The logic of the patch accidentally changed in 240a8d26 (
{,Propertied}User: Reduce the overhead from Developer role processing.,
2019-02-14). This used to be a patch to not consider 'Developer' as
a local role in SimpleUser.getRolesInContext , but it was changed to
not consider 'Developer' role at all - also as a global role.

SimpleUser are users from the the root acl_users, users from
erp5/acl_users will typically be PropertiedUsers, because this is a PAS
user folder.

Developer users are exceptional, we should first check if normal user
is allowed before considering Developer role.

zopefoundation/Products.CMFCore commit 882a697 (- correct check for
request compatibility for CookieCrumbler, 2017-05-12)

Restricted python 4 supports iterators

This drops some patches that should no longer be needed with zope 4 to
only keep the publisher timeout changes

Using make_query makes sure that the query parameters are properly
encoded and also change selection_index to become an int, because
it keeps the type of parameters.

As a consequence, we had to adjust a few place in the code where
selection_index was tested for truthiness: because "0" as a string is
true, but 0 as an int is not. For that, we changed to test the presence
of selection_name instead of testing selection_index, as they are always
used together.

This fixes a problem that & in URL was encoded twice for listbox
anchor links ( bug_module/1137 )

02011d8e (immediateReindexObject: use super user to reindex script,
2015-12-14) did not apply for inventory, because they were overloading
immediateReindexObject.
Introduce a new level, _immediateReindexObject that will hold the actual
reindexing logic.

Previously this method was using PortalContent.reindexObject which was
monkey patched, to make things less complex and more future proof, move
the monkey patch to a method on base class.

This also drops alternateReindexObject on BalanceTransaction, because it
is already defined in Inventory

This is supposed to fix "Should not load state for ${oid of a skin} when
the connection is closed" sometimes happening with live tests (especially
when the test self.publish and the developer access the site while the
test is suspended on a debugger breakpoint).

The object accessed after the connection is closed was a skin (python
script, sometimes page template or form) that was cached in SKINDATA.
The mechanism to prune entries from the cache uses REQUEST.hold API
which expects that REQUEST.close is called on request, but because
requests were not closed at the end of the request, it happened that
the cache was reused from another connection.

This change to close the requests, like ERP5TypeTestCase is doing in
tearDown (the actual close is done by Testing.ZopeTestCase.connections).

By closing requests at the end of tests, we also have to change so that
at the beginning of the test we initialize the request, by using the same
setSite and setupCurrentSkin that are done in ERP5TypeTestCase.

SKINDATA is an implementation detail that should not be accessed from
this level.

In SkinTemplateItem.install, self._objects contains entries for skin
folders and for all skins. objectValues method calls was called for skin
folders (as expected) and also for all skins, which acquire objectValues
from skin folder and do the work again.

This simplifies this by only running this for skin folders.

"default" URL is the URL when listbox column does not use URL column
and when the brains do not have a getListItemUrl method.

Move the computation in a lazyMethod, so that it is computed only once
per line instead of once per cell.

Also remove a try/except, I don't think this code is supposed to get an
AttributeError

Do not call container.reindexObject every time, otherwise the number
of activities explode because of recurseCallMethod.

See merge request nexedi/erp5!1643

of Organisation_viewFinancialInformationList

In xhtml_style, just after clicking next or previous page in the listbox
buttons, the values selected in 'Section Category' and 'Section Category
Strict' were not used, because they were used from request and not from
as normal script parameters.

This fixes only the xhtml_style version, ERP5JS has another problem that
dialog fields values are reset when going to next page, so it would also
need this problem to be fixed.

Zope4 uses urlparse + urlunparse to encode the URL, which has a side
effect of removing the empty fragment from the URL. Using a lower level
API to set the status code and the Location header we achieve the same
result.

On Zope2, there was support for starting a ZServer (in
Testing.ZopeTestCase.utils.startZServer) and there was another makerequest
(in Testing.ZopeTestCase.utils.makerequest) which had knowledge of this
ZServer address. We are using our own implementation of startZServer,
which also knows how to start a WSGI server, but we we were interacting
with Testing.ZopeTestCase.utils so that its makerequest knows about
our web server.

On Zope4 there is no utility to start an http server from the tests and
the makerequest is also gone.

This revisits the web server from test interaction with makerequest:
 - no longer use PortalTestCase._app which creates an app with a request
   to http://nohost, but implement our own _app which creates a request
   to our web server
  - store our server address as class attributes on ProcessingNodeTestCase
   instead of patching pseudo constants in utils (these pseudo constants
   no longer exist)
  - in the case of ERP5TypeLiveTestCase, also rewrap the site to use a
  request to our web server (which was done on Zope2 by using
  Testing.ZopeTestCase.utils.makerequest and not
  Testing.makerequest.makerequest directly)

This is supposed to fix "Should not load state for ${oid of a skin} when
the connection is closed" sometimes happening with live tests (especially
when the test self.publish and the developer access the site while the
test is suspended on a debugger breakpoint).

The object accessed after the connection is closed was a skin (python
script, sometimes page template or form) that was cached in SKINDATA.
The mechanism to prune entries from the cache uses REQUEST.hold API
which expects that REQUEST.close is called on request, but because
requests were not closed at the end of the request, it happened that
the cache was reused from another connection.

This change to close the requests, like ERP5TypeTestCase is doing in
tearDown (the actual close is done by Testing.ZopeTestCase.connections).

By closing requests at the end of tests, we also have to change so that
at the beginning of the test we initialize the request, by using the same
setSite and setupCurrentSkin that are done in ERP5TypeTestCase.

This reverts commit 3db20ae6
and commit 948d8b57.

This commit and the fixup were not good, reverting to make a proper
commit