Ingo Weinhold wrote: >> As for the indexer itself, it sounds like a great idea! I think the >> index data would need to be stored as BFS attributes as others have >> suggested -- a separate database would be a kludge IMO, and likely >> to lead to sync problems. > > I don't think BFS is particularly well suited for this task and other file > systems will need a database anyway. So a database per volume tailored to > that purpose is the way to go IMHO. I think it's important to realize that there are two kinds of data that would be indexed. One is structured data, like tags in image and music files, or author and title of a document. These are typical meta data that should be stored in file attributes. The other is the full text of a document. Some file types, such as media files, only have the former. The idea of automatically populating attributes when files enter the system has been discussed before, and it seems to me that this is definitely a related issue. And more importantly, it would be a very bad idea to store the same information both in a database and in the file attributes because then they may differ, which means that search results will not reflect what Tracker shows for the attributes. The point about other file systems is moot, I think, because a mechanism for supporting live queries on non-BFS volumes will be needed anyway. Saying that file attributes should not be used for features X or Y because it is not supported on other file systems is like saying that we should give up one of the most interesting aspects of BeOS.