GitLab

these were basically for this serialization, but now that we have a
different version of serialization/deserialization, we don't need them
for this purpose.  Until we see the need for them for specific purposes,
it'd be better to clean them up.

This simplifies some internal part of the code, and eliminates the need for
offsets_size_ (so it's removed).  This change is a preparation for making
the "from raw data" construction more generic to support non absolute labels.
This is a purely internal refactoring, and public interfaces aren't changed.

if one is absoulte and the other is not, they won't have common labels,
so the main logic can detect their relationship is NONE.  We don't need
to handle this case separately for performance (if that was the reason)
because we generally don't expect to do this type of mismatch comparison.

unit tests need to be adjusted because the order is now non 0.

if nlabels == 0, it must not be absolute because otherwise the other
should also be absolute, in which case nlabes != 0.

the originally intended behavior was to return non 0 order for these cases,
so this point was adjusted.
the code was also simplified by removing unnecessary conditions like:
-                     ((last_label_ < getLabelCount()) ||
-                      (other.last_label_ < other.getLabelCount())))) {

this is meaningless because getLabelCount() is last_label_ - first_label_,
so the codntion is actually first_label_ < 0, which is always false.
whatever the real intent of this condition was, such additional checks
didn't seem to be necessary for the intended behavior of compare() anyway,
so I simply removed them.
some test cases were adjusted for the change, and some were added to confirm
related cases that were not tested before.

and remove the unnecessary 'explicit' from it ('non-explicit' usage
wouldn't be a copy constructor in the first place).

Like the previous change, the part of the changes are big, but this is
essentially a straightforward refactoring.

many parts of the code had to be updated, but this is basically pretty
straightforward refactoring (except for exception consideration).

When there's a JSON error, we raise an exception with better
explanation.

As suggested by cppcheck.

As it is not generated, the _inc file is not needed. The documentation
is moved inline to the _python.cc one.

In the python part, we play a little with the reference counts. The data
source client wrapper can now hold a life keeper in addition to the
container, as some data source clients don't come from the container or
are not known.

We have a class to keep a reference to it. This is in C++ only yet, the
python part needs to be implemented.

However, it still needs documentation and some stuff with object
lifetime management.

currently it uses the property that the parent of a subtreeroot is a null node

So it will better support the case when created from the client list.

For the ConfigurableClientList. The constructor works and is tested, but
that is of little use yet.

This fixes a build failure on the build bots.