GitLab

Update the build system to use pybuild.

The new binary package with Python 3 support is named
python3-cubicweb-skos according to current standards.
Python 2 package is also updated to use the proper namespace
and a transitional package has been introduced to handle
the migration.

Support for Python 3.8 is disabled as it is currently broken
in Debian unstable.

We currently ignore test/test_ccplugin.py as it requires
a PostgreSQL server. Running it through autopkgtest seems
like a more robust approach, but this remains to be
implemented.

Drop useless __pkginfo__.py symlink

On python3, csv module uses unicode strings. Since we use bytes strings
on our side, we need to decode when calling csv module ourselves.

We now explicitly use a file opened in bytes mode to produce the stream
for LCSV and SKOS data import. Similarly, when using an in-memory
object, we use a BytesIO. This does not change anything in python2, but
this is required to keep things working on python3.

Previously, the "lang" value could be None because of getattr(...,
None). Subsequent usages of this value do not look quite right
(e.g. str(None) or set([None])). Instead, we now catch attribute error
and build the md5hash without the None value as well as insert an empty
set for language_code attribute when importing a label.

This also prepares for upgrade to python3.

Use cubicweb's BaseTestCase since we use datapath() method.
Drop usage of non-standard set_description() method and use subtests.

This used to be needed when the cube was old-style.

Spotted by flake8 version 3.6.0.

In SAEM we need to render the concept tree as a jquery tree. So we
introduce a render_tree method intended to be overidden by specific
cubes.

This module has been deprecated (yesterday) but I now update downstream
projects to avoid relying on it.