To make it easier for users to change their preferences, so that they do not have to create the preference themselves, we tried to pre-create a user preference ready to be configured for each user.  

It was 59860df3 :  an interaction to create a user preference on `Person.setReference` which is more or less the time when this person become a user ( but not really - this was already a weakness of this approach).

This calls `Person_createUserPreference` that initialize the preference by introspecting the assignments of the person. This already had a problem that it was working only if the assignments were created before the reference was set on the person.

With the new user management introduced in !185 this interaction moved to `Person.setUserId`, which is called in Person's init script. This had the following problems:
 - All persons have a user id, so all persons have a preference. For sites with many persons that are not actually users, this create useless preferences.
 - During init, person does not have assignments yet, so `Person_createUserPreference` could not use information from assignment to create preference.

The suggested change is to create the preference only when the user click on *Edit my preferences* button.

This is done by adding a new `portal_preferences.getActiveUserPreference` method that returns the active user preference and create it if not already existing, this way we do not have to put logic in the user interface scripts.

All *Edit my preferences* links should use it like it was done in f62e6651

The `person_interaction_workflow` was completely remove, as the other interaction it was containing - clearCache when deleting the person - was useless . We had to adjust a few tests that was passing thanks to this interaction.

/cc @gabriel @vpelletier @kazuhiko @tc 

/reviewed-on nexedi/erp5!273

Base.edit has this feature of not actually modifying the properties when
the new property value is same as the current one, so when we do
`movement.edit(price=x)`, this will cause an implicit getPrice.

As price lookup is a bit slow, do not lookup price in this case.

/cc @vpelletier @kazuhiko @romain 

/reviewed-on nexedi/erp5!259

Fix for [#20160719-1B69F57]( https://nexedi.erp5.net/bug_module/20160719-1B69F57 ) 

These after clone methods in transformed resources are being too clever and behave very badly when the transformation itself is cloned.

The suggested approach is to stop initializing int index & reference when cloning transformed resources ("transformation lines"). This behavior is kept when adding new transformed resource. This become consistent with what we have, for example, with Sale Order Lines in Sale Orders. 

/cc @luke @seb @gabriel @kazuhiko @romain @Nicolas 

( see also !148 for more )


/reviewed-on !258

Abort transaction, we do not need changes made by test to persist.

Execute pending activities before removing persons, the pending
activities might be relying on the existence of persons.

This changes API on a script that some projects have customized, but
getActiveUserPreferecnce "should work" even with an old script not
returning preferences.

In that case, first call to getActiveUserPreference will create
preference but not return it, but next time getActiveUserPreference is
called, the preference will probably be indexed and will be returned.

This method creates a user preference if no preference exists.

We do not need to pre-create user preferences, especially that creating
them too early creates one preferences for each person created in person
module and does not allow Person_createUserPreference to create
preference based on person's assignments are they are not created yet.

We do not need to clear cache when deleting user logins anymore. This
problem was fixed differently.

Adjust tests that was relying on cache being cleared when persons ar
deleted:

testERP5Web.TestERP5Web.test_15_Check_LastModified_Header was never
isolated from test_14.

test_14 was filling Base_getWebDocumentDrivenModificationDate cache and
this cache got clear when the persons where being deleted during
tearDown. When removing this interaction clearing cache, we revealed
this weakness. Choosen solution was to clear cache before checking
response headers, to make sure we don't get something filled by a
previous test.

Tests should not need to reindex modules

In the same time, stop showing range of cores like "9 to 15" cores. The infrastructure should be
important enough to have all the time the maximum of cores in the range. So just use the maximum
in the range, "15 cores" here.

Make sure we find section if we render https://my_site/my_section/foo.html. This is required
to take into account configuration of the web section, like specific content security policy.

Should make test output more expressive on failure.

To make it more likely detect killed transactions before entering
tpc_finish phase.

Saves one round-trip to server.

This parameter does not force a reconnection, it allows reconnecting if
a closed-connection is detected when querying on a transactional
connection on which the transaction was already started.

Otherwise, all subsequent tests fail too.

This reverts commit 699759bc

Do not give conflicting instructions to portal_catalog.

Preserve compatibility with in-zodb worklist cache code, but do not use
*_expression when calling catalog.

These are a bit more complex, there are more to come.