But still rely on it by default.

Use Owner role to help filling user_set: owners must be users, never
groups.
Only look for users among group_ids when they are candidate for indexation
into a catalog_role column. This should very significantly reduce the
number of user lookups, even bringing it to zero on instance with default
catalog setting of catalog_role_set = set(["viewable_owner"]) .
Finally, only look for users if there are candidate for lookup.

Test loop invariant outside of loop.
Use clearer and more consistent variable names.
Avoid iterating twice when once is enough, allowing removal of
"local_role_dict" local and simplification of corresponding dict
structure.
Avoid creating new instances when the existing one can be mutated.
Evaluate dict as boolean instead of constructing an empty dict to compare
against.
Avoid using %-operator just to concatenate strings (allows en-passant
typechecking).
Drop unused locals.
Do not use a dict when a set would be sufficient.

So that it can be reused outside of group manager without having to
duplicate the backward-compatibility order.
Also, simplify code.

Move variable initialisations closer to where they are actually used.
Do not initialise a variable which will always be initilised again before
its first use.
Reduce argument indentation.
Drop unhelpful comments.
User document may not be a Person instance, use a more neutral variable
name.
Avoid single-use locals.
Take advantage of zLOG.LOG's "error=True" idiom to not have to pull
exception information ourselves.

Drop failed activities instead of retrying execution and expecting it to
both fail and do something (execute alarm, then index result).

And not the string "None". I thought ZSQLMethods were supposed to complain
when receiving None for non-optional fields...

manage_afterClone clears workflow history, which is precisely what
"duplicate" is trying to avoid.
This fixes a regression introduced in
  commit 7e810d5a
  Author: Vincent Pelletier <vincent@nexedi.com>
  Date:   Tue Feb 20 17:06:46 2018 +0900

      CopySupport: Factorise clipboard-related methods and code paths.

The Diff view in new UI will be using IFrames to display the diff.

Also, add changes in tests:

erp5_ui_test: Update history test according to use of diff gadget
erp5_web_renderjs_ui_test: Update tests according to new Diff view for Historical Revisions
testXHTML: Add Historical Comparison View to the list of listbox with no selection name
testXHTML: Simplify test_html_file and include way to ignore single html file

Language modification is done on another page, accessible from many places for now.

The original need to touch this comment is that it mentions now-removed
Folder_reindexObjectList script. But also:
- it mentions multi-ZEO setups, which are now discouraged if favour of
  using NEO
- it mentions pre-conceived optimisations, which can be wrong and hence
  misleading (up-to-date measures are needed before deciding what to
  optimise)
- it mentions a number of activity nodes which is nowadays exceeded
  without seeing any obvious issue caused by activity table deadlocks
  (which are nevertheless still likely an issue, just not yet breaking
  scalability)

These are now superseded by recursiveReindexObject.

This method does not belong to Folder (nor Base, even as an alias), as
it recurses wihout any limit in a tree whose node count is unknown.
Such generic recursion should either be deferred (split in multiple
transactions, each processing a limited chunk of the tree) or not be
recursive (which of course depends on the need).
There is a single caller left (constructInstance) where the subtree size
is assumed to be reasonable, as it was constructed inside current
transaction. So essentially rename this method so it is only used in such
conditions.

isAncestryIndexable is a parent-recursive version of isIndexable: it
will return False if any parent is non-indeable. This simplifies
disabling indexation of an entire tree, from its root to its deepest
leaves.
isSubtreeIndexable allows preventing any subobject from being indexable,
without having to mark each individual subobject as non-indexable. This
is different from isAncestryIndexable, as it allows having an indexable
root, but no children indexable. This is especially useful for Trash Bins,
where the bin itself should be indexable (because there is no reason not to)
but its content must not be indexed, while being left as it was found in its
original location.

Should make Folder_reindexAll and most custom indexation methods obsolete.
Remaining valid reindexation methods are:
- reindexObject: for a single document, which may contain subdocuments
  which indexation is not necessary
- recursiveReindexobject: for any subtree of documents
- ERP5Site_reindexAll: for site-wide reindexations, as there is a semantic-
  dependent indexation order.

Also, uniformise and factorise spawning immediateReindexObject.

Also:
- testSupply: Drop check for the previous magic threshold.
  _recurseCallMethod takes care of it all now.
- testXMLMatrix: Let activities execute before changing cell id.
  This works only because recursiveReindexObject on the matrix spawns a
  single recursiveImmediateReindexObject activity on that context. Now,
  up to 1k immediateReindexObject activities (for the first 1k sub-objects)
  are spawned immediately, preventing their renaming immediately after
  commit.
  So let test wait for indexation before trying to rename.
- testERP5Security: More activities are now spawned immediately, adapt.

after_method_id is too aggressive a dependency.
after_path_and_method_id is marginally better, but is bad practice
nevertheless, as such activity dependencies remain very hard to introspect.
Tags allow more variations, and are hence easier to follow.
Also, add XXX comments when (recursive) reindexation is only added to
represent the dependency, without certainty that a reindexation is actually
needed per-se: maybe another place in the code chould just appropriately
tag its reindexation activities instead, saving at least the cost of
spawning one more activity set, and at most the cost of actually
reindexing these documents.

Also, drop isIndexable instance property, unmasking class property, which
already has the value we need.
Also, factorise catalog clearing by reusing ERP5Site_reindexAll's.

Allows having a more consistent API between this tool and all other
Nexedi-maintained tools.

Reduces code duplication.

portal_activities contains documents which need to be (and are) indexed.
For consistency, the tool itself should be indexable.

So it is not just an unused constant (OFS_HANDLER) and no implementation.

Also, clarify what exactly was the broken BTree marker value, fixing
isBTree accessor. Update associated tests.

erp5.util.testbrowser: Refactor of openNoVisit() and {Image,Submit}Control for recent zope.testbrowser (followup of 5a0b3763).

And specifically, Domains (more portal types may be added to the relation
string field widget as needed).
Predicates get expanded to the list of documents they match.