[req,web/request] Move language prefix handling into web request This changes part of d8830e2bd2e0 by unsetting "lang" attribute of RequestSessionBase in particular (because it appears to break many downstream applications). The new implementation relies on a new "build_url_path" method on RequestSessionBase that is overridden in _CubicWebRequestBase to handle the language prefix in URL. As a consequence build_url in RequestSessionBase is not aware of this prefix anymore (contrary previous implementation from d8830e2bd2e0). Move respected tests from unittest_req.py into unittest_request.py. Commenting out part of test_handle_request_no_lang_negotiation_fixed_language in unittest_application.py because it now fails because of some obscure bug that cannot be resolved at the moment. Related to #15743487.

[test] Use stdlib unittest instead of lgc.testlib in unittest_request.py

[test] Distinguish "request" tests from accept parser ones in unittest_request.py

[pkg] Add setuptools to RPM build requirements

[pkg] Version 3.24.0

[req,web] Make it possible to handle page language from URL prefix Adding a short language prefix to URL (like "/en" or "/fr") changes the language the pages are displayed in. This prefix is kept during navigation. This way it is not necessary to do language negotiation, nor to use user preferences to determine which language to apply. This behavior is controlled through a new configuration option "language-mode", which replaces "language-negociation" option and which values may be "http-negotiation", "url-prefix" or "" (to disable language setting and force using "ui.language" property). Migration from previous option is not handled because I could not manage to have it working (users will get prompted with the configuration file diff anyways). Add some tests checking various scenarios. Closes #15743487

[config] fix the load_site_cubicweb() method for to 'new-style' cubes (closes #16059402) We first try to load the site_cubicweb module from the cubicweb_<cube> package, and if it fails, revert back to old cube path.

[massive store] remove on_commit / on_rollback parameters No actual use of this was found in client code. If one wants to control this behaviour, he can always override the commit method and add a try / except around flush_entities in its own store

[massive store] master_init should commit by default Default client usage is to create a master store to control slave ones. They need to access cwmassive_initialized from another transaction

[massive store] isolate SQL statements that one may want to customize i.e. provide public methods for meta data insertions

[massive store] Store entities in temporary table as well * on some entity type is encountered by a slave, create a dedicated table for insertion of entities of this type by this slave, similarly to what is done for relation - this should lower changes of conflicts in master/slaves mode ; * delay drop of constraints and indexes to `finish` method, where copy from temporary tables to regular table is done ; * insertion of metadata is done by scanning temporary tables, which may be way shorter than theier associated regular table ; * drop drop_metadata_constraints with its constraint_dropped friend attribute, there are no more necessary since this is done once in the `finish`. Related to #15538303

[massive store] Drop useless check for empty buffer If something went wrong, an error will be raised.

[massive store] Docstring / comment cleanups

[massive store] Turn a runtime error into an assertion geting there is a bug.

[massive store] Delay metadata attributes and default values handling to flush_entities This may be done once for all there, instead of for each entity in prepare_insert_entity.

[massive store] Use a slave specific table for relation insertion in the massive store * attribute a random id to the store * add a column containing this id in the cwmassive_initialized table * separate generic `master_init` which create the cwmassive_initialized table from rtype specific initialization which must be done in each slave now * delay removal of table indexes to `finish` * get back relations from each table on `finish` Similar work still has to be done for entities insertion - which will allow more cleanup to the constraints handling which is still rough for now. Related to #15538303

[massive store] docstring and __init__ cleanup

[massive store] Drop dead code This function is never used

[massive store] Reintroduce methods that are necessary to properly handle master/slave configuration Related to #15538303

[massive store] Follow configuration of the metadata generator Don't drop constraints and indexes for tables that are ignored by the metadata generator given to the store. One may now easily disable insertion of e.g. created_by / owned_by by removing them from the MetadataGenerator.META_RELATIONS set, in which case indexes for associated table won't be removed by the massive store.

[dataimport] make MetadataGenerator.META_RELATIONS customizable This should be done on the instance rather than on the class

[massive store] Rework constraint/index handling The basic idea is to keep the primary constraint on entities.eid since it's heavily used in metadata insertions. Other option would have been to drop / recreate but its very costly on big database, and the index is used for insertion into the entities table itself, so it's not worth droping it at a first glance. Also, keeping it avoids to systematically drop all constraints which depends on it. We may thus now lazily drop constraints, only on insertion of some etype/rtype for the related table. Related to #15538359

[massive store] Lazy removal of constraints and metadata indexes They should not be removed during store's init, because we may want to query the database with its index between store creation and call to prepare_insert* (e.g. to build the extid2eid map). Along the way: * rename drop_metadata_constraints into drop_metadata_indexes, because that's what it does * rework a bit impacted tests Closes #15538359

[repository] Drop the index on entities.type It should be now only used in maintenance queries in checkintegrity. Those are not worth maintaining it. Closes #15538317

[source] Drop source mapping handling It was only used by the cwxmlparser which has been deleted. This is too complex for litle benefit, we don't want to maintain that in cubicweb.

[repository] Drop the entities.extid column and associated cache This was not necessary anymore with promoted usage of the new data import API. Turn repository's _type_extid_cache to _type_cache with only the entity's type as key. This introduces an backward incompatible change: entity_metas dict doesn't contains anymore the extid key, but it doesn't seem used at all anywhere, so this sounds acceptable. Closes #15538317

[repository] Drop the entities.asource column It not used anymore thanks to previous patches. Closes #15538288

[integrity check] Stop using entities.asource column in integrity checks Simply check if entities have a cw_source, and link it to the system source when not found. Related to #15538288

[rql2sql/test] Ensure test results predictability Without this, test were randomly failing within tox due to different order of solutions.

[rql2sql] Stop generating SQL query from RQL using entities.type This was necessary in the "true multi-sources" time, it's not anymore, while maintaining this index is costly. Related to #15538317