doc/changes/3.22.rst
author Guillaume Vandevelde <gvandevelde@logilab.fr>
Thu, 13 Feb 2020 13:56:12 +0100
branch3.26
changeset 12876 fe9bd0e937e4
parent 11104 269317987dc6
permissions -rw-r--r--
[pkg] version 3.26.15

3.22 (4 January 2016)
=====================

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

* a huge amount of changes were done towards python 3.x support (as yet
  incomplete).  This introduces a new dependency on six, to handle
  python2/python3 compatibility.

* new cubicweb.dataimport.massive_store module, a postgresql-specific store
  using the COPY statement to accelerate massive data imports.  This
  functionality was previously part of cubicweb-dataio (there are some API
  differences with that previous version, however).

* cubes custom sql scripts are executed before creating tables.  This allows
  them to create new types or extensions.

* the ``ejsonexport`` view can be specialized using the new ``ISerializable``
  entity adapter.  By default, it will return an entity's (non-Bytes and
  non-Password) attributes plus the special ``cw_etype`` and ``cw_source``
  keys.

* cubes that define custom final types are now handled by the ``add_cube``
  migration command.

* synchronization of external sources can be triggered from the web interface
  by suitably privileged users with a new ``cw.source-sync`` action.

User-visible changes
--------------------

* the ldapfeed source now depends on the `ldap3` module instead of
  `python-ldap`.

* replies don't get an ``Expires`` header by default.  However when they do,
  they also get a coherent ``Cache-Control``.

* data files are regenerated at each request, they are no longer cached by
  ``cubicweb.web.PropertySheet``.  Requests for data files missing the instance
  hash are handled with a redirection instead of a direct reply, to allow
  correct cache-related reply headers.

API changes
-----------

* ``config.repository()`` creates a new Repository object each time, instead of
  returning a cached object.  WARNING: this may cause unexpected issues if
  several repositories end up being used.

* migration scripts, as well as other scripts executed by ``cubicweb-ctl
  shell``, are loaded with the print_function flag enabled (for backwards
  compatibility, if that fails they are re-loaded without that flag)

* the ``cw_fti_index_rql_queries`` method on entity classes is replaced by
  ``cw_fti_index_rql_limit``, a generator which yields ``ResultSet`` objects
  containing entities to be indexed.  By default, entities are returned 1000 at
  a time.

* ``IDownloadableAdapter`` API is clarified: ``download_url``,
  ``download_content_type`` and ``download_file_name`` return unicode objects,
  ``download_data`` returns bytes.

* the ``Repository.extid2eid()`` entry point for external sources is deprecated.
  Imports should use one of the stores from the ``cubicweb.dataimport`` package
  instead.

* the ``cubicweb.repoapi.get_repository()`` function's ``uri`` argument should
  no longer be used.

* the generic datafeed xml parser is deprecated in favor of the "store" API
  introduced in cubicweb 3.21.

* the session manager lives in the ``sessions`` registry instead of ``components``.

* ``TZDatetime`` attributes are returned as timezone-aware python datetime
  objects.  WARNING: this will break client applications that compare or use
  arithmetic involving timezone-naive datetime objects.

* creation_date and modification_date attributes for all entities are now
  timezone-aware (``TZDatetime``) instead of localtime (``Datetime``).  More
  generally, the ``Datetime`` type should be considered as deprecated.

Deprecated code drops
---------------------

* the ``cubicweb.server.hooksmanager`` module was removed

* the ``Repository.pinfo()`` method was removed

* the ``cubicweb.utils.SizeConstrainedList`` class was removed

* the 'startorder' file in configuration directory is no longer honored