otherwise it is acquired from BudgetLine, which is bad because:
- the value of catalog.has_cell_content will be wrong
- hasCellContent is slow and this was called for all cells in a budget line

 * support budget cell's getInventoryQueryDict
 * raise NotImplementedError for budget cell variation. We can't
   implement this unless we generate dynamically a date time field on
   budget line's view.

Still TODO:
 * Test
 * group_by slot_index should not be needed
 * think more about a generic "domain budget variation" ?

TODO: discuss wether this can be an useful addition

We have in ZODB some PropertyGetter as new style classes (there are also
some cases in the XML exports in the repo, search for _dt_reconstructor
)

We no longer use pypdf but use pypdf2, due to some bugs, we have some
pypdf objects in object database, so create a fake module so that we can
load them.

In a prefious version of this patch we have left some
Products.ERP5.NameObject in the database and saving new version of the
docuement (edit) fails with:
PicklingError: Can't pickle <class 'Products.ERP5.NameObject'>:
attribute lookup Products.ERP5.NameObject failed

HACK: disable bank_reconciliation_aggregate_* related keys, since we added these columns directly in stock table for performance

Signed-off-by: Aurélien Calonne <aurel@nexedi.com>

Note that after applying this change it's important to clear preferred timezone
for system preferences, as they have priority over user preferences

This is wrong, I still have to understand how DMS can be used to upload
a file while making sure that if a file with same reference already
exists it is not edited in place but always a new version is created.

Only user action transitions needs permission/role guards.

This workflow is not really related to security

edit method has security definition on the class, gard is not needed
here.
This cause compatibility issues, in the past it was not necessary to
have any permission to call edit from restricted code.

This action is typically used to add notes in history in scripts,
including in cases where user does not have modify portal content
permission on the document.

Wrapping a method in a workflow method should not change the
publishable state the method.
If the original method is not publishable, wrapping it in a workflow
method should not make it publishable.  If the original method is
publishable, then the wrapped method should still be publishable.

This was always intended to work like this, as we can see in the code
comment in `WorkflowMethod.__init__` but was not properly tested and got
broken at some point.

It's important to restore the behavior, because workflow methods such as
`validate` should not be published, users must only be able to use the
user interface transitions freely, workflow methods transitions are
only available if developer expose them in a script - and perform the
necessary consistency and security checks in that script.

Only "user action" methods needs a security declaration.

This can hide bugs, especially when updating business templates.

TODO: if we drop this we can also drop the command line flag