93e30e5e introduce a  new `user` table that is listed as a *search table*, so until this table is created all catalog queries are failing, and alarm tool, which relies on catalog fail with such a traceback:

```
ERROR TimerService Process timer error
Traceback (most recent call last):
  File "parts/erp5/product/TimerService/TimerService.py", line 102, in process_timer
    DateTime(prev_tick), DateTime(next_tick))
  File "parts/erp5/product/ERP5/Tool/AlarmTool.py", line 175, in process_timer
    self.tic()
  File "parts/erp5/product/ERP5/Tool/AlarmTool.py", line 135, in tic
    for alarm in self.getAlarmList(to_active=1):
  File "parts/erp5/product/ERP5/Tool/AlarmTool.py", line 111, in getAlarmList
    alarm_date={'query':now,'range':'ngt'}
  File "parts/erp5/product/ERP5Catalog/CatalogTool.py", line 702, in unrestrictedSearchResults
    return ZCatalog.searchResults(self, **kw)
  File "parts/erp5/product/ZSQLCatalog/ZSQLCatalog.py", line 1091, in searchResults
    return catalog.searchResults(REQUEST, **kw)
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2585, in searchResults
    **kw
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2554, in queryResults
    **kw
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2418, in buildSQLQuery
    ignore_unknown_columns=ignore_unknown_columns,
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2394, in buildEntireQuery
    query=self.buildQuery(kw, ignore_empty_string=ignore_empty_string, ignore_unknown_columns=ignore_unknown_columns),
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2295, in buildQuery
    result = self.buildSingleQuery(key, value)
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2087, in buildSingleQuery
    search_key, related_key_definition = self.getColumnSearchKey(key, search_key_name)
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 2049, in getColumnSearchKey
    related_key_definition = self.getRelatedKeyDefinition(key)
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 1999, in getRelatedKeyDefinition
    for entire_definition in self.getSQLCatalogRelatedKeyList([key]):
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 1935, in getSQLCatalogRelatedKeyList
    column_map = self._getSQLCatalogRelatedKeySet()
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 130, in wrapper
    result = transactional_cache[cache_id] = method(wrapped_self)
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 1908, in _getSQLCatalogRelatedKeySet
    column_map = self.getColumnMap()
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 130, in wrapper
    result = transactional_cache[cache_id] = method(wrapped_self)
  File "parts/erp5/product/ZSQLCatalog/SQLCatalog.py", line 1146, in getColumnMap
    for field in table_dict[table]:
KeyError: 'user'
```

This means that even if upgrader's post upgrade constraint are supposed to fix this by calling `portal_catalog.upgradeSchema`, because this constraint rely on alarm tool, it's already too late.

The suggested way to fix this is to also call [`portal_catalog.upgradeSchema`](https://lab.nexedi.com/nexedi/erp5/blob/c99fb9163503c5afdef59ecb124b3060b05330a4/bt5/erp5_upgrader/SkinTemplateItem/portal_skins/erp5_upgrader/TemplateTool_checkTableConsistency.py#L7) in the same transaction that the transaction upgrading business templates.

I have not completely removed `TemplateToolTableConsistencyConstraint`, because maybe there will be cases where just call post upgrade alarm to execute data migration steps after installing a business template manually. But you guys think it's better to completely merge `TemplateToolTableConsistencyConstraint` with `TemplateToolBusinessTemplateInstallationConstraint` I would be OK with that too.

Of course, we cannot just make
`TemplateToolTableConsistencyConstraint` an `upgrade` constraint, because there would be no guarantee that it's called after `TemplateToolBusinessTemplateInstallationConstraint`.

Also, I have not considered making [`getColumnMap` and friends](https://lab.nexedi.com/nexedi/erp5/blob/master/product/ZSQLCatalog/SQLCatalog.py#L1142) resilient to the case where a table listed in search tables does not exist, because it's would just have been failing later anyway and it's better to fail early in such case.

cc: @vpelletier @jm @seb @tiwariayush @gabriel

/reviewed-on nexedi/erp5!247

We had an incident in a in instance were a user changed state of 70K invoices using https://www.erp5.com/howto/erp5-developer-howto/erp5-HowTo.Change.Workflow.State.Of.Multiple.Documents  and leads to user not receiving the reports they requested.

`Folder_modifyWorkflowStatus` creates a lot of `callMethodOnObjectList` activities (for all the selected documents) with priority 2, then these activities will cause more reindex an expand activities.

Until all these `callMethodOnObjectList` are processed, no new activities with priority > 2 were processed. Some "important for users" activities such as erp5_deferred_style reports where waiting in the queue.

I believe we should just set a lower priority to these `callMethodOnObjectList`, eventhough I considered a more clever way of giving the priority:

 - if the number of selected documents is reasonably small, process them with a very high priority, this way the user can see his document changing state almost immediately as when using the synchronous change state. This should not cause congestion because there are not too many documents.
 - when there are a lot of selected documents, process them with a very low priority, because anyway it will take time and user will not "wait" for documents to change state.

... but I realized this is trying to be too clever.

So any objections to just lower priority here ?

/reviewed-on nexedi/erp5!235

As discussed in  !248 this approach allows to set any mail header.

I also included a not so related patch of email header handlings 88d40b40 so that we review all this together.

/cc @gabriel @kazuhiko 

/reviewed-on !256

I just simulated an upgrade for an ERP5 running a version from before !185,  `ERP5UserManager Non Existence_constraint` properly disabled old `acl_users/erp5_users` and migrated all persons, but  `ERP5UserLoginManager Existence_constraint` did not activate `acl_users/erp5_login_users`.

This is because  `ERP5UserLoginManager Existence_constraint` is registered as a **pre-upgrade** constraint, but with my upgrade scenario, pre-upgrade constraint are installed too late. 

The upgrade scenario was:

 1. install new ERP5 SR on slapos
 2. manually install new versions of  erp5_upgrader and customer_configuration_upgrader business template
 3. run upgrade ( sense then fix on portal_alarms/promise_check_upgrade )

During step 3, when pre-upgrade constraints are executed, erp5_base is not installed yet, so  `ERP5UserLoginManager Existence_constraint` can not run because it's not installed yet. 

My suggestion is to change it to a post-upgrade constraint.

If I understand correctly and my upgrade scenario is really the supported one, it means we can only have pre-upgrade constraints in erp5_upgrader.

( I also fix a meaningless typo in the same code)

/cc @kazuhiko  @vpelletier @seb @gabriel 

/reviewed-on nexedi/erp5!262

To be able to use pre_upgrade we would need to install the bt first

Because of ERP5Type.UnrestrictedMethod, 'System Processes' can own objects.
Such objects can be proxy-role'd scripts, and proxy-role mechanism
triggers many users look-ups (each time security is evaluated, which is
virtually every getattr). Each such lookup will do a query for 'System
Processes' user, which will (hopefully) find nothing anyway.
So special-case 'System Processes' when looking by user_id by skipping
the search altogether (enforcing the inability to locate this user,
consistently with Zope assumptions, and consistently with previous
behaviour).

The code was not working as expected when we had several test nodes with same name
(which should not happen). So add a timestamp in the title.

test_result: fixed random issues where test result was staying confirmed while all tests were finished

When two test results line were stopped in parallel, both transactions were thinking that
there was still one ongoing test line, thus none of them were setting the test result in stopped.

Use activities instead of serialization to avoid having testnode getting conflict errors.

This makes a difference for SQLDict as not all messages are accepted for
insertion.

Firefox and Xvfb aren not provided by the tested SR,
but rather by a separate SR that is installed by the testnode.

Charting libraries are better within iframe to avoid enabling too permissive
Content Security Policy globally

Fix error display after save
Fix parameters not recognized
Fix python linting
the same way as ERP5 Script Python

Add PortalType field to distinguish different types of Python Scripts, add referencence and use source code editor

We could not find a portable way to use verifyElementPositionRangeTop.

See internal discussion `[test][erp5] update selenium test` for more
details.

In addition to TypeEror when a 'date' is given, this also accepts duration=0.

If Test Result is associated with Test Node and thus distributor, and
distributor defines process timeout use this timeout to now when to
restart test lines. This is useful to avoid needless restart of tests
running on nodes related to long time outing distributors.

Otherwise fallback to default.

This link allows to navigate from Test Result via its Test Result Node
to Test Node and distributor related to given Test Result.

Note: It happens only, if client will do subscribeNode before calling
createTestResult.

This is accompanying commit to "erp5_test_result: stop affecting last
remaining tests to all test nodes", where alarms redrafts running tests
for long running nodes.

In case of very long running test nodes, to which support is introduced
in this work, it is acceptable situation that test will get redrafted
and shall be stopped.

Unfinished test
(test_18_checkConfigurationGenerationERP5ScalabilityDistributor) uses
return value of subscribeNode and assumes it is ERP5 object.

This work changes return value of subscribeNode to JSON object,
following startTestSuite return value, thus rendering this test failure.

As test is unfinished, non asserting part has been commented out.

Also if process_timeout is available in received configuration setup
process_manager with transmitted value.

Backward compatibility is assured, that if no process_timeout is
provided, nothing is being updated.

Also process_timeout not set on master (the default), will not change
local default.