On 25 Sep 2008, at 4:24 AM, Henri Sivonen wrote: >> [...] > > Here's an example of a table with at least one headers/id cycle: > http://www.usdoj.gov/jmd/fass/afvreport2005.htm > > Regardless of how common this is, if headers cells can be chained > AND authors can add arbitrary header associations, it is *possible* > to construct a cyclic header chain. A proper spec must, then, > define in detail what an agent is to do with cycles. Here's one processing rule that I would submit for consideration: Grow a running set of the nodes that are included in self +recursivelyRelatedBy@headers. When an @headers arc (one ID appearing in an @headers attribute) leads to a node already in that set, drop _that arc_ (quietly) and continue to follow the rest. This allows for mutual dependency without creating a non-terminating process. > (What does existing client software do, btw?) I'll try to get this confirmed, but AFAIK the present implementation does not chain @headers references. I just reads, when asked, the contents of the cells mentioned in the @headers of the cell where the focus was when the AT was asked for more info. Maybe in conjunction with the first-in-row, first-in-column heuristic headers. Al > -- > Henri Sivonen > hsivonen@iki.fi > http://hsivonen.iki.fi/ > > >Received on Sunday, 5 October 2008 14:12:38 GMT
This archive was generated by hypermail 2.2.0+W3C-0.50 : Sunday, 22 March 2009 03:32:10 GMT