To preserve the consistency of code and data, let the new service worker wait until all tabs and windows of the old version are closed.

Don't give request object itself to cache.match. Firefox's Cache Storage does not work properly when VARY contains Accept-Language. Give URL string instead, then cache.match works on both Firefox and Chrome.

[erp5_web_renderjs_ui] Fix rjs_gadget_erp5_serviceworker.js. Use cache.add because safari does not support cache.addAll.

Collect a list of files from service worker code.

Fetch usage can be bypassed to do not use service worker when not needed.

As appcache has been dropped on Firefox, this change will improve the speed on Firefox.

No change is expected on Chrome/Safari.

erp5_trade: Fix PackingList_getODTDataDict.py. Get source reference from supply paths. Movements are not predicates.

Otherwise, this test fails because of duplicate uids.

Display precalculated listbox content returned after a form validation
error.

Ensure the cell error_text is displayed and can be removed.

Return the submitted value in case if present in the REQUEST.

Return the cell error_text if present.

Execute the listbox listmethod in case of ValidationError.
Ensure that it is rendered with the same query parameters.

This commit does not change the behaviour of the gadget.

The goal is only to make the next commit more readable.

Even if it does not lead to ConflictError being raised, updating _max_uid
has a cost and a transaction-serialising effect (most noticeable with NEO).
This value is not used (outside of a bootstrap value for portal_ids
generator), so stop doing this unnecessary work.
Also, fold produceUid into its only caller.
Also, change the folded pattern to react to an empty buffer rather than
test for emptiness before each access.

The way of truncating the index file was an error in git 2.11.0, but it
was not an error with git 2.17.0.

/reviewed-on nexedi/erp5!863

Follow xhtml_style CSS.

This is nearly a copy/paste of renderJS error rendering.

While rendering a field_list gadget in non-editable mode,
the `text_content` parameter was given only in the first `render` call.

/bug #20190204-39AF19
https://nexedijs.erp5.net/#/bug_module/20190204-39AF19

/reviewed-on nexedi/erp5!860

Before we were using up to 10 testnodes per test suite. Push it to 15 testnodes to have
chance to run big test suite (like whole ERP5 tests) in less than 2 hours.

/reviewed-on nexedi/erp5!859

ZODB Components: Use regex to check whether a class of the same name as the Component reference is defined.

Until now, there was a very basic string check which didn't handle such cases:
  class DocumentReferenceAnything():
    pass
  class DocumentReference():
    pass
=> _message_reference_class_not_defined was incorrectly raised as the first class was considered.

This allows to know the node working on test, thus we can redraft
lines when a runTestSuite command fails

Before we were looking only create date of test results to give priority of work to do for a test nodes.
So we had cases where some test suites with medium priority had 5 testnodes, and test suites with
high priority having only 3 testnodes.

So clearly distribute the work depending on the maximum quantity of testnode we want.

Right now we have this scenario:
- test result line is started
- sometimes, runTestSuite fails (like timeout), failure is reported
  but test result line remains started (we don't know yet which
  line is associated with testnode)
- when a test result line is "started" since more than 4 hours,
  test result line is redrafted
- test can be reexecuted
Speed up by removing the need of waiting alarm, by knowing which
test result line is executed by which test node, and by redrafting
immediately the test result line on test node failure

Put auto-generated _getAcquireLocalRoles at the end of mro, removing it
from Base.
Also, document why it is treated specially here.
Also, add _getAcquireLocalRoles methods on a few leaf classes whose
instances fake being portal types, without actually being proper document
instances. At least, the ones which are detected in unit tests. The
proper fix would likely rather be to make them proper document classes,
but this carries data migration concerns which go beyond the scope of
this regression fix (_getAcquireLocalRoles was not possible to override
anymore).