Current and suggested file systems that provide query support treat the ``name space'' associated with the query-based access to files as logically different from the name space associated with path name-based access. This makes it very difficult (if not impossible) to offer both forms of naming within the same system. For example, it is not possible to create new files within the virtual directories of SFS [gjso:91], and it is not possible to combine views of Nebula with directories in the ``underlying'' file system [bdbcp:94]. The Multistructured Naming system [sm:92] comes close. It allows users to specify certain relationships between queries (or ``labels'') so that users can organize queries and their results in a hierarchy. (Unlike SFS, if two queries in this system are related to each other in a hierarchy, their query-results do not necessarily have to be related in any way.) However, they still do not have the freedom to group files of their choice together within a label: they must also think of a query that matches the contents of exactly these files (and no others), and associate the query with this label.
Our approach to this problem is radically different: instead of starting with a query-based naming system and imposing a hierarchy or other relationships on queries, we start with a hierarchical naming system and extend it to support query (content) based naming. We show that this approach has many advantages: it gives users a lot of flexibility and power, and at the same time it makes the system easy and intuitive to use.
The first step is to map queries and their results onto file system abstractions. For obvious reasons, we decided to map queries into directories in the HAC file system. We call such directories semantic directories. When users create a new semantic directory, they specify both its path name and its query. HAC then creates a new directory, associates it with the query, and contacts the CBA mechanism to evaluate the query. In the new directory, HAC automatically creates new symbolic links to all files that satisfy the query. These symbolic links can co-exist with other information in the semantic directory, including other symbolic links or other regular files. The symbolic links can also point to files in other semantic directories in the file system, or even to remote file systems. HAC also provides a mechanism by which the user can easily extract the results of the query from these files. Semantic directories provide the abstraction and utility of virtual directories, but in HAC they are also regular hierarchical directories for all purposes. Users can add files to them, modify them, run applications from them, and so on.
HAC allows both ordinary syntactic directories to co-exist in the same file system. Directories (whether semantic or syntactic) can be accessed by specifying path names, and they can contain files, sub-directories, symbolic links, etc., as usual. Semantic directories contain additional information that helps HAC to maintain them and keep them consistent with whatever the user is doing. The consistency problem is a new non-trivial problem that we discuss next.