[sobjects/test] check that register_user properly sets user name and email address Related to #3020639

[server] Refactor Repository.register_user into a CubicWeb service (closes #3020639) [jcr: move commit to the caller; add item in release notes; fix NameErrors]

[migration] set a cnxset before process_script() Not pretty, but I don't have a better idea right now and this at least ensures migration scripts get run in the expected environment and can access sql stuff as before.

[server] make internal_cnx get and keep a connection to the db - internal_session worked that way - it's less surprising for users if their internal connections "just work", rather than having to deal with the cnxset when using low-level APIs Includes a backout of 3f62606c01a2 "[repo] Fix register_user" which is no longer needed. Depend on logilab-database 1.12.1 to avoid errors due to pointless check in python's sqlite3 binding.

[server/session] make commit/rollback obey ensure_cnx_set Inside the ensure_cnx_set context manager, commit and rollback aren't allowed to dispose of the cnxset.

[repo] fix deprecation messages for stats and gc_stats

[cwconfig] use "foo not in bar" instead of "not foo in bar"

[debian] Add Breaks on a few cubes that need changes for 3.19 - mercurial_server is affected by composite deletion changes - forgotpwd relied on session data to be forwarded to notification view - registration hooks into the authentication stack

[hooks] run cascading delete hook later Applications may have their own before_delete_entity hooks that rely on the composite relations still existing.

[devtools] set PGPORT/PGHOST when calling pg_ctl start Helps pg_ctl find the socket on squeeze, otherwise it looks in /var/run/postgresql.

[server/test] update to deal with lgdb sqlserver fix See https://www.logilab.org/ticket/235748

[test] Fix regressions from 96549de9dd70 "[test] use assertIn where appropriate"

[repo] Fix register_user Make sure we have a connection to the db around user creation.

[test] use assertNotIn where appropriate Automatically generated with: find -name unittest_*.py -o -name test_*.py | xargs sed -i -e 's/assertFalse(\([^,]*\) in \([^,]*\))/assertNotIn(\1, \2)/' plus revert of a couple false matches.

[test] use assertIn where appropriate Automatically generated with: find -name unittest_*.py -o -name test_*.py | xargs sed -i -e 's/assertTrue(\([^,]*\) in \([^,]*\))/assertIn(\1, \2)/'

[doc] Add removal of old multi-sources to 3.19 release notes

[pkg] Bump to 3.19.0

[web/debug] remove code broken by session refactoring session.cnx and cnx.check() no longer exist.

[web/debug] Fix check for open sessions itervalues returns an iterator, which never evaluates to False. Regression from 0bb18407c053 "[toward py3k] rewrite dict.keys() and dict.values() (part of #2711624)".

[cwconfig] Remove useless 'ui-cube' configuration option (closes #3728904) The mechanism introduced could never actually be used as it brings dependency loops to the cube graph. Just get rid of it. This reverts commit 4032499c701e.

[test] Use RepoAccess.shell() in unittest_predicates

[devtools] add 'shell' method to RepoAccess, deprecate CubicWebTC's Lets us control the connection's lifetime.

[repo] make extid2eid work with either a session or a connection The only place it should matter is for restoring pending_operations in case of failure. Which is not exercised by the test suite. Here be dragons.

[repository] extid2eid's 'commit' argument no more necessary This removes the need for creating an internal session.

[repository] complete argument is no more needed since we have no more 'true' multisource. It is always false.

[repository] 'session' argument is always given to type[_and_source]_from_eid, make it mandatory and simplify code accordingly

[repository] 'session' argument is always given to extid2eid, make it mandatory and simplify code accordingly

[web/data] Remove broken jquery-json plugin (closes #3590335) Conversion of Date() objects is somehow broken (don't know why beyond that it's due to jquery-json's monkeypatched Date.toJSON() method). Use the now-standard JSON.stringify() instead. http://caniuse.com/#feat=json says it's supported by all browsers, and even IE going back to version 8.