nodes,
    so tracker-store would give them unique IDs like <urn:uuid:1234...>

  * All extractors created nco:Contact resources for content publishers,
    but previously some would assign fixed URIs based on the name
    <urn:contact:James%20Joyce>, while others would insert them as blank
    nodes so they would be assigned unique IDs like <urn:uuid:1234...>.
    Now, all extractors create nco:Contact resources with fixed URIs
    based on the content creator's name.

https://bugzilla.gnome.org/show_bug.cgi?id=767472

Now that extraction is performed after the file got initially
inserted and has an URN, there's no reason we can't pass the
URN to the TrackerExtractInfo. This would make URNs available
to extract modules, which have some places where the SPARQL
produced would be clearer.

This actually opens the path for quite a cleanup, we might
pass a single TrackerSparqlBuilder to extract modules instead
of doing the postprocessing we currently perform.

They had gotten swapped around. See:
http://www.cipa.jp/std/documents/e/DC-010-2012_E.pdf

https://bugzilla.gnome.org/show_bug.cgi?id=758295

After commit ede17cc2 this was failing to compile.

https://bugzilla.gnome.org/show_bug.cgi?id=745583

- Return FALSE when we don't add to the TrackerSparqlBuilder, TRUE otherwise
- Skip any unit tests when --enable-guarantee-metadata is not configured

This reverts commit 898ac3ca.

Conflicts:
	tests/libtracker-extract/tracker-guarantee-test.c

https://bugzilla.gnome.org/show_bug.cgi?id=733536

tests: add missing newline in test_guarantee_title that caused a test failure with the new tap driver

https://bugzilla.gnome.org/show_bug.cgi?id=734464

The --enable-guarantee-metadata flag ensures that nie:title and
nie:contentCreated will always be set for a given file, even if they
need to be guessed based on the filename and mtime.

It was previously disabled by default, although the functional tests
rely on it being enabled.

There should be no harm in making this behaviour the default, and
removing the option. If this turns out to have unintended consequences,
we should either fix the code in question, or remove it completely.

https://bugzilla.gnome.org/show_bug.cgi?id=733536

Don't install libtracker-extract system-wide and don't generate any
introspection data.

This is a follow-up commit to 60fdc9b9

We imported Makefile.decl from glib when it was using Subversion, a long time
ago. The script had a few issues (like running all unit tests twice on make
distcheck) and needed to be updated.

I've stolen a latest version from GLib's glib.mk and also their m4 macros
which are required to make this all work too.

This gives us the advantage of defining installable and uninstallable test
cases as well as test data, scripts and other useful stuff.

The debug output also looks quite good.
And distcheck passes :)

Now we're using g_test_trap_subprocess() instead and that requires GLib 2.38.

Tracker now depends on GLib 2.38.

We require GLib 2.35.x, no need to check for previous versions.

When a title is generated from the filename (--enable-guarantee-metadata), return it to the
calling extractor so that it can be used for media art. This allows storing video poster art
for any video, provided guaranteed metadata is enabled.

Calling g_thread_init with GLib 2.31 requires explicit dependency on
gthread-2.0. However, g_type_init initializes threads since GLib 2.24,
so it is no longer needed.

Milliseconds were removed in a special case. Not needed anymore, because
now we want milliseconds in the dates.

tracker_test_helpers_cmpstr_equals was written when g_assert_cmpstr
didn't exist but both had the same intention.