Users can create, remove, rename (move), or modify any data in the file system at will. There is therefore a possibility that the set of transient symbolic links in a semantic directory sd may not represent the current result of evaluating its query. This gives rise to a data-inconsistency problem. Data inconsistencies manifest themselves in the following ways: (i) A query result can contain an invalid link to a file that no longer exists, has been renamed, or has been modified so that it no longer satisfies the query, (ii) A query result may not contain a link to a new or modified file when it actually should. For example, new email that match the fingerprint query should be added to that semantic directory, and at the same time if a certain matching file has been moved to an area outside the scope of the query (e.g., it was deemed old and moved to archive), it should be removed from the semantic directory.
Though HAC removes scope-inconsistencies from the file system as soon as possible, HAC does not remove data-inconsistencies instantly. We could have adopted such a policy, similar to databases, but we believe that file systems typically do not require it, and the extra cost (determining when files have changed, re-indexing files automatically, etc.) will not warrant it. At present, HAC invokes the CBA mechanism to reindex the file system periodically (say, once a day or once an hour), determined by the user. At reindexing time, all scope and data consistencies are settled. HAC also allows users to initiate reindexing at any time, and for any part of the file system. So, for example, users can decide to update certain semantic directories as soon as new mail comes in, but not when an application modifies some files in the file system. In future, we plan to explore more sophisticated mechanisms to enforce data consistency in file systems.