[wsgi] Honor the 'CONTENT_TYPE' wsgi variable See http://legacy.python.org/dev/peps/pep-0333/#environ-variables Closes #4191812

Added tag cubicweb-version-3.17.17, cubicweb-debian-version-3.17.17-1, cubicweb-centos-version-3.17.17-1 for changeset 57e9d1c70512

[pkg] prepare 3.17.17

[web/headers] don't list a request header twice in "Vary"

[test] silence a few more deprecation warnings in unittest_hooks self.session → cnx.

[session] call rollback in Connection.__exit__ If we just free the cnxset and clear the Connection, we're missing the pending operation's rollback_event callbacks, which may be needed for cleanup.

[devtools] pre_setup_database takes a Connection, not a Session

[source/native] cPickle is available in all supported pythons

devtools: deprecate .req_from_url Which uses .request() And instead provide .admin_request_from_url, which is a context manager yielding the request.

[tests] Add a webtest based TestCase class Closes #4002134

[tests] Move 'anonymous_allowed' property from CubicWebServerTC to CubicWebTC The effective set of the property on the config is now done later that before, ie at the end of CubicWebTC.init_config instead of before. It is believed not to have an impact on the test suites that uses `anonymous_allowed`. Related to #4002134

[compat] Remove imports of "any" and "all" from lgc (closes #4306044) They're builtin since python 2.5.

[css] Remove the `div` tag specification of the pendingDelete css rule The HTML tag was changed by e2f96b16c3bd from a div to a span without changing the css rule accordingly. Closes #4285248.

[cwvreg] cleanup the event manager when reloading modules Closes #3848995 The event manager callbacks are not cleaned during reloading. So the callback defined in the reloaded module appears twice (old and new version). This may cause problem when the old version is called.

[test] Add missing attribute 'add' permission in test schema Silences yams warning.

[facet html] Add surrounding div with "facetBody" class to "has_text" facet. This makes its html structure consistent with the other facets and fixes both css and a javascript behaviour (hide the facet body when clicking on its title). Closes #3797501.

[cwctl] don't prompt for cube options in automatic mode Closes #3984223

[serverctl] fix default value for db-init config-level Otherwise cubicweb-ctl db-init --automatic complains that "--automatic and --config-level should not be used together" when they weren't. Closes #3984336

[web] Return useful error messages when exceptions arise in ajax controllers (closes #3932503) The call to super() in RemoteCallException passes 'reason' all the way to python's Exception.__init__() which is then stored in Exception.args. This way, CubicWebPublisher.ajax_error_handler() gets a useful error message when calling unicode(exc). The change to uilib is just to prepend the exception type name. Just a small improvement.

[test] Fix test breakage uncovered by previous changeset Now that Binary.__eq__ doesn't always return true, this test started failing. CWAttribute.defaultval is zpickled, and the description_format attribute is a String, meaning unicode, so adjust the expected test result.

[etwist] use more specific REQUEST_ENTITY_TOO_LARGE instead of BAD_REQUEST When a request body exceeds the configured limit, return 413 to the client instead of a generic 400.

[entities, view] delete dead code The auto_unwrap_bw_compat metaclass calls unwrap_adapter_compat, which was removed in changeset 697a8181ba30 "remove 3.9 bw compat". So this can't possibly work anymore, meaning we can get rid of it.

[book] stop talking about the hg `forest` extension

[book] update sections about dependencies This is probably incomplete, but it's a start.

[web/component] move unicode() call around Makes Aurélien happier.

[web/js] replace callAddOrDeleteThenReload with more generic callAjaxFuncThenReload That's going to be a more useful replacement for callUserCallbackThenReload. Closes #4178566.

[hooks/syncsession] try to remove cnx vs session confusion

Fix Binary.__eq__ (closes #4172701) Due to a typo we always returned true, which was unhelpful.