Introduce something similar to "cache cookie", but more suitable for
preferences that depend both on the site configuration and user
preferences.

We store for each user a value in _preference_cache and we modify this
value each time user change his user preference. This value is then
used part of the a cache key.
The same mechanism happen with a global value for site and group
preferences. See _getCacheId docstring for more details.

This way, when one user modify their preference this does not reset
cache for every user.

Because we have proper invalidation, increase the case duration by using
erp5_ui_short instead of erp5_ui_long

This is applied to both preference methods and to templates documents
inside preferences.

test_system_preference_value_prefererred_clear_cache_disabled is updated
to use another preference name, because it was same as in a previous
test, so that we don't have test isolation problems.

When we use preference.edit(preferred_something=something_else) the
cache was cleared, because there is an interaction on _edit, but when we
do preference.setPreferredSomething(something_else) cache was not
cleared.

This still have a problem cache is cleared too much but at least it
eliminates this incorrect behavior.

Update tests at the same time:
 - testEditorField,testAuthenticationPolicy: no need to clear cache,
   it's now done automatically.
 - testUpgrader: the workflow chain of preference changed. This test is
   asserting the actual workflow chain, so we have to update the test
   everytime workflow chain is modified.

when erp5_authentication_policy is installed, a cached method is used,
when it's not installed we just return False. We don't need to wrap this
in another level of cache.

preferences are based on catalog, it's required to tic() before a new
activated preference is effective. This test was working by chance.

ZODB Components: PortalTransforms: No need to reload() 'erp5.' dynamic modules as this done through reset.

Calling manage_reloadAllTransforms was raising KeyError.

ZODB Components: image_to_{bmp,pcx} have already been migrated but the filesystem modules still existed.

You may have to either restart your instance or call portal_transforms/manage_reloadAllTransforms
for this change to take effect.

This is a mirror of the late import of getCurrentNode within
product/ERP5Type/Core/Folder.py
Both are required depending on which module is imported first.

This reverts commit 0af3e8f6.

This is not required anymore since ERP5Type.PropertySheet is a dynamic module
(commit 9ddcbfc1).

from a41f71c3 :
> Remove the TALES for lines, columns and tab on the matrix box because
> the default behavior is to call this updateCellRange script

that's not true, we have to list the asCellRange script explicitly.

ZODB Components: ModuleSecurityInfo() should also apply to the alias module, not only to the versioned one.

This fixes an Unauthorized error when GitLoginError is imported from
erp5.component.module.Git defining ModuleSecurityInfo() for __name__
(erp5.component.module.erp5_version.Git).

Note that when migrating the content of a Product, ModuleSecurityInfo() first
parameter must be __name__.

Traceback (innermost last):
  [...]
  Module script, line 1, in BusinessTemplate_handleException
   - <PythonScript at /erp5/BusinessTemplate_handleException used for /erp5/portal_templates/832>
   - Line 1
    from erp5.component.module.Git import GitLoginError
  Module Products.ERP5Type.patches.Restricted, line 305, in guarded_import
    return orig_guarded_import(mname, globals, locals, fromlist, level)
  Module AccessControl.ZopeGuards, line 305, in guarded_import
    raise Unauthorized("import of '%s' is unauthorized" % mname)
Unauthorized: import of 'erp5.component.module.Git' is unauthorized

ZODB Components: Alias module was not added to sys.modules if the module it points to had already been loaded before.

- revert parallel gadget loading
- minor refactoring on appcache storage creation

- new skin folder with project quick view form, scripts and gadgets
- front page redirects to custom project module

with namedtuples and unicode literals enabled, our patch break with:

    Module py2stdlib, line 266, in infer_named_tuple
      ''' % {'name': name, 'fields': attributes})
    Module Products.ERP5Type.patches.pylint, line 74, in string_build
      encoding = _guess_encoding(data)
    Module astroid.builder, line 65, in _guess_encoding
      if string.startswith('\xef\xbb\xbf'):
    UnicodeDecodeError: 'ascii' codec can't decode byte 0xef in position 0: ordinal not in range(128)

because pylint/astroid uses string_build to evaluate the named tuple and the source code enabled 
unicode_literals future, string_build is called with an unicode object.

    -> for inferred in node.func.infer():
      astroid-1.3.8-py2.7.egg/astroid/bases.py(391)infer()
    -> return self._explicit_inference(self, context, **kwargs)
      astroid-1.3.8-py2.7.egg/astroid/brain/py2stdlib.py(266)infer_named_tuple()
    -> ''' % {'name': name, 'fields': attributes})
    > erp5/product/ERP5Type/patches/pylint.py(77)string_build()
    -> data = data.encode('utf-8')
    (Pdb) up
    > astroid-1.3.8-py2.7.egg/astroid/brain/py2stdlib.py(266)infer_named_tuple()
    -> ''' % {'name': name, 'fields': attributes})
    (Pdb) l
    249     def infer_named_tuple(node, context=None):
    250         """Specific inference function for namedtuple CallFunc node"""
    251         class_node, name, attributes = infer_func_form(node, nodes.Tuple._proxied,
    252                                                        context=context)
    253         fake = AstroidBuilder(MANAGER).string_build('''
    254     class %(name)s(tuple):
    255         _fields = %(fields)r
    256         def _asdict(self):
    257             return self.__dict__
    258         @classmethod
    259         def _make(cls, iterable, new=tuple.__new__, len=len):
    260             return new(cls, iterable)
    261         def _replace(_self, **kwds):
    262             result = _self._make(map(kwds.pop, %(fields)r, _self))
    263             if kwds:
    264                 raise ValueError('Got unexpected field names: %%r' %% list(kwds))
    265             return result
    266  ->     ''' % {'name': name, 'fields': attributes})
    (Pdb) pp name
    u'NamedTuple'
    (Pdb) pp attributes
    [u'foo', u'bar']
    (Pdb) pp [ (arg, arg.value) for arg in node.args ]
    [(<Const(unicode) l.4 [checkPythonSourceCodelXOzr3] at 0x7f9f1caee250>,
      u'NamedTuple'),
     (<Const(unicode) l.4 [checkPythonSourceCodelXOzr3] at 0x7f9f1caeebd0>,
      u'foo bar')]
    
    

/reviewed-on !978

To avoid duplicating _processSearchValue.

Scriptable keys are already handled by other branches below.
If one wants to provide a string or dict value to a scriptable key, they
need to wrap it into another type to disable catalog processing of these
values.

Also, add a missing ignore_unknown_columns parameter.

These are only accessed within methods which are not reachable from
this SearckKey type, so this was essentially dead code.

This allow using measure.updateCellRange to create varianted measures
programmatically.

Remove the TALES for lines, columns and tab on the matrix box because
the default behavior is to call this updateCellRange script

/reviewed-on nexedi/erp5!974

As was initially intended, but visible a directory level got lost in an
intermediate patch file.

* This didn't work anyway with concurrent processes using the same SVN working
  copy (which is a frequent use case nowadays).
* Now that it has been migrated to ZODB Components, this also breaks on reset:
  1) Thread T1 acquires the lock.
  2) Thread T2 performs a reset.
  3) Thread T2 loads Subversion and thus a new RLock instance is created.
  => T2 can acquire the lock.

Moved 'git_askpass' shell script to product/ERP5/bin (considering that this
is a very short shell script which hasn't changed in 7 years, no need to move
it to the ZODB which would require creating a temporary file...).

After updating erp5_forge, you should delete 'product/ERP5VCS/' directory
as this will only contain '.pyc' files.

/reviewed-on !973

This fixes the issue that a truncated traceback followed the json.

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