Until now it was only used in ERP5Catalog where the expression property is
`expression` for both Python Script and SQL Method but this is a common pattern
and will also needed for ERP5Workflow.

Instead of generating the report in one big activity, split the calculation
in several batches, so that the processing can be distributed on zope nodes.

The limit for the number of lines was also increased from 10000 to 100000

Using an absolute URL for gadgets cause the gadget to have a different URL
per language when the web site uses "Language in URL" feature.

Using absolute URL is not good for cache and also it seems to cause
incompatibilities with recent versions of ERP5JS. Using relative URL is
what we usually do, it does not have cache problem and works fine.

See merge request nexedi/erp5!1345

for consistency

for consistency

Even though it was enabled on packing list, this report was showing orders.

This report can be used for purchase, then it shows supplier. In the
report dialog, we have "By Third Party" report mode, so use consistent term.

This report shows packing lists, but in the dialog we could only select
the states of orders.

a928abd2 (web_renderjs_ui: use lxml to extract data-i18n messages, 2020-12-14)
was an old version of the patch that was pushed accidentally. This applies
the missing bits.

This is already working for multi-column conditions (among other malformed
values), but in the monocolumn case the type-check treats None in the same
branch as sequences, which cause the error:
  TypeError: 'NoneType' object is not iterable
As the condition is already covered for multi-column case, put this test
in the monocolumn codepath.

form_discussion_view embeds an editor, but it did not expose notifySubmit
method to child gadgets. Because using Ctrl+S in the editor calls
notifySubmit on parent gadget, Ctrl+S was causing a page crash.

Expose such a method that would submit a comment when called.

The previous regular expression based approach sometimes could not extract
message properly. Using xml parser simplify code and fix several messages
that were not extracted properly, like messages containing ", [] or {}

This also fix some problems when looking for messages sources:
  - archived web pages were sometimes used instead of published ones
  - messages from gadgets implemented as page templates/OFS files were not
    extracted.

A few more unit tests for the scripts involved in this process are added.

Fixes the handling of:
- non-sequence dependency_value
- incorrect dependency_value length
- None in dependency_value
- None in dependency_value items

So more sequence types are accepted by _getExecutableMessageSet.

See nexedi/erp5@6c4f7600

Allows to keep UI tests working independently of the login process
implementation details.

This reverts commit e5c160bc.

in API script AccountingTransaction_createReversalTransaction

Opening an accounting period is refused if the start date of the period
is more than one day after the stop date of the previous period, but this
check did not take into account that "next day" might be more than 24
hours, like it's the case when daylight saving happen between these dates.
Instead we check that the difference is less than 1.9 days.

Reorganise tests to group accounting period related tests in a dedicated
test class and add missing tests for period validation checks.

Also fix a few race conditions with catalog indexing that are probably not
a problem in real life but were revealed by the test.

See SQLBase._getExecutableMessageSet for operation principle.
Removes the notion of order_validation_text: activity validation is no
longer evaluated per-activity , but per-dependency for multiple activities
at a time. In this context, order_validation_text does not make sense as
it flattens all dependency types for a given activity.
Rework activity-dependency-to-SQL methods: use a dict rather
dynamically-generated method names.
Based on initial work by Julien Muchembled.

Seen happen on SQLDict with 94 indexation activities being stuck assigned
to a processing node on which load failed because of an SQL deadlock.

The main reason for not using cache.addAll was to fill the cache as a low
priority task, not to generate too much traffic, the safari compatibility
was secondary. Now that safari seems to support cache.addAll, remove the
comment about safari.