[migration] reorder system tables / meta-schema migration into bootstrap_migration Also fix a bug in the backwards-compat eid_type_source method where it returned an extid (possibly null) instead of a source name. With these changes, migrating a database from cw 3.12 is now possible. Closes #4962681

[dataimport] add missing import (closes #4985916)

[debian] cubicweb-dev needs cubicweb-etwist Closes #4965410

[pkg] 3.19.11

Binary initializer should accept bytearrays (closes #5084075) Recent versions of pyodbc return bytearray objects for binary data.

[native source] log the used connection in doexecmany as in doexec

[server] use unicode wherever appropriate in transaction/undo related methods Silences lgdb/sqlite warnings about using str instead of unicode.

[web] typo

[entities] return the entity itself from notification_references if the view uses a random message-id Useful for entity update notifications to reference the entity creation message.

[notifications] use non-fixed message-id for EntityUpdatedNotificationView More than one of those can be sent for a given entity, so they shouldn't all have the same message-id. Closes #4962712.

[server/test] kill remaining 3.19 deprecation warnings in undo tests

[devtools] accept str objects in CubicWebTC.new_access and .create_user

[utils] use the "predictable" argument to js_dumps Fixes doctest under python -R. Related to #4959402.

[server/test] don't rely on a sorted output from .object_relations() Fixes test under python -R. Related to #4959402.

[req] make rebuild_url predictable Fixes test with python -R. Related to #4959402.

[web/test] don't depend on iteration order on a set Fixes test under python -R ('pending_insert' in session data is a set, so we can't rely on any particular ordering). Related to #4959402.

[web/test] fix cubicweb 3.19 deprecation warning

[web/test] use unicode instead of str for user logins makes lgdb/sqlite quieter.

[test] Don't depend on dict iteration ordering Fixes test with python -R. Related to #4959402.

[entities] make cw_fti_index_rql_queries predictable Don't depend on the hash function, fixing test with python -R. Related to #4959402.

[hooks/syncschema] use a list instead of a set for UpdateFTIndexOp The order in which etypes get indexed matters. Related to #4959402.

[rqlrewrite] sort possible types when turning is_instance_of into is Helps get predictable output even with python -R. Related to #4959402.

[server/test] use unicode strings for user logins

[web/test] use unicode object as user login (py.warnings) WARNING: /home/jcr/src/cw/cubicweb/devtools/testlib.py:206: DeprecationWarning: Sanitizing an input dictionary with str values, please check your data (key = 'login')

[server/querier] stop passing a rqlst to empty_rset (py.warnings) WARNING: /home/jcr/src/cw/cubicweb/server/querier.py:46: DeprecationWarning: [3.20] rqlst parameter is deprecated return ResultSet([], rql, args, rqlst=rqlst)

[web/views] Use predictable ordering for attributes in entity XML view Fixes test under python -R. Related to #4959402.

[web/views] set explicit order for ManageSourcesAction Gives predictable ordering wrt ManageUsersAction under python -R. Related to #4959402.

[web/test] don't depend on dict iteration ordering Fixes tests under python -R. Related to #4959402.

[web/test] pass login as unicode string silences lgdb/sqlite warning.

[web/test] fix 3.19 deprecation warning

[sobjects/test] Fix dependency on predictable hashing Fix test under python -R by ignoring url query parameter ordering. Related to #4959402.

[test] don't rely on a stable ordering for appobject_path Fixes test under python -R. Related to #4959402.

[server] remove "standalone RQL server" from possible configuration choices Related to #2919297.

Remove leftover bits from start-repository command, and document its demise Related to #2919297

Remove the remote repository-access-through-zmq support Modern methods such as the rqlcontroller cube + the cwclientlib library are the way forward. Closes #2919297.

Remove remote repository-access-through-pyro support Modern methods such as the rqlcontroller cube + the cwclientlib library are the way forward. Closes #2919309.

[web/views/rdf] Take the second element of XY equivalent for non final relation When given a Yams snippet of the form `<etype> <rtype>` ``xy.xeq()`` will return the XML snippet equivalent to this relation in the form of a tuple `(subject, rtype, object)` so ``item`` will always be a tuple here. In fact, the correct code appears just a few lines above (around line 88) for final relations so I guess this is a copy-paste mistake. Closes #4745929.

[server] Make sure 'zmq-address-sub' config is always iterable

[web] split timeline in an individual cube Closes #1279544

[devtools] don't hide warnings and errors during tests The markdown module in jessie turns all warnings into logging calls, so this allows me to still see deprecation warnings when running tests.

[test/i18n] remove unneeded hack

merge 3.19.10 into 3.20

[server/session] Fix typo "meta vs. metas" in describe()

Added tag cubicweb-version-3.19.10, cubicweb-debian-version-3.19.10-1, cubicweb-centos-version-3.19.10-1 for changeset 3bab0b9b0ee7

[pkg] 3.19.10

[i18n] Update translations Related to #5012581.

[config] Lowercase the FQDN we get from the OS (closes #5040345) Browsers convert the host name part of URLs to lowercase. This has the unfortunate effect of triggering various CORS error messages in CubicWeb (because the Origin: header will contain lowercase letters). As per RFC 4343, host name comparison should be case-insensitive. So let's put it in lowercase when we grab it from the host system. If admins put uppercase letters in "host" or "base-url" in their all-in-one.conf, then it's their fault.

[web/views] Honor 'action' attribute on AutomaticEntityForm, closes #4943392 On AutomaticEntityForms, get_action used to bypass the logic from FieldsForm.form_action() which handles the 'action' attribute itself, preventing callers from overriding it at selection time. This capability is used by the inlinedit cube. In a nutshell, we let the deprecated AutomaticEntityForm.get_action() raise an AttributeError, which allows FieldsForm.form_action() to handle it properly.

[docstrings] fix project-wide English language mistake Even though 'u' is a wovel, it may be pronounced 'yu' which is a consonant sound. In such cases precede it with 'a' instead of 'an'.

[doc] Don't pretend we know cubicweb runs on mysql As far as I know it hasn't been tested in years.

[test/i18n] Update reference po file following 75cd7df5c32e This changeset added a (missing) message "add a <etype>", so reference po file needed an update.

[devtools] add a method on CubicWebTC to fake data posted by standard cw forms. Closes #4985805

[web/wdoc] fix links They're relative to ${base-url}/doc/, not ${base-url}. Closes #3161099.

[i18n generation] add missing 'add a EntityType' message to generated catalogs. Closes #5012581

[server/test] fix test_zmq We were only asserting in a thread, thus ignoring the results.

[views,js] remove user_callback feature Related to #3975260.

[newcube] Make the generated code PEP-8 compliant Closes #4925396

Remove unused imports

[ext] Remove use of cStringIO We create a StringIO never to read anything from it. Docutils seems to support setting the 'warning_stream' to False to ignore warnings. So let's just do that.

[view] Use io.BytesIO as stream for binary views

[undo/test] check that the entity was restored even though the fiche relation wasn't

Use a moved_entities table to record external entities moved to the system source Instead of using a negative eid in the entities table, move the record to a new table so we don't have an interval with a missing eid in entities. Related to #4846892

Add unique index on entities.extid

[undo/test] if we try to delete an entity referenced by an inline relation, we fail

[undo] when we can't undo an inlined relation change, point at None, not a broken eid This whole thing is "a bit" weird.

Ask sqlite to check foreign keys Related to #4846892

Add reference from etype table's eid column to the entities table Related to #4846892.

[server] change order of entities table modification vs entity creation/deletion In order to have a foreign key referencing the entities table, the insertion into entities must happen first, and the deletion must happen last. For the deletion case, this means we need to: 1. delete all relations (cascade) 2. delete the entities themselves 3. delete the corresponding lines from the entities table This means the _delete_info{,_multi} methods can't keep doing 1 and 3. Thankfully the "public" delete_info method appears to be unused, so drop it. Related to #4846892.

Use foreign keys for relations tables Closes #4846892

Add foreign key for inline relations Related to #4846892

Use our version of schema2sql Related to #4846892

Steal schema2sql module from yams It has no user inside yams, and we need it here if we want to be able to reference the (cubicweb-only) entities table from other tables. Related to #4846892

[dataimport] fix pylint detected error in SQLGenObjectStore

[restore/portable] properly restore the numrange generator If the dump format version is 1.1, read and restore numranges. Closes #4870338

[backup/portable] properly write the eid numrange generator (the eid is no more a sequence) Bump format version to 1.1. Related to #4870338

Added tag cubicweb-version-3.20.4, cubicweb-debian-version-3.20.4-1, cubicweb-centos-version-3.20.4-1 for changeset 49831fdc84dc

[pkg] 3.20.4

[schema] set permissions that do not allow edition on computed relation. Closes #4903918 Also, drop 'del schema' from test which clutters the code for no benefit.

[computed attribute] ensure attribute's formula apply only to the correct type. Closes #4901163

[schemaserial] fix detection of CWComputedRType table, closes #4901150

[dataimport] source.add_info doesn't take anymore a 'complete' argument Closes #4891550

[dataimport] massive renaming of session to cnx and make NoHookObjectStore working with the new connection handling. Closes #4891547

[web/data] use the right selector for hiding jquery treeview placeholders Closes #4903927

merge 3.19.9 into 3.20 branch

Added tag cubicweb-version-3.19.9, cubicweb-debian-version-3.19.9-1, cubicweb-centos-version-3.19.9-1 for changeset b7c373d74754

[pkg] 3.19.9

[hooks] don't insert use_email relation if the entity on either side is going away Closes #4912946

[undo] fix deletion of cw_source relation Closes #4912882

[hooks] don't insert an owned_by relation for deleted entities If, in the same transaction, a composed entity is created and deleted, we'll have scheduled a SyncOwnersOp which would end up creating an owned_by relation for the deleted entity. Closes #4846883

[etwist/service] adjust for a signature change Closes #4878570.

[rset] fix crash on displaying rset where some symmetric relation is used. Closes #4739253 Avoid a KeyError.

Fix handling of limit within the primary view for related elements. Closes #4723579 * respect limit set by uicfg (like the 'autolimited' view itself does) * set limit in rql query to one more than the specified one so the view knows there are more things to display

[native] Change SQL errors log level. - The SQL error itself is logged as 'info'. - The 'transaction has been rolled back' message is logged as 'debug'. These log entries should not be critical: - If an error occur at the sql level, it will be transmitted as a CW error. The exception is raised and handled in a higher level function, which will decide if it is critical or not. - Admitting that rolling-back automatically on errors is wanted in some cases, it cannot be considered 'critical'. The caller will receive the original exception anyway, it is its job to decide if it is critical or not. Closes #4801120

Added tag cubicweb-version-3.20.3, cubicweb-debian-version-3.20.3-1, cubicweb-centos-version-3.20.3-1 for changeset 7d3a583ed539

[debian] build-depend on markdown for documentation build

[devtools] Make PostgresTestDataBaseHandler multi-use friendly Dont init the pgdb twice on the same datadir - If two test case with the same apphome use a Postgres configuration, the handler is initialised twice, so it has to check if the pgdb directory has already been initialized. - Work with the realpath of the pgdb In some cases, the self.config.apphome will resolve symbolic links, but not always. It can result in an attempt to start twice the pg server for the same directory, in the same test run... resulting in failure. Closes #4875827

[devtools] restore i18n of string removed in 4001cfe2f44d Removing that string from the po files means that after running i18ncube against newer cubicweb, we lose the translation when running on older cubicweb's, which is probably not a good idea.

[pkg] fix setup.py install One day I'll get this thing right. Maybe.

[pkg] 3.20.3

[server] Increase the stacklevel (due to @_open_only decorator)

[pkg] install jquery-treeview

[migration/3.20] sync rtypes before etypes We need to make sure the rtypes are inlined before adding unique indices.

[doc] Fix class name for ComputedRelation; fix code block about computed attributes

[web] Allow propagating 'klass' parameter on tree views recursively Closes #4881302

[utils] Fix define_var in ajax context (closes #4881301) Use namespaced <cubicweb:script> instead of just <script>

[web/data] Never hide JS errors in our Deferred implementation Modern browsers have good debugging tools, let exceptions go all the way up. This only affects exceptions raised in the js code, if there's an issue on the server/python side, they'll keep going through our error callbacks. Closes #4881300

[web/data] Fix userCallbackThenUpdateUI to completely replace old DOM elements Without the 'swap' parameter, userCallbackThenUpdateUI() would replace the _children_ of the component with a new version of the componenent, thus creating nested divs/spans/buttons/etc. Closes #4881299

[web/views] Stop mishandling the fieldset name in the default form renderer While not really documented, it is generally assumed that: * the fieldset is a user-visible string (much like a title), * it will be translated by whichever piece of code uses it. So: * fix AutomaticEntityForm to generate untranslated fieldset names, * fix FormRenderer to properly translate the fieldset name before rendering it, * stop using the field set (which may contain whitespace) as a CSS class name (the feature was unusable, so just drop it), * properly handle the colon as part of the translated fieldset name Closes #4881298

Added tag cubicweb-version-3.20.2, cubicweb-debian-version-3.20.2-1, cubicweb-centos-version-3.20.2-1 for changeset 138464fc1c33

merge 3.19.8 into 3.20 branch

[pkg] fix mistake in c7b218125c25 We ended up with the wrong filenames

Added tag cubicweb-version-3.19.8, cubicweb-debian-version-3.19.8-1, cubicweb-centos-version-3.19.8-1 for changeset efc8645ece43

[server/test] convert new test to 3.19 API

[pkg] 3.19.9

merge 3.18.8 into 3.19 branch

Added tag cubicweb-version-3.18.8, cubicweb-debian-version-3.18.8-1, cubicweb-centos-version-3.18.8-1 for changeset 231094063d62

[pkg] 3.18.8

[security] Test case and fix for an INSERT security hole 7099bbd685aa introduced a untested corner case in which an Entity with no attribute specified could be created whatever the permissions. Report and test case by Christophe de Vienne, fix by Aurelien Campeas. Thanks! Closes #4854359

[pkg] 3.20.2

[debian] cubicweb 3.20 breaks cubicweb-bootstrap << 0.6.6 Closes #4842316