With distributed caches, portal_caches.clearAllCache() does not clear cache, so
using ERP5Site_cleanupSupportRequestUITestDataSet is not enough to flush the
"last comment" from portal sessions, so we need to delete explicitly the
session - and also commit the transaction because distributed caches are also
transactional.

when using a RAM cache for session, we can modify the temp objects directly
and the changes are kept, because the same object kept in RAM is accessed by
subsequent requests, but when using memcached every request get a different
object, so we need to put back the object in the session so that the new
version is saved in cache, so that subsequent requests use the modified object

When sessions uses a RAMCache, the portal_caches.clearAllCache is enough to
reset the session, but not when using a distributed cache.

It was using the default UserDict.update implementation which did not remove
acquisition wrapper and eventually lead to ConnectionStateError like for example

> Shouldn't load state for erp5.portal_type.Interaction Workflow Interaction 0x6637 when the connection is closed

which was observed when deleting sale order lines (wich triggers some interaction)

Previously `Ram Cache` was used, causing anonymous clients to never use portal_session properly.
Mainly for public forum captcha purpose.
Now `Distributed Ram Cache` is used to share the session between the zope nodes.
It's required to restart all zope nodes after this BT5 upgrade.

This was only supported for the RAMCache version of session tool, but is also
good to support them in DistributedRamCache, so that we can use DistributedRamCache
by default.

One breaking change is that when using RAMCache, changes made to temp objects
in session were persisting for the session lifetime, after this change, each
time we get the object from the session we get a new fresh object, so if we
make change we have to explicitly put it back to the session, which is surprising
for developers used to the convenience of ZODB transparent persistence.

Stop using Sequence, use more high level assertion methods instead of
self.assert_, define the shared tests in a test case and make tests for each
cache plugins inherit from the test case.

Also add a few tests:
  - storing temp objects in Distributed cache
  - storing "recursive" temp objects (like a sale order with sale order lines)

What should not be allowed is to store persistent objects in ZODB, but pickling
them should be OK. This is necessary if we want to store them in memcached.

This also modifies loadTempPortalTypeClass which was returning a class with a
different name than requested, by prepending "Temporary ". This prevented
pickling, because pickle checks that the class is can be imported under its
same name. This was problematic because before this change, we had such a behavior:

    (Pdb) tmp = self.portal.person_module.newContent(portal_type='Person', temp_object=True)
    (Pdb) tmp.__class__
    <class 'erp5.temp_portal_type.Temporary Person'>
    (Pdb) import erp5.temp_portal_type
    (Pdb) getattr(erp5.temp_portal_type, 'Temporary Person')
    <class 'erp5.temp_portal_type.Temporary Temporary Person'>

My understanding is that this prefix in the class name was not useful, that
class lives in another module than the "real" Person class.

 * expose firefox console and error logs in `geckodriver.log`
 * expose `geckodriver.log` in the folder testnodes expose on http
 * fix some cases where test were detected as "did not run" and cases where "did not run" tests were not reported as failures

See merge request nexedi/erp5!1444

It can happen that a test running selenium fail in python, but not in selenium,
like for example test_result_module/20210615-CDADEC14/183
To prevent such tests from being reported as PASS, we make the total number of
failures being the sum of the python unittest failures and the selenium failures.

It happens, for example with erp5_officejs_ui_test:testFunctionalOfficeJSPyodideNotebook
on test_result_module/20210615-CDADEC14/183 that the test is considered as
not running, because the test result table was not populated in time.

Add an explicit wait to prevent such problems.

This fixes commit ea4cb55b.

Our 'rmtree' helper has never been supposed to be called on a
non-existing folder: the ENOENT handling is an unrelated
implementation detail for Python 2.

Also drop erp5.util's rmtree and use the one from slapos.core,
which is a [testnode] dependency.

This front page gadget allows user to:
* create any kind of document
* launch any kind of report
* contribute a new documents

This gadget can be used to replicate the xhtml access page.

Eventhough _getBudgetDict was cached, getAvailableBudgetDict also
calculates getCurrentBudget for each cell, which is also heavy.

Duplicate actions are actions from the same category with the same name.

See merge request nexedi/erp5!1442

With software release test suites, there are probably
files/folders with permission issues.

See commit 380037f8.

so that we can access this log directly via http

This firefox setting cause console.log messages or javascript errors to be
printed on stdout, which should help diagnosing failures sometimes.

See merge request nexedi/erp5!1439

By setting XORG_LOCK_DIR, we can isolate processes when running on slapos test
nodes, so that instead of having all test runners used the shared /tmp, which
eventually fail with

    EnvironmentError: All displays locked : [':123', ':124', ':125', ...

errors when too many testnodes are currently running or too many testnodes have
left some stale lock files. Eventhough since fbdb4ec9 (test: Do not check
presence of lockfile in /tmp/ to assume Xvfb is running for this display, 2017-06-07)
things have improved, it happens that the lock files are owned by a different
unix user (because it was left by another testnode), then Xvfb is not able to
remove the stale lock file.

This is backward compatibility code for previous commit.

actions in erp5_jio_view category does not seem to be directly presented
to user, so rename them so that they don't use the same name "View" for
coding style test_DuplicateActions

Duplicate actions are actions from the same category with the same name.

`inst/test0-0`  ->  `i/0`

See merge request nexedi/erp5!1441

With bf57228a ([erp5_core] RelationField listbox must also fetch
default_params value from the relationfield, 2018-05-09), we introduced a
regression that listboxs from relation fields "proxy listbox ids" default params
where no longer used. As a result, if we have a listbox which define some
catalog search with default parameters, like for example the "Suppliers" or
"Clients" listboxs from erp5_trade are doing, these parameters were not
taken into account. In the case of trade listbox, this was showing all nodes.

This changes Base_getRelatedObjectParameter to get catalog search from the
proxy listbox if there is one.

trade_renderjs_ui_test: tests for "Suppliers", "Clients", "Organisations" and "Persons" Proxy Listbox Ids

This is mostly for editor fields, to fix the problem that components were
edited by the text editor and not the code editor.

Editor is selected by Base_getEditorFieldPreferredTextEditor, which will
use a text editor by default, unless the document has a content type, in
which case it will return the most appropriate editor, which is by default
the source code editor for documents whose text content is not text/html.

It's anyway not wrong to have a content type defined here, Components
are TextDocuments and TextDocuments usually have getContentType method.

See merge request !1432

also enable more plugins:
  - https://ckeditor.com/cke4/addon/autolink
  - https://ckeditor.com/cke4/addon/pastebase64

See merge request nexedi/erp5!1428

Instead of using an always-different logger name which confuses log
analytics system, use the warning category as logger name (or what zLOG.LOG
calls "subsystem")