This commit changes the way we obtain the original
Products.ERP5Type.Base.Base.__setstate__ method (before it is
monkey-patched (1)).

Previously, Products.ERP5Type.Base.Base.__setstate__ was inherited from
persistent.Persistent.

Since the previous commit, Products.ERP5Type.Base.Base overrides it with
its own method.

So the test was no longer looking at the right thing.

---
(1): See the code inside product/ERP5Type/dynamic/lazy_class.py and
product/ERP5Type/dynamic/persistent_migration.py

Without it, it is not possible to upgrade instances since:
48c45fbd

[test/upgradeOldDataFS] use UI callables
Ensure the admin will able to trigger the upgrader

erp5_corporate_identity_test: test pdfs are only used to store data then check converted image, so one pdf is enough

Do not limit by default

Querying the catalog does not limit the result by default..
See nexedi/erp5@241a87a1

what's wrong with this anyway

See merge request nexedi/erp5!1226

When a request got retried after a conflict error, a:

    TypeError: getPublisherDeadlineValue() takes exactly 1 argument (0 given)

was raised.

Instead of adding a new test to cover this, run all tests from
ZPublisher module (except a few which had suspicious errors).

This actual error happened when running
ZPublisher.tests.testPublish.testPublisher before the fix.

See merge request nexedi/erp5!1220

Install more business templates such as erp5_web or erp5_dms, to confirm that we can also update these business templates.

Run the tests with "full indexing", to be more realistic and to be sure that we don't hide problem that would happen when reindexing action definitions or property sheets.

See merge request nexedi/erp5!1228

See nexedi/erp5@981f2e61

In a real ERP5 actions, roles etc are all indexed, so it's more
realistic to simulate update with full indexing.

We are using html2canvas to render the page in a canvas element, but
because the canvas element was created in the selenium runner frame (and
not the tested frame), if some extra css fonts were loaded in the custom
page they did not apply.

To make the snapshot use proper fonts, we create the canvas in the
tested page.

Selenium asynchronicity does not use Promise, but calls function while
it returns true, we were implementing this logic in two places and
implementing wrongly because some case of errors where not reported.

Introduce a wrapper that makes turns Promise into a function returning a
boolean like selenium is using and use it in setFile and
verifyImageMatchSnapshot commands

Also a bit more of jsdoc comments for typescript.

This can help debugging test failures on test nodes

ERP5: Remove Capacity/GLPK: never used and broken since glpk Python module is not in ERP5 Software Release neither.

* Properly override WebSite Document in erp5_web_shadir by creating a new
  version_priority.
* Keep Web{Section,Site}TraversalHook in Product/ERP5/Document as these are
  not ERP5 objects. Also, add `kept_for_backward_compatibility_only` to
  importLocalDocument() to handle pattern where the Document class has been
  migrated but some code remains.
* Move WebSection_get*PrecacheManifestList Python Script from bt5 not depending
  on erp5_web to erp5_web_renderjs as this is required by CodingStyleTest
  (WebSection being now in erp5_web).

When a first migration was reverted for any reason (e.g. ConflictError),
the assertion failed because __setstate__ was not called yet to fix the
class back to the old one.