[etwist] pass the repository to the root resource This is another step toward a cleaner instantiation scheme for the repo. The http test now pass an already created object to the etwist server instead of relying on the config cache. Related to #2249513

[repo] move repo.gc_stats to Service API (closes #2951068) We currently have a method on the repo that anyone can call. `call_service` gives a similar thing, modulo: * can be restricted access to manager only (thanks to selection on services), * less unrelated code on the repository class, * no need to fetch a reference to a repo object from the client side * (`req._cnx.repo` is not an API and config.repository() is on its way out). The old way to access this information (repo.gc_stats()) is deprecated.

[repo] move repo stats to Service (closes #2951067) We currently add a new method on repo that anyone can call. Call service allows to reach the same result with: * Restricted access to manager only (thanks to selection on services), * Less unrelated code on the repo. * No need to fetch a reference to a repo object from client side. `req._cnx.repo` is not an API and config.repository() is on its way out. The old way to access this information (repo.stats()) is deprecated.

[service] drop the asynchronous execution possibility Call_service was able of both sync and sync execution making the API confusing. There is not user of the async case. We drop the async argument in favor of synchronous execution only. This makes call_service the official API to call server side code from the client side. This is a remplacement for the usual monkey patching of the repo object. The zmq notification bus is a solid alternative for codes that needs to start an async execution.

[auth] pass `repo` instead of `vreg` to SessionManager and AuthenticationManager Those object actually need a repo object to connect and create new session object. They used to retrieve the repo object using ``vreg.config.repository()``. I'm trying to sanitise the initialisation stack and get ride of ``config.repository()``, a function that mix factory and cache. Retrieving the vreg from the repo is trivial (``repo.vreg``) so we now pass the repo object instead of the vreg.

web/application: instantiate the repository outside of CubicWebPublisher This improves decoupling and allows ``CubicWebPublisher`` user to choose which argument are used during ``Repository`` instantiation. In particular is allows caller to provide a ``TaskManager``. The `vreg` argument of publisher is made mandatory. See CubicWebPublisher docstring for details. Related to #2249513

remove vreg argument to CWPublisher The passing a vreg to CWPublisher was used by test to allow reused of an existing vreg. We dropped this feature two commit ago and can safely removes it. Removing this argument allows later cleanup and improvement on the CWPublisher and Repository front. closes #2944813

[devtools] vreg becomes a class ppty After multiple devtools refactoring, the vreg used for test is now *always* got from the repository. Making this explicit is a step toward simplification of the vreg handling during various initialisation phases. Related to #2944813

[devtools] drop unused `reset_schema` and `reset_vreg` class attribute They have no effect and no user... Related to #2944813

[connection] move security control logic on Connection The _security_enabled context manager now operate on connection. We have to keep a specific entry point in Session to ensure Connection object are properly cleaned up when using session to "manage" Connection live cycle. Related to #2503918

[connection] move hook control logic on Connection The _hook_control context manager now operate on connection. We have to keep a specific entry point in Session to ensure Connection object are properly cleaned up when using session to "manage" Connection live cycle. Related to #2503918

[session] update _hook_control docstring

[connection] reinstall cnx.data as cnx.transaction_data Too much code expects self._cw.data to be session data. backout 9b2f68916474 Related to #2503918 Related to #2912807

[connection] give access to is_internal_session boolean Needed if we are to use this object like we currently do with session. Related to #2503918

[req] drop from_controller on non WebRequest object (Closes #2901079) The `controller` concept is purely a web things. It does not belong to generic `req` object. The `build_url` code is slightly changed to handle that. This probably requires extended cleanup later. This is a barely used and very internal API. No deprecation period

[connection] add logging method on connection If we are to use this object like we currently do with session, we need to have the logging method as a lot of code use them.

rename server.session.transaction into server.session.connection The ongoing rework of the API to access the database include the splitting of Session in two objects: The ``Session`` object will only hold credential (user) and global session data. A new object will publicly emerge to handle a database access. This object is named ``Connection`` since that the way database accessors with the same property are named in other system. So we renamed the ``Transaction`` object into ``Connection``. The ``Transaction`` object have already grown in the direction of something directly usable by generic code, but the name ``Transaction`` is ill suited as such object can be used for multiple database transaction in a row. Related to #2503918

[application] call req.set_session in application.main_handle_request The Session handling chain is no more responsible for calling req.set_session. It just returns a valid session and lets the caller link it to the Request. This opens the way to explicitly creating and closing a connection/transaction in ``application.main_handle_request``. Related to #2503918

[session-handler] use session directly to update last usage We don't really need the WebRequest for that. Not using the WebRequest to access the cubicweb repository here will allow a delayed set_session. Related to #2503918

[application/connect] simplify connection logic ``application.connect`` now either sets a full featured ``DBAPISession`` to the ``WebRequest`` object or raises ``AuthenticationError``. The creation and usage of a fake DBAPISession is now handled by ``main_handle_request`` when needed. This means that fake DBAPISession are no longer tracked by the session manager and that user are not given anyway to retrieve them for a later request. This fake DBAPISession is still passed to ``core_handle`` because multiple cubes like registration or forgotten password need this behavior. We would like to get ride of it in the future. This clarification of the connection API greatly simplifies ``DBAPISession`` retrieval//creation process opening the way to improvements in this area. Related to #2503918

Drop hijack user (closes #2901093) Deprecated since 3.17.0. Dropping this function help the session cleanup business.

[sources] drop support for ldapuser source (closes #2936496) The ldapfeed source is a replacement for ldapuser. Use it instead.

merge stable back into default

3.17 is the new stable

Added tag cubicweb-version-3.16.5, cubicweb-debiann-version-3.16.5-1, cubicweb-centos-version-3.16.5-1 for changeset 810a05fba1a4

[pkg] prepare 3.16.5

Added tag cubicweb-version-3.17.2, cubicweb-debian-version-3.17.2-1 for changeset 195e519fe97c

[pkg] prepare 3.17.2

repository: make tests pass again The tests use mono-threaded pyro, which breaks assumptions made in e27337dfec8c. To fix that, remove sessions from the _pyro_sessions dict when they're closed, and force the test pyro client to actually disconnect when it's done.

repository: monkey patch pyro connection handling to detect clients going away When a pyro client goes away we need to close their session (to release their cnxset). Add a dict to the repository mapping pyro client threads to session object. Closes #2932058