3.25 (UNRELEASED)
=================

New features
------------

* A new option `connections-pooler-enabled` (default yes) has been added. This
  allow to switch off internal connection pooling for use with others poolers
  such as pgbouncer_.

.. _pgbouncer: https://pgbouncer.github.io/

* In `deleteconf` view (confirmation before deletion), the list of first-level
  composite objects that would be deleted along with the primary entity is
  displayed (01eeea97e549).

* The ``cubicweb.pyramid`` module now provides a Paste application factory
  registered as an entry point named ``pyramid_main`` and that can be used to
  run a Pyramid WSGI application bound to a CubicWeb repository.

* A new configuration type ``pyramid`` has been added to create CubicWeb's
  instances (through ``cubicweb-ctl create -c pyramid <basecube> <appid>``).
  This configuration bootstraps a CubicWeb instance that is essentially a
  repository plus the minimal setup to run a Pyramid WSGI application on top
  of it. Noticeably, it does not ship all *web* configuration but rather
  relies on configuration declared in a ``development.ini`` file for any
  Pyramid application.

* A new way to declare workflows as simple data structure (dict/list) has been
  introduced. Respective utility functions live in ``cubicweb.wfutils``
  module. This handles both the creation and migration of workflows.

* A new IDublinCore adapter has been introduced to control the generation of
  Dublin Core metadata that are used in several base views.

* It is now possible to *derive* rtags using their ``derive`` method
  (0849a5eb57b8). Derived rtags keep a reference to the original rtag and only
  hold custom rules, allowing changes which are done in the original rtag after
  derivation to be still considered.

* A new ``cubicweb-ctl scheduler <appid>`` command has been introduced to run
  background and periodic tasks of the repository (previously called *looping
  tasks*). In a production environment, a process with this command should be
  run alongside with a WSGI server process (possibly running multiple
  processes itself).


Backwards incompatible changes
------------------------------

* As a consequence of the replacement of the old looping tasks manager by a
  scheduler, all cubicweb-ctl's "start" commands (i.e. ``start``, ``pyramid``,
  ``wsgi``) do not start repository *looping tasks manager* anymore, nor do
  they start the scheduler. Site administrators are thus expected to start
  this scheduler as a separate process. Also, registering looping tasks (i.e.
  calling ``repo.looping_tasks()``) is a no-op when the repository has no
  scheduler set; a warning is issued in such cases. Application developers may
  rely on repository's ``has_scheduler`` method to determine if they should
  register a looping task or not.

* In ``cubicweb.pyramid``, function ``make_cubicweb_application`` got renamed
  into ``config_from_cwconfig`` (950ce7d9f642).

* Several cleanups in repository's session management have been conducted
  resulting from changes introduced in 3.19 release. Among others, the
  ``cubicweb.server.session.Session`` class has been dropped, and request
  ``session`` attribute is now tight to a web session whose implementation
  depends on the front-end used (twisted or pyramid). Hence this attribute
  should not be accessed from "repository side" code (e.g. hooks or operations)
  and has lost some of his former attributes like ``repo`` which used to
  reference the repository instance. Due to this, you don't have anymore access
  to session's data through the connection, which leds to deprecation of the
  ``data`` attribute and removal of ``get_shared_data`` and ``set_shared_data``
  methods which are deprecated since 3.19.

* Support for 'https-url' configuration option has been removed
  (4516c3956d46).

* The `next_tabindex` method of request class has been removed (011730a4af73).

* The `cubicweb.hook.logstats.start` hook was dropped because it's looping
  task would not be run in a web instance (see first point about repository
  scheduler).

* ``uicfg`` rules  to hide the opposite relation of inlined form are not anymore
  automatically added, because this was actually done randomly and so not
  reliable, so you'll have to add them manually:

  ::

    autoform_section.tag_subject_of(('CWUser', 'use_email', 'EmailAddress'),
                                    'main', 'inlined')
    autoform_section.tag_object_of(('CWUser', 'use_email', 'EmailAddress'),
                                   'inlined', 'hidden')