This is a follow up commit to cd18daf4,
to fix the behavior in WSGI.

Zope WSGI no longer works like Medusa in the case of "response.write".
Before this change, the response contained the contents of
"response.stdout" (which we want), plus the contents of the body (the
traceback created by the raised Exception, which we do not want). Note
that the "content-length" header was not correctly set either, and parts
of the response were cut off.

Reminder: raising an exception is done in order to doom the transaction.

We are working on reimplementing WSGI/Medusa compatible streaming in a
better way in the near future; this commit is a temporary fix.

This is a backport from Zope (see commit
9c8dd7672cbef67573b3641178f12337c80de074).

>>> True and [(1,2)]
[(1, 2)]
>>> False and [(1,2)]
False
>>> [] and [(1,2)]
[]

This is needed for compatibility with existing actions.

Fields must propagate all values requested by ERP5.

Add a new url_generator mode

Do not redirect to 'view': let the js handle the default navigation

Do not generate view url if document does not acquire the hal web section

This is to support SLAPOS-EGG-TEST which runs tests for multiple
repositories from the same test suite. Evaluating each commit from
each repository with the result of the test result is problematic because
every time one test fail, all repositories are marked as failed.

Introduce a new property on test result repository to optionally define
the pattern of test result line to consider for this repository. This way,
with our SLAPOS-EGG-TEST running tests for slapos, slapos.core,
... etc and producing test result line with the same title as the repository
name, we can use something like `slapos$` (the pattern is a regular expression)
to match only slapos,  `slapos\.core` to match only slapos.core etc.

This also allow to define more complex relation if necessary, for example kedifa
and caucase seems related, maybe we want to configure so that a failure on kedifa
test suite marks the commits from both kedifa and caucase as failed (or maybe
not, but at least this way should make it possible)

/reviewed-on nexedi/erp5!964

When a commit is no longer reachable from a branch, gitlab API does not
allow annotating the commit.

This happens quite often when we push force in the branch after test
result started to run on this branch.

/reviewed-on nexedi/erp5!972

The previous way of using portal.ERP5Site_absoluteUrl() +
test_result.getRelativeUrl() could not work with ERP5JS URLs.

To be compatible with this and also maybe erp5_web-style URLs if we want
to link to public tests results, we should have more flexibility in how
we generate URLs.

Introduce a "test result url template" property on gitlab connector, see
unit test for a renderjs ui compatible usage.

/reviewed-on nexedi/erp5!963

Because gitlab way of passing project ID as user/project URL-encoded
does not support dots (it would needs to be encoded as %2E) and that
requests rewrites %2F from URL to . (which is probably correct, that's
the RFC), we cannot use encoded paths for when they contain a dot.

To workaround this, when user or repo as a dot, make another API call to
get the numerical project ID and use that instead.

We only do that for projects with a dot because that's an extra request
we'd like to avoid.

/reviewed-on nexedi/erp5!962

While deploying nexedi/erp5!924 we found out that this script was not working in these cases:

 * when buildout section id contains `-repository` there's a convention that this is stripped and does not appear in test result reference ( implemented [here](https://lab.nexedi.com/nexedi/erp5/blob/31804f683fd36322fb38aeb9654bee70cebe4fdb/erp5/util/testnode/NodeTestSuite.py#L95) )
 * when buildout section id contains `-` this script crash
 * when there's no buildout section id and there's no fix for that, because buildout section id is a required property. If we encounter this failure, we fix the data by putting a buildout section id on test suite repository.

/reviewed-on nexedi/erp5!961

Guarded with Manager role, as they allow exporting entire source files & scripts.

ZODB Components: Make 'Origin' field read-only as this is set during migration and should not be modified later on (cf632afb).

ZODB Components: Revert "Also, allow developer to call validate action from 'modified' state to handle import use case" (part of 1cb2dc4c).

This was badly implemented as it didn't handle the following use cases:
  * {draft,invalidated} => validated.
  * clearRecordedPropertyDict() was not called on re-validate.

For now, re-validate through modifying {text_content,reference,version}.

Until now it was disabled which means that if Component A imports Component B
and the latter is not in {validated,modified} state, Component A could be
validated without raising any error and failed at execution time.

As this relies on Pylint transform/hook and avoiding monkey patch as much
as possible, make Products.ERP5Type.patches.pylint available for Python3 (not
actually tested with ERP5 but no AttributeError nor ImportError with Python3
with this code).

Also, allow developer to call validate action from 'modified' state to handle
import use case:
  1. Edit A which raises an error stating that B.
  2. Fix B and validate it.
  3. Validate again A without requiring a modification of {reference,version,text_content}.

ZODB Components: Source code was checked even when setting text_content_{error,warning}_message_list properties.

So only check source code and validate when _set{TextContent,Reference,Version}
are called and remove wildcards.

Because this is using SQLDict, when we had multiple conversion
activities for the same path with different conversion parameters, only
one was executed.

We don't want this here, as the recent failures on
testDmsWithPreConversion.TestDocumentWithPreConversion are showing, when
the same image is referenced with different conversion parameters we
want to pre-convert for all conversion parameters.

Use SQLQueue, which allows calling the same method more than once.

/reviewed-on nexedi/erp5!971

Sometimes validator.erp5.net replies with error code 502 and this cause
ERP5 tests to fail. While this is a bit like ignoring problems, problems
with validator.erp5.net should not impact ERP5 tests results.

/reviewed-on !968

This is a fixup for commit 1519d81a.

There was a typo.

This test was using a blind `pause` instead of an explicit waitFor*, so
it was failing when machine was not fast enough.

Also refactor the test a bit to simplify the assertion by first
selecting the iframe we are going to inspect and make the test
compatible with both chrome and firefox - they behave differently when
interpreting new lines in selenium test tables.

/reviewed-on nexedi/erp5!967

We want to have FAILED test results before PASSED, to easily see the
failed tests.

Apparently this was a local change on nexedi ERP5 for some time but the
change got lost in last update. Let's commit it to make this
permanent.

/reviewed-on nexedi/erp5!960