3.24 (2 November 2016)
======================

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

* Various bits of a CubicWeb application configuration can be now be
  overridden through environments variables matching configuration option
  names prefixed by ``CW_`` (for instance ``CW_BASE_URL``).

* Cubes are now standard Python packages named as ``cubicweb_<cubename>``.
  They are not anymore installed in ``<prefix>/share/cubicweb/cubes``. Their
  discovery by CubicWeb is handled by a new setuptools entry point
  ``cubicweb.cubes``. A backward compatibility layer is kept for "legacy"
  cubes.

* Pyramid support made it into CubicWeb core. Applications that use it
  should now declare the `cubicweb[pyramid]` dependency instead of
  `cubicweb-pyramid`.

* New `NullStore` class in ``cubicweb.dataimport.stores`` as new base class for
  every store, and allowing to test your dataimport chain without actually
  importing anything.


Major changes
-------------

There has been several important changes to the core internals of CubicWeb:

* Dropped `asource` and `extid` columns from the `entities` table as well as the
  index on the `type` column, for a sensible optimization on both massive data
  insertion and database size / index rebuilding.

* Dropped the `moved_entities` table and related mecanism to remember that an
  entity has been moved from a source to the system database - this is now the
  responsability of source's parser to detect this (usually by remembering its
  original external id as `cwuri`).

* Dropped the original 'give me an eid for this extid, but call me back on
  another line if it has to be created' mecanism on which the ``cwxmlparser`` was
  relying, in favor of parsers using the dataimport API. This includes dropping
  the ``cwxmlparser``. If you're using it, you'll have to write a specific
  parser, examples to come.

* Dropped source mapping handling (schema, views, logic) with its client
  the ``cwxmlparser``. This is not worth the burden, specific parsers should be
  preferred.

The above changes lead to the following API changes:

* `req.entity_metas(eid)` doesn't return anymore a 'type' nor 'source' keys, use
  `req.entity_type(eid)` instead or 'cw_source' relation to get those
  information,

* deprecated `entity.cw_metainformation()`, which doesn't return anymore it's
  'source' key,

* dropped `repository.type_and_source_from_eid(eid, cnx)`,
  `repository.extid2eid(...)` and `source.eid_type_source(cnx, eid)`,

* dropped `source.support_entity(etype)` and `source.support_relation(rtype)`,

* dropped 'cw_source' key from default JSON representation of an entity,

* dropped `source_uris()` and `handle_deletion(...)` method from datafeed parser
  base class, deletion of entities is now the responsability of specific
  implementation (see ``ldapparser`` for example),

* entities from external source are always displayed in the UI with a link
  to the local entity, not the original one simplifying `entity.absolute_url()`
  implementation and allowing to drop `use_ext_eid` argument of
  `entity.rest_path()` (though it's still supported for backward compat).


Changes to the massive store
-----------------------------

Several improvments have been done to ``cubicweb.dataimport.massive_store``,
with among the more important ones:

* Extended store API to provide more control to end-users:
  `fill_entities_table`, `fill_relation_table`, `fill_meta_relation_table`.

* Dropped `on_commit` / `on_rollback` arguments of the constructor.

* Use a slave specific temporary table for entities insertion as for relations
  (should improve concurrency when using in master/slaves mode).

* Delay dropping of constraint to the `finish` method, avoiding performance
  problem that was occuring because indexes were dropped at store creation time.

* Consider the given metadata generator when looking for which metadata tables
  should have their constraints dropped.

* Don't drop index on `entities.eid`, it's too costly to rebuild on database
  with some million of entities.