Terry Kerr wrote at 2003-4-21 09:04 +1000:
I don't think its a bug...it looks very purposefull! The KeywordIndex uses the same _apply_index() as the FieldIndex, so will match on the same values. The method uses BTree.IITreeSet.intersection if the operator is 'and' on the keywords, and as clearly documented in IITreeSet.Interfaces, the intersection method if provided two sets, one which is None, will return the other set. So if one keyword has no matches, then the result is essentially ignored.
That's an implementation argument. However, semantics is more essential than implementation or, more precisely, the implementation must be such that its implements the semantics. When your query is keywordindex= {'query' : [key1,key2], operator='and',} then the semantics (of the "and" operator) dictates that you get (the) objects indexed under both "key1" and "key2". Thus, when (e.g.) no object is indexed under "key1" you must get the empty set.
....
Terry Kerr wrote at 2003-4-20 15:54 +1000:
I have a Keyword Index setup in a Catalog, and when I search for more than one keyword using a query such as {'query':['one', 'two'], 'operator':'and'}, and there are no indexed records with either of the keywords, all records in the Catalog are returned. In this instance, I want NO records to be returned. Is this easily possible?
Sounds like a bug...
File a collector report (<http://collector.zope.org>).
Look at the implementation and try to fix the bug.
Dieter