In case of ConflictError, request will be retied for
HTTPRequest.retry_max_count times. For Zope 2, the default value of the
class attribute was 3, but on Zope 4, default value is 0 and this is set
to 3 on startup (by Zope2.Startup.handlers.root_wsgi_handler).
This startup code is not executed by runUnitTest, we need to set
HTTPRequest.retry_max_count somewhere, otherwise ConflictError are not
retried during tests.

The status message no longer use system-msg class, use simpler
"assertTextPresent" that should be enough.

Simplify the waiting for static pages, clickAndWait is enough.

I'll make a commit to handle all tests

This reverts commit 82595cd8.

This supports both python2 and python3. For python3, this decode() properly
arguments returned by string_literal() to have str(), otherwise the query string
is an str() (DocumentTemplate) and the arguments are bytes().

In other places, bytes() is used directly but in this case this is not needed
and would require monkey patching (at least) DocumentTemplate.

And _mysql/mysqldb API (_mysql.connection.query()) converts the query string to
bytes() (additionally, cursor.execute(QUERY, ARGS) calls query() after
converting everything to bytes() too).

TODO: Instead of doing this ourself, should we use cursor.execute(QUERY, ARGS)
      which does conversion to bytes()? Performances?

{encode,decode}string() aliases available but clearer this way...

And Zope now mandates it to be a string ({Shared.DC.ZRDB.TM.TM, ITransactionManager}.sortKey()).

When rendering a proxy field, 3 different fields can come in play:
1. the field to be rendered
3. the template field (i.e. not a proxy field) that knows how to render
2. possibly an intermediate proxy field that contains the value to render

What's difficult when rendering a proxy field is to take the above 3 fields
into account and this commit does it by creating a temporary field:
1. 'field' variable in TALES
2. the value
3. the code

Before this commit, 1 could be wrong.

This was deprectated because we don't have get*ById for other modules
and tools, we just use OFS API. This should also be slightly faster
because one less method call (and one less call to warning)

exiting the context manager restores the original DateTime behavior

This tests the test framework, because this code is tricky.

I'm not sure if this is used, but not being valid python code cause
problems with code analysis tools