Henk Hangyi <[EMAIL PROTECTED]> wrote: > I agree, we really need a crystal clear definition of directionality.
I propose we simply start avoiding the word, because everybody seems to mean something else with it. > A. On CREATION of a relation the definition of the relation-type defines > the result. > > A.1. If the underlying RelDef object is BI-directional a relation is > added in the direction it was created. ANY SWAPPING OF SOURCE AND > DESTIONATION SHOULD BE CONSIDERED A BUG. > > A.2. If the underlying RelDef object is UNI-directional a relation may > only be added in the direction specified in the RelDef object. Creating > a relation in the opposite direction should result in an error message > like: "ERROR: the relation from images to news is not possible. Maybe > you wanted to create a relation from news to images, which is possible". I think that on creation the 'dir' field is more or less irrelevant. It is filled only in the 'dir' field of the insrel. Swapping of source and destination (should) only occur(s) if there is only an allowed relation in that direction (and not the in the requested one). > B.On USING a relation the result is defined by: > > B.1. If the underlying RelDef object is BI-directional the searchdir > determines the result. > > B.2. If the underlying RelDef object is UNI-directional the relation > will only be used in the direction specified in the RelDef. Specifying a > searchdir on an UNI-directional relation should result in an error > messag like: "ERROR: the relation from images to news is > uni-directional. The searchdir parameter is only available for > BI-directional relations." Searchdir could also be used in uni-directionial relations, but it does make little sense. searchdir = source will not find such relations. The dir field of relations and reldefs means only something for the _visibility_ of it, and nothing for the direction or possible directions. One can justify the field to be named 'dir' only because this visibility is expressed in a direction (visible from 2 sides, or from 1 side). > > What do you think? I think your B is more or less accurate, in A you perhaps too much stress the relative unimport bi/uni-directional visibility. Michiel -- mihxil' Michiel Meeuwissen Mediapark C101 Hilversum +31 (0)35 6772979 []()
