See commit c45c2295
("erp5_web_shadir: more useful checks").

This fixes how the builder processes the result of IMovementGroup.test.
Testing deliveries to find one that can be updated should have no impact
on properties if the builder ends up creating a delivery.

When i upgrade/install erp5_notebook in an instance which use neo storage instead
of zeo, Neo stopped immediately, as consequence, erp5_notebook can't be installed/upgraded.

The reason is in erp5_notebook, there has a one big file scipy.data.bin 157MB [1] which can't be
installed as it is  in neo. The solution suggested by @jm is to wrapper it by Pdata when install
a BT5.

See merge request !953

[1] https://lab.nexedi.com/nexedi/erp5-bin/blob/7d3505fd/bt5/erp5_notebook/SkinTemplateItem/portal_skins/erp5_notebook/scipy.data.bin

In 44e0b22f (Move to new API, detailed properties take precedence,
2011-09-09) we introduce a new data model for coordinates, in the case
of emails, we have "url_string" fro the "detailed" form and
"coordinate_text", for the "store what user entered" form.

In some places of erp5_credential, we use setDefaultEmailText, which
calls Coordinate.setText which uses the deprecated Coordinate.fromText,
which sets the same value for "url_string" and "coordinate_text", which
seems a questionable behavior, because the data is saved twice and
some code might be using the wrong property.

This changes every usage to the new coordinate_text property

This backports parts of zope 4 commits [bb7837c66 (Explicitly make all
classes new-style., 2017-09-15)](https://github.com/zopefoundation/Zope/commit/bb7837c66) and [173658008 (Fix unpickling of instances
those base class changed to a new-style class. (#208), 2017-10-24)](https://github.com/zopefoundation/Zope/commit/173658008) for zope2.

This is done for two reasons:
 - so that business template XML are same when exported from zope4
   and from zope2
 - so that a business template created from zope4 can still be
   installed on zope2.

See merge request nexedi/erp5!1596

This is used to trigger recompilation of bytecode, but we don't
export the byte code so we don't need to export it.
This also prevent diffs when the Script_magic changed, like it's the
case with Products.PythonScripts 4.2 , with commit 590125a (Force
recompilation of scripts., 2017-10-23).

This is patched to DateTime.DateTime._dt_reconstructor but we don't
want _dt_reconstructor to appear in pickles, especially business
templates ones.

These properties are supposed to never be set on ERP5 documents.

See https://lab.nexedi.com/nexedi/erp5/blob/34f4ae37/product/ERP5Type/Constraint/PropertyTypeValidity.py#L65

This backports parts of zope commit bb7837c66 (Explicitly make all
classes new-style., 2017-09-15) and 173658008 (Fix unpickling of instances
those base class changed to a new-style class. (#208), 2017-10-24).

This is done for two reasons:
 - so that business template XML are same when exported from zope4
   and from zope2
 - so that a business template created from zope4 can still be
   installed on zope2.

This properties are not used, no need to have them exported

If there's an error during the --save setup, we don't need to run
the --load steps, because test failed already and if they don't
fail in --load step, it would be a case where some errors are not
reported.

Usually we run tests with two nodes, so when there's a conflict it's
always between the "main" node and the processing node, so we always
have a distributing node during conflict.

But when running with, for example, 2 nodes (the main node + 2 nodes),
the conflict might occur between the two processing nodes and after
the resolution we still don't have a distributing node.

When looking up a document by reference, using getDocumentValueList from
the context of a web site / web section might not find the documents,
depending on the predicates defined on the web site / web section.

This gadget looks up a document by reference, by using
getDocumentValueList on the context of the web section.
getDocumentValueList also applies the criterions from the context, but
because web page references are already unique, adding more criterions
can just cause the document to be not found and the gadget to display
the "No document" error.

This sort was based on the id of the role definition but this did not
work when the type definition had more than 10 documents, because:

    >>> sorted(['1', '2', '10'])
    ['1', '10', '2']

This switches to an approach of converting the id to integer, with a
fallback in case the id generator is customized.

See merge request !1732

See merge request !1731

cause by nexedi/erp5!1726

On the way to Zope 4, the biggest change was DateTime >= 3, which has these main differences:

 - `DateTime` instances are new style classes
 - `DateTime.__eq__` now take into the timezone into account
 - `DateTime` supports timezone naive ( https://docs.python.org/3/library/datetime.html#aware-and-naive-objects ) - actually this is since DateTime 2.12 , but since we fully patched the method where the parsing happens, ERP5's patched DateTime never used timezone naive dates.
 - pickles are also different.

These are changes to prepare ERP5 code base to support DateTime >= 3, while keeping compatibility with the previous behavior, this means that we keep patching to keep the same behavior as DateTime 2, so that `DateTime.__eq__` ignores timezone difference and `DateTime` does not support timezone naive dates - these two different behavior seemed to introduce too much breakages, especially in project code, with no immediate benefit for our usages. For now, while ERP5 is still based on Zope 2 / DateTime 2 the fact that DateTime are new style classes has no impact, this will only impact once we switch to Zope 4.

This also change the code to adjust the part where we were depending on implementation details of DateTime and where the details changed on DateTime 3, sometimes exposing higher level API, like the new `timeZoneContext` to use in tests.



See merge request nexedi/erp5!1593

See merge request nexedi/erp5!1725

See merge request nexedi/erp5!1726

use calendar module and public DateTime API instead.

This ensures that we don't break more.

We have code depending on these attributes, they were used in pattern
like this:

```py
try:
  DateTime(user_input)
except DateTime.DateError:
  error("invalid date")
```

so it's safer to keep the attributes for now, since we are already
patching a lot

We have too much code depending on this behavior, so we keep this patch
for now.