We now have getRelationCountForDeletion API for this.

The general approach is:
- replace deprecated lambdas with list comprehensions
- remove unused variables/imports
- replace dangerous arguments [] with ()
- replace dangerous arguments {} with None and "if None -> {}"
- replace builtin names with something similar

Now that all problems reported by pylint are fixed, enable coding style
test.

These scripts are not used and does not respect naming conventions or
have pylint errors.

PropertySheetChild is not an acceptable prefix, use Base as the context
is used. Technically, this uses the "id as reference" mixin, but this
mixin is a bit special as it has takes the suffix as parameter and does
not reside in mixin module, so we just use Base as a prefix.

FieldValidator is not an acceptable prefix,
Base_validateDialogConfirmation seems better

Rounding scripts:

Planning is not an acceptable prefix, this script is called on the
current context, so let's use Base_ as a prefix.

Planning_roundBoundToMinute → Base_roundPlanningBoxBoundToMinute
Planning_roundBoundToDay → Base_roundPlanningBoxBoundToDay
Planning_roundBoundToInt → Base_roundPlanningBoxBoundToInt

Fix an unexpected typo.
See nexedi/erp5@71a3f9af

* Fix the problem that preference cache was not reset when `setPreferredSomething` was called, only when `edit(preferred_something=)`
* Implement something more advanced than just `portal_cache.clearCache()` so that when a user change their preference, only preference of this user are affected.
* Increase default cache duration, now that we invalidate caches properly.

/reviewed-on nexedi/erp5!983

By using a magic comment such as

```javascript
   // erp5_monaco_editor_include: ../other.js other
```
we can tell monaco editor that other.js will be loaded before the
current javascript file, which helps the typescript compiler in
diagnosing errors and providing completions.

rsvp, renderjs and jio are still always loaded because they are used in
almost all our javscripts.

/reviewed-on nexedi/erp5!969

/reviewed-on nexedi/erp5!993

When updating persistent to >= 4.4 `__repr__` of zope objects display
information about ZODB connection but it's more useful to have
information about the acquisition chain at this level.

This happens because `persistent.Persistent.__repr__` is used instead of 
`OFS.SimpleItem.Item.__repr__` (as of zope 2.13 ).

One notable case was proxy field error messages, it was supposed to
include the path of the proxy field in the error message. This one is
fixed by fixing the MRO so that persistent.Persistent does not appear
first.

/reviewed-on nexedi/erp5!991

testXHTML was testing only document contained in modules, this take another approach to test all portal types and fixes the problems found now that we test views of all portal types.

The number of tests in testXHTML also increase from 864 to 1137.

/reviewed-on nexedi/erp5!992

Without this, subclassing defining this method will have arguments-differ
pylint warning because the signature of Base._checkConsistency is
different.

These scripts are used to check the site configuration and also in
CodingStyleTest.

This is just the minimal so that CodingStyleTest can run, but this could
be extended to several scripts currently in erp5_toolbox skinfolder of
erp5_forge which overlaps with erp5_administration. Also, regarding
CodingStyleTest, it seems that erp5_administration contain a lot of
things, so we may want to extract a smaller business template containing
only the scripts used for static analysis of business template.

We want this check to run on all components.

Module is an acceptable prefix for modules

Introduce a list of skins IDs that does not match our conventions but
are acceptable because this is what external API are calling.

Some portal types such as Business Template were not tested by testXHTML
because this tests started by modules and recursively tests views based
on allowed content types. Because of this approach, types that are not
created in a module but in a tool were never tested.

With unittest, the only way to dynamically add test methods to a class
is to generate a test class in test_suite function. At this stage, the
ERP5 site is not created yet, so the test had to introspects business
templates XML.

This now uses a slightly different approach, instead of finding modules
and chain of allowed content types from business template XML, we only
use business template to introspect the list of actions.
The lookup of the appropriate containers is no longer done before setup
by static analysis of business templates XML, but later once the site is
created, by dynamic analysis of the modules and allowed content types on
the running ERP5 site during the test method.
If we don't find a chain of portal types, we create the test document in
portal_trash, a tool without filter of content types.

This way, we can test all views of all portal types. This revealed a few
problems:
 - we need developer role to create components in portal_components, for
this we add developer role to the current user.
 - Delivery Cell portal type looks not used, there are no container
accepting it. We don't test delivery cell views for this reason.
 - VCS view on business template needs preferences and working copy
setup. We just mark this test as expected failure for now.
 - Solver Decision has a form conditionnaly displayed when there's a
relation to a solver, but this test does not evaluate action conditions
and does not allow to call a script (that would made it possible to
modify the document so that the condition is true). For now we also
mark this as expected failure.

Delivery Cell is not really used these days and cannot be created in
testXHTML, so drop these useless views.

For reference, the same views still exists on all other movements portal
types, so we don't "loose anything important" by removing the views
here.

The corresponding form, BudgetTransfer_viewBudgetTransferLine does not
exist, so mark view as not visible for now so that it's not tested.

This redirects to another site, so it is not really a view action.

This also confuses testXHTML.

Maybe what's missing is an interaction to update cache internals with
updateCache when a cache is added, but that's enough to make TestXHTML
pass now.

DublinCore was missing, so getCompactTitle was causing AttributeError
_baseGetTranslatedTitle

When listboxs has an editable field, but the field is non editable, it
produces a markup like:
  <a href="link to the line"><EditableField/></a>
This is fine for most fields, but some editable fields are rendered as
an <input> even if they are not editable - this is the case for
CheckBoxField.
To prevent rendering <a><input></a> which is not valid HTML, configure
the enabled field as non-editable on alarm list view, with this change
listbox treats the field as an editable field and just render the field
as <EditableField/> without the <a>.
To keep the same visual appearance of having a disabled checkbox field,
make this field disabled with extra.

The action was in erp5_core, but the form in erp5_forge. Move the action
in erp5_forge.

The listbox list method was raising error because it was using catalog
with unsupported installation_state=. Change to use contentValues
instead.

This listbox had "Listbox" as a title, use "History" instead, which
makes a little more sense in this context.

Keep the same panel/header links than on form_view.

Translation table seems to be often picked before other conditions by
MariaDB's query planner.
When there is at least one other condition on a query which require a
join (ex: category-based relation, delivery-table-based date
comparison, ...) this may not be the most efficient execution plan.
Instead, put translation lookup into its own separate query, in a way
similar to security_uid lookup. This way, two separate execution plans
can be used to execute the overall operation.
One drawback is that the new query does not know of any portal_type
condition, so all possible original values are listed along with their
respective portal types.
This does not seem to affect query performance, and on most common
(and simple) cases where a translated value is used, like module
listboxes, the query planner should be able to eliminate irrelevant
entries as such queries are already expected to provide a list of
relevant portal types and to lack condition nesting.
Preserve custom related keys: these are used when selecting or sorting,
while the scriptable keys are used when applying a condition.
Note on UI tests: mass workflow transition listbox is not supposed to be
sorted. It's bad that these tests all rely on row order, and also bad
that they all check the whole listbox. They should pick the line they
care about and not check the rest - except for one test whose purpose
would be to check the whole list anyway.

Add support for "like" auto-detection and use as default comparison
operator.
Allows mimicking KeywordKey and DefaultKey with 2 parameters, in addition
to already-supported simple SearchKey.

MissingCategoryDocumentConstraint was incorrectly using
unrestrictedTraverse, unlike category API which uses its own
resolveCategory method with better semantics.

Because of this, some problems were not reported:
 - categories using paths that resolve by acquisition, but does not
exists as such (like in the test)
 - categories using path with dynamic objects, like a website language.

/reviewed-on nexedi/erp5!986

checkRevision now returns a callable, return a do-nothing lambda when
the expected revision is already checked out

/reviewed-on nexedi/erp5!990