GitLab

Its two class methods get_schema and get_schema_hash will only compute
the schema once. They where computed every time the schema was
requested.

Closes #1.

This object generates a schema usable by the cwclientlib API.
Enable the schema generation to be done by the backend.

- Python 2 support removed.
- File cubicweb-rqlcontroller.spec seemed to be usefull only for
  Python 2.6, so we removed it.
- We use mainly Debian 10 "Buster" that ships python3.7 for
  python3, so we consider that we won't support previous versions
  anymore. This cube is maintained just a bit and users should
  upgrade, so it does not sound a problem to remove support for
  things that should be upgraded.
- With the same logic, the current last version of CubicWeb
  (3.27.3) is now required. In fact, the needed code in CubicWeb
  is not yet in a version, so it is not accurate, because the
  next one will be needed, but we would like not to wait to
  publish a new version of rqlcontroller.

We decide to support only the newest version of CubicWeb. Like
this, it is easier to maintain and encourages users to upgrade.

This introduces a version 2.0 of the rqlio protocol that is able to answer
select RQL queries with symbolic bindings (with the names of the selected
variables) instead of the positional rows. Because this breaks compatibility
with the 1.0 protocol, this change warrants a new version. The previous version
continues to work alongside the new one.

The end result is the ability to answer the query Any X WHERE X is CWEtype with
[ { 'rows': [ [101], [102], [103], ...],
  'variables': ['X'] } ]
instead of
[ [[101], [102], [103], ...] ]

In protocol version 2.0, when there are no variable names, the variables member
is simply empty. In addition, the value of the 'rows' member of the response
object is exactly the same as the total response in the previous protocol, which
should help write fallback code in client libraries.

sometimes webob (or cgi) decide that json field content should be
BufferedRandom and not BytesIO in this case we cannot use getvalue
method

- move cube files to new cube folder
- update setup.py, MANIFEST.in, __pkginfo__.py

The ValueError only concerns json.loads() call.

This is the same logic as in 59eab509372e which handled the
non-multipart case. Since this multipart requests mode had no test
coverage, we add a test for it. Notice that this is the mode used by
cwclientlib's rqlio().

This is expected by CubicWebPublisher.
We use request's encoding to encode the JSON value.

In Python3, json.load() only accepts text content. So we decode
_cw.content's value using request's encoding before passing it to
json.load(). As a consequence of using .getvalue() on BytesIO (content
is a BytesIO), we no longer need to .seek() after reading the content.

The "json" name in cubicweb.utils is just the "json" module from stdlib.
Make this explicit.

We replace CubicWebServerTC, which uses Twisted, by CubicWebWsgiTC.
In particular, the latter will work on Python 3 whereas this is not
supported in the former.

This is needed to make the project installable with Python 3 (which will
be used by tox by default).

Closes #17047898.

Related to #17047898.

Related to #17047898.

Drop useless stuff and make the file flake8-compliant.

Related to #17047898.

Related to #17047898.