[test/notification] merge related test classe They have no reason to be distinct.

merge with stable

[server] add a looping task to free unused cnxsets capture by dead pyro When a pyro client disconnects, we're not necessarily notified so they can hang on to their session/cnxset forever. Don't let them do that. The hacky looping task is only installed when pyro is enabled. This optional activation save this very hacky patch from being rejected. closes #2825380

[pkg/debian] update changelog to 3.17.0

[devtools] Use XMLValidator instead deprecated SaxOnlyValidator The renaming was introduced in 198fdadafed6.o This prevent a meta class conflict introduced by logilab-common 0fcd2e050621.

[pkg] update changelog

[pkg] prepare 3.17.0

[pkg] prepare 3.16.4

[repo] straightforward bootstrap sequence. Closes #2841188 * kill fill_schema method which was doing to much different things, in favor of a deserialize_method which is simply reading schema from the db and returning it * set_schema doesn't bother with cubes This allows to easilly follow what's going on and so to cleanup init_cnxset_pool to get the following boostrap order: 1. load cubes (implies loading site_cubicweb) 2. load schema 3. init data sources 4. init cnx sets

[repo] minor cleanups to bootstrap sequence * fix some docstring/comments * move initialization of HooksManager to the __init__.py, this has nothing to do with pool initialization / boostraping

[schemaserial] serialize additional yams parameter for customs type Yams have a new custom type feature. We add support for extra parameter on CWAttribute in schema serialization handling. Closes #2847151

[sql] preprocess_entity uses lgdb helper's SQL converters. The preprocess_entity does not yield anymore the conversion logic for different base types, but relies on a new SQL_CONVERTERS dictionary of the db helper. This will allow later inclusion of new base types.

[repo] kill rebuildinfered feature from Repository.set_schema This clutters thing and is useless in most cases: when schema just have been readen from the database or from the file-system, infered relations have just be calculated. The only use-case where it should be rebuilt is on migration (ie schema modification by editing the meta-model using rql), so move this feature there.

[test] stop deepcopying for nothing

drop xhtml content-type support (closes #2065651) * HTMLStream does not care about xml any more * reqquest.demote_to_html and .xhtml_browser are deprecated * web config: drop force-html-content-type option * adjust tests

[htmlparser] exclude <script> tag from html source lxml are confused by them :-(

[htmlparser] store unaltered source in pageinfo.source Do not store the parser to preserve the source aspect. Store the initial input in the source argument instead. This is very useful when parsing html. In such case we need to drop ``<script>`` tag.

[htmlparser] rename SaxOnlyValidator to XMLValidator

[htmlparser] add missing deprecation message

[form] remove hard coded id html id should be unique. It is clearly not achieved here.

[server/test] use server config in migration This allows a lighter initialisation (no rtag needed). Execution of those test move from 300 secondes to 220 seconds here. A 25% speedup

[session] deprecate `hijack_user` method The semantic of this method is wicked and lead to very hard violation of sanity. (multiple transaction using the very same cursors). We deprecated the API to be able to drop it as soon as possible. The method was added long ago for some notification trick. It is not needed since we sent notification on ``postcommit_event``. (closes #2781782)

[mail] drop hijack_user usage The transaction is now commited before notification. We can create a new session directly and drop `hijack_user` usage. This prepare #2781782

[notification] merge BaseNotificationView and NotificationView They are in the same module now.

[notification] move notification view in ``sobject.notification`` It has no user outside this module. This enforce serversideness of notification and allow future cleanup. No backward compat is set up to prevent circular import. The class has no other user anyway. (closes 2845144)

remove unused import pyflakes say they are unused. Tests suite confirms.

one more merge

[notification] introduce an official `notify_on_commit` function It provides a proper notification api instead of replacing it by yet another clumsy one. Closes #2837251

[cubicweb/doc] Replace dc_type() by cw_etype dc_type() is may be translated. Leading to terrible terrible damage.

[dataimport] backout 6947201033be (related to #2788402) (and add a try: except to cache the intended error) The problem actually comes from the ``MassiveObjectStore`` in the ``dataio`` cube, so it should be corrected there. Here, we only protect it with a ``RuntimeWarning`` so that the user can see the problem. ``value`` is set to ``None`` (whence to ``NULL`` from a database standpoint), so that the data can be nevertheless inserted in the database. However, only the keys present in ``row`` are actually non-'``NULL``'. The real solution is to work out the issue in ``MassiveObjectStore`` directly. The current try/except should only be a temporary hack.