[css] replace old rounded corners hack by css rules Closes #3381426.

[views/templates] provide a modal main template (closes #3274672) The current NonTemplatableViewTemplate may seem to provide the desired feature if a '__notemplate' is provided in the form, unfortunately, it: * wraps the view in a spurious div * calls view.set_stream(), which has the effect of wiping any js/css resource recorded until then by the view, leading to hard-to-track missing resources symptoms

[js/widgets] fix the InOut widget with modern jQuery versions Several things are done there: * reduction in size and complexity of the code * the unused defaultsettings are removed * the initial `unlinked` list is now correctly populated from python-side * the unit test is adjusted because it tested an irrelevant implementation detail which is no longer true (but the widget of course still handles correctly the linkto information) Tested with ie7, ie9, chromium, firefox. Tested with jQuery 1.6 (cw 3.17.x) and 1.10. Closes #3154531.

[autoform] Allow overridding of permissions checks Closes #3063930

[schema] fix unique together index handling We now provide a more compact indexname, using the schema constraint entity type and the position of the columns set in the entity type unique constraints list. This avoids a nasty name truncation issue. The UniqueTogetherError object is made smarter: it computes the rtypes, abstracting the underlying backend (pg/sqlserver vs sqlite). The `user friendly` adapter is much simplified since there is no longer any truncation issue. Uses a new logilab.database version (ticket #151507) and a new yams version (ticket #189299) Closes #2514939 [jcr: disable hooks when temporarily dropping CWUniqueTogetherConstraint entities]

[pyrorql] deprecate pyrorql source (closes #2919299)

[web/facet] do NOT drop offset/limit of the base rql (closes #3344579)

[webctl] Generate static data directory on upgrade (closes #2167873) - if the folder already exists, ``upgrade`` asks for deletion, - add an option (``generate-staticdir``) to allow skipping this task. - add an option (``staticdir-path``) to specify the static data folder path. The ``gen-static-datadir`` command allows to specify the target folder but there is otherwise no way to retrieve this information during upgrade.

[cwctl] Don't check for old lgc versions Support for check_duplicated_command in CommandLine was added in 2011 for 0.55, and we already depend on 0.59.0.

[migration/3.18] add sanity checks before changing symmetric relations Delete supposedly impossible relations from the db. Related to #3259713.

[migration/3.18] disable all hooks when setting up symmetric relations This operation is semantically a no-op, so it should be invisible for the application. Related to #3259713.

[devtools] bases for instrumenting / debugging standard propagation hooks * a class that helps writing command to check propagation, detecting standard errors and picture the propagation as a graph * a class to instrument sets used to configure standard propagation hooks closes #3171980

instrument cwvreg, so we may know what's being loaded by asking the registry store

set proper PACKAGE information for yams 0.39 Yams 0.39 adds an unobstrusive API to get the package, or cube, defining a part of the schema (entity type, relation type/def). As CubicWeb overrides part of the schema reading process, we have to handle this information on our side so it will be properly set.

[yams] follow yams 0.39 api change introduced by ad95fd2c46de (package attribute on etype/rtype/rdef)

Make the GROUP_CONCAT aggregate function not repeat values (closes #3223975) Work on sets instead of arrays, so if the same value appears twice it's not repeated in the concatenated output. This patch handles the postgresql and sqlite backends, mysql is left alone at this point (seems doable, but I don't have time or motivation to fix and test it).

[mail] allow to specify SMTP's MAIL FROM address to config.sendmails(). Closes #3373620 Also modify testlib.Email so we don't loose this information and may test it.

[mail] format_mail copy/paste fix: we want to test sender-name, not sender-addr

symmetric relations: replace bogus rql2sql translation by a hook The hook ensures X r Y => Y r X iff r is symmetric. The rql-no-hook data importer receives a small amendment but note that: * there exist no test for it * its actual semantics are undefined Hence we cannot prove this hunk breaks nothing, because we cannot prove anything. Closes #3259713.

[migration] fix handling of default value for boolean attributes We can't assert that the old value is 'True' or 'False', because False used to be stored as an empty string in pre-3.18 versions.

[entity/url publishing] fetch_rqlst should use 'is_instance_of' instead of 'is' We usually want to get any subclasses as well (think Training subclass of Workcase). Closes #3210365.

[rewriter] fix latent bug: arbitrary etype may be substituted when using is_instance_of type restriction Note: the test is not deterministic because the bug depends on the iteration order on a set.

[test] drop no more used test data file

[doc] Refer to RsetTableView in EntityTableView docstring instead of TableView

Fix use of vreg.config.anonymous_user() If no anonymous user is declared, anonymous_user() returns 'None, None', not 'None'.

[server] avoid race-condition in timer events With a granularity of one second, there's a good chance calling localtime twice will return the same value. We should consider the event ready immediately if that happens, not wait one second.

[server/pyro] try to shutdown the repository properly If RepositoryServer.trigger_events is not called we might miss a QuitEvent and thus never shutdown the repository and its looping tasks.

[book] remove broken links to web.views.wdoc.ChangeLog{View,Action} ChangeLogView and ChangeLogAction were removed in 3.16.0 (changeset fa044b9157d7, https://www.cubicweb.org/2423532) Closes #3369687

[book] fix Session.{set,free}_cnxset autodoc They're Session methods, not bare functions. Without this building the docs says: reading sources... [ 39%] devrepo/repo/sessions Traceback (most recent call last): File "/usr/lib/pymodules/python2.7/sphinx/ext/autodoc.py", line 326, in import_object obj = self.get_attr(obj, part) File "/usr/lib/pymodules/python2.7/sphinx/ext/autodoc.py", line 232, in get_attr return safe_getattr(obj, name, *defargs) File "/usr/lib/pymodules/python2.7/sphinx/util/inspect.py", line 70, in safe_getattr raise AttributeError(name) AttributeError: set_cnxset Traceback (most recent call last): File "/usr/lib/pymodules/python2.7/sphinx/ext/autodoc.py", line 326, in import_object obj = self.get_attr(obj, part) File "/usr/lib/pymodules/python2.7/sphinx/ext/autodoc.py", line 232, in get_attr return safe_getattr(obj, name, *defargs) File "/usr/lib/pymodules/python2.7/sphinx/util/inspect.py", line 70, in safe_getattr raise AttributeError(name) AttributeError: free_cnxset

cwetype is a *class* property, rename its argument accordingly

typo

[req] New method: RequestSessionBase.find(). This method does what find_entities and find_one_entity did, except it returns the resultset itself. In addition, it accepts 'reverse_' arguments and check that the relations actually exists on the entity before executing the query. Also, reimplement find_one_entity and find_entity based on the new function so they benefit from the more complete implementation, and deprecate them. Note: List of values in kwargs are NOT supported in this initial implementation. Closes #3361290

[rset] New method: ResultSet.one() This method will return exactly one entity while enforcing its existence and unicity. The idea is shamelessly borowed from SQLAlchemy Query.one(). Closes #3352314 [jcr: use len(self) instead of len(self.rows)]

Added tag cubicweb-debian-version-3.17.11-2 for changeset b02e2912cad5

[debian] add lintian override for a false positive

[predicates] allow multiple transition names in on_fire_transition (closes #3013720) Also add a warning about the unimplemented from_state_name argument.

[css] add the entypo font as available ressource (closes #2930356) This provides a standard set of pictograms for use in CubicWeb and cubes. To use a pictogram, the "cubicweb.pictograms.css" must be explicitly loaded (using, e.g. view.add_css(...)).

[skeleton i18n] Add PO-Revision-Date header to make emacs po-mode happy Closes #3276389

[uicfg] give a set_fields_order method to the primary view display control Hence it is nicely symmetrical with the fields kwargs. Closes #2741963.

merge 3.17.11

Added tag cubicweb-centos-version-3.17.11-1, cubicweb-version-3.17.11, cubicweb-debian-version-3.17.11-1 for changeset 7f67db7c848e

prepare 3.17.11

[devtools] write db config through a temporary file Avoids other test processes reading partially written config and exploding in pickle.loads. There are still concurrency issues, presumably with the database itself. Something for another friday.

[server] Add missing indices for undo support (closes #3259691) The tables used by the 'undo' feature were missing some indices, which among other things triggered repeated sequential scans when deleting old transactions.

[sqlutils] fix sqlite group_concat harder (related to #3331906)

[sqlutils] avoid a crash with sqlite when using group_concat (closes #3331906)

[server/initrepo] show failed sql statements and abort if necessary (closes #3308564)

[views/table] use cubicweb.ajax.js for pagination Else we fail in js-land on a "AJAX_PREFIX_URL is not defined". Closes #3329072.

[rset] Fix regression with rset copying (closes #3344410) Don't override passed-in descriptions in ResultSet.__init__, even if they're empty.

[doc] Fix a sample request

[tableview] Don't share column renderers between all instances of a table (closes #3351872)

[tableview] Add some unique element to AbstractColumnRenderer.__str__ (related to #3351872) Otherwise different instances look the same which is confusing.

add failing test case related to #3013554

[web] make sign_text unicode aware, avoid crash with non-ascii chars. Closes #3289774

[doc] Use string debug mode in debugged docstring

[rql2sql] fix relation table scope when some variable from subquery is used. Closes #3252270 While it feels weird to drop this explicit ColumnAlias handling, hg history has no clue on its introduction (prior to rev 0) and no older test case fails without it, so let's drop it.

[cwctl] backout 5d5b3a865eb1, which breaks interactive shell usage

[ReST] Implement a rql-table reST directive. Closes #3252856 allowing to call table or derivated view specify headers / cellvids. Also, rql may be split accross several lines which greatly improve readability.

[server] drop support for antediluvian lgdb versions That API was changed in 2010, before logilab-database was moved out of logilab-common.

[doc/book/security] update description of entity update (Related to #2932033)

[test] test can_use_rest_path directly No need to go through the Entity.rest_path if what we want is to unittest that one function.

[ctl] set proper encoding for sys.stdout/sys.stderr. Closes #1669144

[web] typo in config file comment

[book] Typo and formatting fixes in devrepo/entityclasses/application-logic

[cwctl] Only start the instance after upgrade if it was running before Closes #3207676.

[book] Improve match_rtype selector docstring and "Relation modification related events" section

[book] properly escape backslash

[test] use assertIn / assertIsInstance

update pylint extension to astroid API

Fix example in cw_set method

[merge] incorporate cw 3.17.10 fixes

Added tag cubicweb-centos-version-3.17.10-1, cubicweb-version-3.17.10, cubicweb-debian-version-3.17.10-1 for changeset fe0e1863a137

[pkg] prepare 3.17.10

[web] don't rename resources in gen-static-datadir We should use the result from WebConfiguration.locate_resource to know which file to copy, not what name to copy it to. The destination file should be the name of the resource, as the client will request it. This matters in the case of cubicweb.css which locate_resource might rewrite to cubicweb.old.css. Closes #3206129

[notification] avoid leaking cnxsets (closes #3243810) When sending notifications, we get each recipient as either an email address or a CWUser. In the latter case, we create a temporary session for that user and use it to send the mail. However, if we later decided to not send the mail after all, we'd leak the session and its cnxset. Add a try block inside the loop to make sure the temporary sessions are closed properly.

[book] English fix

[serverctl] complete the schema-diff command with options that will fix & enhance it. Notably: - filter out eids (doesn't make sense for the filesystem schema) - handle permissions, trying to detect the default ones Needs yams #174042. Closes #2996547.

[server] deprecate the old multi-source system It is removed in 3.19. Closes #2919299

[doc] document changed attribute default value storage

defaultval migration for sqlite

[schema] store default attribute values in a Bytes field, allowing python objects as default values Notes: * Binary objects grow two methods * we depend on a newer yams version * the code is quite simplified as a result of the whole change * a very old and nasty bug where bool(default values) evaluating to False not being properly migrated is killed Closes #2414591.

[source/native] fix sqlserver pattern matcher for unique constraints (closes #3227593) The matcher is used to parse the low-level error message and extract meaningful information for the end users. If broken, the users see an alarming and bewildering message. A full & definitive fix is scheduled for CubicWeb 3.18.

[dbapi] Fix broken inmemory:// URLs parsing (closes #3225832) ParseResult.path is what comes after the "domain name" in standard URLs. So in the case of 'inmemory://<instance' URLs, it is always empty. ParseResult.netloc contains the "domain name" part that we really want.

[metadata hook] rename an operation for the sake of readability

[book] Typo fix and rephrasing in devrepo/migration

[cwshell] rename history file to .cwshell_history (erudi is dead, long live cw)

[entity] Entity.related(): add a targettypes argument (closes #2957313) This allows to restrict the entity types returned by the method (passed to cw_related_rql). When such an argument is given, no caching happens.

[source configuration] silent lgc 0.60 deprecation warning

[web] remove action "Got Rhythm" (closes #3093362)

[debian] Break old geocoding It wants cubicweb.interfaces.

[wsgi] remove ages old hack to set base_url Its meaning has been lost in the mist of time.

[wsgi] also provide an example using werkzeug (if available) Related to #3005509.

[wsgi] add the simplest possible wsgi (debug) server This server is able to: * serve on a given port using the stdlib SimpleHTTPServer * run looping tasks Closes #3005509.

[querier/security] instrument a bit the querier read security checks Related to #2920304

[book] Improve PostgreSQL configuration section * Do not advice for specific PostgreSQL version (point to meta-packages if available in Linux distros); * First mention packages from Linux distros, put the link to official postgres documentation at the end; Closes #2725302.

[js] add a selector string escaping function will help use id strings in jquery selector expressions, e.g: "foo.[subject]:42" -> "foo\.\[subject\]\:42" Related to #3154531.

[merge] backport stable fixes

[test/ldapsource] backout 192a748550c7 which broke RQL2LDAPFilterTC

[rset] make sure rset.description is always a list It's more consistent, and avoids pylint warning "Instance of 'tuple' has no 'append' member"

[schema] edit syntax tree instead of playing with strings for RQLExpressions Lets us support more complex expressions involving e.g. "HAVING" clauses. Closes #3202855

[py3k] note about future warning on __eq__ vs __hash__ Closes #3179561.

[formfields] py3k kill future compat warning Related to #3179561.

[entity] do not raise an AssertionError if the kwargs given to set_relations/attributes is empty Regression was introduced in 3.16 with the new cw_set API. Old set_relations and set_attributes methods did not crash on empty kwargs. Closes #3104019.

[packaging] setup.py: don't exclude skeleton/debian/* from being installed Closes #3192725.

[tests/hooks/integrity] kill old and empty skipped test

[server/session] demote log to debug It can pollute the log very fast. Closes #3203391.

[test] use assertCountEqual instead of assertItemsEqual

[server] fix a number of typos, mostly in docstrings

[skeleton] Install cubes README in Debian packages (Closes #3171971)

[debian] Build-Depends on python-unittest2 | python (>= 2.7) Closes #3175895.

[doc] typos and reformulation

[doc] additional documentation about pip install under both Windows and Debian

spelling: rollbacked -> rolled back

Added tag cubicweb-version-3.17.9, cubicweb-debian-version-3.17.9-1 for changeset 5668d210e49c

merge two default heads

[pkg] prepare 3.17.9

Use the list of sources instead of an iterator in update-feeds looping task This prevents RuntimeError due to dictionary size change that may occur (as a result of a new source being added) during iteration. Closes #3155843.

[security] fix dumb attribute error when inserting read security. Closes #3196891 This has gone undetected because of erroneous tests...

[entity] give a default reasonnable __eq__ and __hash__ to Entity Using the eid. Closes #3179560.

[debian] add some Breaks for backwards compatibility removal The person and comment cubes used to import the cubicweb.mixins module. The inlinedit cube relied on String.startsWith in javascript.

[js] remove 3.9 bw compat (where apparently unused) - cubicweb.ajax.js - loadxhtml: form.callback support removal - removal of top-level functions: preprocessAjaxLoad, reloadBox, replacePageChunk, loadxhtml - cubicweb.compat.js: - map is undeprecated (jquery.map being not an acceptable replacement) - removal of noop, contains, findValue, filter, addElementClass, removeElementClass, hasElementClass, KEYS mapping - htmlhelpers.js: use non-deprecated functions cw.urlEncode - cubicweb.js: - removal of startsWith and endsWith monkeypatches - note deprecated but still used stuff (for action) - test_utils.js: use cw.utils.sliceList instead of global function Closes #2782004.

Rename cleanup_interface_sobjects into cleanup_unused_appobjects Interfaces are gone in 3.18 Related to #2782004.

remove cw 3.9 bw compat (bw compat other than the interface -> adapter changes) - cwconfig, doc/admin/setup: docstring adjustment wrt 3.9 & virtualenv - testing/rst: windmill (which is gone from the testing infrastructure) - other docs: "stuff introduced in 3.9" cleanup, as we don't care about the version old features were introduced - server/hooky.py: bw compat for propagation hooks - server/schemaserial.py: pre CWUniqueTogetherConstraint migration support - web.__init__.py: dumps bw import - autoform.py: .action ppty is gone, also deprecate set_action and get_action - baseviews: CSVView.subvid gone long ago, secondary view removal - primary.py: _render_attribute & _render_relation signature change - reledit.py: rvid/default_value bw compat, should_edit_* dropped - webconfig.py: resourcefile is gone - formfields.py: old vocabulary handling - request.py: build_ajax_replace_url, external_resource Related to #2782004.

remove 3.9 bw compat In cw 3.9, interfaces are deprecated and replaced with adapters, yielding a lot of bw compat in many places -- most if this patch is concerned with the interface bw compat - cwvreg: interface cleanup - doc/adapters.rst: interface cleanup - entities/adapters.py, wfobjs.py: interfaces bw compat - entity.py: interfaces bw compat, also get_value, delete, attr_metadata, has_perm, set_related_cache, clear_related_cache, clear_related_cache, related_rql - predicates.py: score_interfaces & implements - interfaces.py & mixins.py: 100% gone - view.py: implement_adapter_compat, unwrap_adapter_compat - calendar.py, editcontroller.py, ibreadcrumbs.py, navigation.py, xmlrss.py: interface bw compat - treeview.py: salvage one function from mixins.py Related to #2782004.

remove 3.8 bw compat - cwconfig: old keys such as depends_cubes and non-dict dependencies - cwctl: fall back to docstring if no short descr, forget 'short_descr' - dbapi: session_data, get_session_data, set_session_data, del_session_data - dbapi & testlib, session: eid_key is gone - migractions: cachekey is gone - doc/datamodel/definition, schemaserial: BoundConstraint -> BoundaryConstraint - formwidgets: separator - primary view: render_entity_metadata Related to #2782004.

[hooks/security] Defer entity permission checks to an Operation. Some of these checks may currently happen twice within the same transaction and be costly. This should be semantically safe. If people rely on some internal transaction ordering to be allowed early (thus pass) while the condition wouldn't be met at precommit time, their application is broken. It however seems unlikely to happen in the real life (tm). Closes #2932033

[schema] drop very old bw compat (pre 3.5.10) Closes #2925085.

[devtools,etwist] rename TwistedConfiguration to WebConfigurationBase (follows #2919310)

remove "twisted" configuration (closes #2919310) This was also known as "web only" instances. Not used in production anywhere today.

merge 3.17.8 into default

Added tag cubicweb-centos-version-3.17.8-1, cubicweb-version-3.17.8, cubicweb-debian-version-3.17.8-1 for changeset 909eb8b584c4

[pkg] prepare 3.17.8

[date picker] revert #ec65ca70aac9 which breaks the date picker (closes #3182844)

[merge] backport 3.17 fixes into the future 3.18

Added tag cubicweb-centos-version-3.17.7-1, cubicweb-version-3.17.7, cubicweb-debian-version-3.17.7-1 for changeset 483181543899

[pkg] prepare 3.17.7

i18n update

fix typos in workflow constraint error messages

[migractions] rschema.final.inlined -> rschema.inlined Closes #3153086.

[server] Make internal sessions not reset 'safe'-ness on first commit Increment the ctx_count when disabling integrity hooks so the next commit/rollback doesn't reset our transaction and magically turn us into a safe session. Closes #3168027

[facets] Correctly replace old 'edit box' HTML on facet-induced page refresh (closes #3161100)

[debian] replace find | xargs rm with find -delete in cube skeleton. Closes #3161797

[debian] don't require (fake)root to run the clean target. Closes #3161797 It works just fine as a normal user.

[debian] Add ${misc:Depends} to cube packaging skeleton. Closes #3161797 Best current practice is to have that in Depends in all packages using debhelper.

[devtools] fix race when creating backupdir If somebody else creates backupdir between our stat() and mkdir() (like, say, another test running in parallel), we shouldn't crash.

[staticcontrollers] Raise Forbidden, not Unauthorized Unauthorized means "log in to get access", as it results in a HTTP 401. Here, the error is pretty much permanent, and returning 401 instead of 403 confuses things terribly. (This seems to be a pretty widespread confusion :/)

[web] allow /data/ url again (closes #2464798) This feature was broken by 65fecbeb9c3a (which fixed another one itself).

[utils] fix typos in make_uid docstring

[c-c i18ncubicweb] fix crash, closes #3096647

[rql2sql] fix bad sql generated when outer joining 'identity' relation and lhs var comes from a subquery. Closes #3099418 The generated SQL was attempting to access table.cw_eid which doesn't exist.

[web] stop using deprecated StatusResponse. Closes #3098215

[deprecation] add (approximate) version number to deprecation message and set proper stacklevel

[schema] fix spurious warning when rqlexpr/constraint mainvars specify a non predefined variable. Closes #3098165

[repository] properly use IUserFriendlyError when UniqueTogetherError is raised during entity update. Closes #3096638

[deprecation] add cw version number to the deprecation message and help user to understand the change

[datafeed] fix crash due to bad http_timeout handling. Closes #3096585

[date picker widgets] properly distinguish DOM id and input name. Closes #3096575

[date picker widgets] use bare jquery expression rather than jqNode

Use return in CnxSetTracker context manager methods Just in case.

[web/views] make PrimaryView.entity_call accept kwargs It's useful when using this view in a different context to have it not crash on random arguments.

[ldapparser] demote some logs from warning to debug (closes #2713671)

[doc] add a what's new file for 3.18

merge stable in default (3.18) branch

Added tag cubicweb-centos-version-3.17.6-1, cubicweb-version-3.17.6, cubicweb-debian-version-3.17.6-1 for changeset 5b9fedf67a29

[pkg] prepare 3.17.6

Do not use cubicweb-card in rest_path tests Card overrides the rest_path method, so it does not make sense to use it in cw tests.

[inlined form field] fix regression introduced in 3.16.4/570208f74a84. Closes #3064653 In the above changeset, we changed ordered entities to be edited according to mandatory inlined constraints, though this leads to unpredictable order in cases where there are no constraint. This leads to case where inlined relation field may think there is no value related to a mandatory relation, while this is simply because there is a value specified but to an entity that will be created later (together with the relation). To fix this, we've to use the fact that the RelationField.process_form_value return None in cases where there is some entity to be created, while an empty set when there are no linked entity. The new no_value() method allows to differentiate between those different notion of 'no-value'.

[testlib] temporary_appobjects should call __registered__ if it exists. Closes #3064075

[entities] properly escape in EmailAddress.printable_value when format is html. Closes #3064025

Added tag cubicweb-version-3.17.5, cubicweb-debian-version-3.17.5-1, cubicweb-centos-version-3.17.5-1 for changeset 15dd5b37998b

[pkg] prepare 3.17.5

[pkg] fix debian packaging (closes #3058542) - remove useless Python 2.4 reference, - simplify dh_install configuration to install files consistently instead of having half in /usr/lib/pythonXX and half in /usr/share/pyshared, - incidentally, this makes the package work on squeeze. Regression from eaa58d1c7d5f

[testing] add missing generate_bigint method to ValueGenerator. Closes #3059327

[rql / querier] fix bad interpretation of some RQL SET query when some neged relation is involved. Closes #3058527.

merge stable in default (3.18) branch

Added tag cubicweb-version-3.17.4, cubicweb-debian-version-3.17.4-1, cubicweb-centos-version-3.17.4-1 for changeset c7ba8e5d2e45

[i18n] cleanup dead code also fix small bug introduced in b0f6e8c14e7f

[i18n test] hack to make i18n tests run using a Python interpreter they used to work fine when executed via pytest only. This should be removed as soon as logilab.common.registry is fixed.

[i18ncube] fix crash due to duplicated messages in generated schema.pot Add a test case and fix regression introduced in e713c47a993d. Relates to #2811282

[i18n test] load_po: fix bug in case of multi-lines msgid and ensure there are no duplicated messages

[i18n test] simplify test cube a bit by only providing a views module, no package

[web] consider inlined relations in has_editable_relation. Closes #3049970 In some cases where the user can only add/edit inlined relations (though actually edit perms of the inlined relation doesn't seem checked), the "modify" action doesn't appear while it should.

[web test] silent 3.13 warning

[editcontrollers] Account for role in the ordering of entities (Complements #3031719) Only role='object' was correctly accounted for by 570208f74a84

[migration.sync_schema_props_perms] ensure all participants to a unique together constraint are there before adding CWUniqueConstraint. Closes #3038345

[session commit] save back exception context to avoid potentiel cluttering if some revert operation raise an exception

[pkg] python-central has been removed from Debian

[pkg] prepare 3.17.4

[pkg] merge centos 3.17.1-2 in stable (prepare 3.17.4)

Added tag cubicweb-centos-version-3.17.1-2 for changeset 965f894b63cb

[rqlrewrite] fix rqlrewrite unpredictability vs relation sharing. Closes #3036144 The relations index used to determine if relation may be shared only considered a single relation node per relation type. So when the same relation type occurs several time, dict order give unpredictable result. We shall properly consider all relations instead. Tentative test case included (but bug reproduction is by definition unpredictable...)

[rqlrewrite] rewrite doesn't need a solutions argument, always use the Select'ones Also, do a copy systematically else it may lead to empty solutions depending on the rewrite path. Test case tentative included in the following changeset.