Pgraphite: an experiment in persistent typed object management

Reviewer: Daniel M. Berry

The paper describes PGRAPHITE, a prototype tool in an Arcadia environment for managing persistent typed objects. A typed object is a datum whose type, in the sense of Ada's strong type system, is known to all programs that use it. The way the program accesses the object is determined by its internal structure as specified by its type. A persistent object is a datum which exists longer than any activation of any program that may access it; the datum is usually stored on a disk medium. The usual program data that are allocated and deallocated during one activation of the program are nonpersistent. It is desirable that persistence of an object be orthogonal to all other aspects of the object. That is, the code which uses an object based on its type structure should be the same for persistent and nonpersistent objects. The paper correctly observes that for persistence to be truly orthogonal in this sense, it is essential that an object be deallocated only after it has ceased to be accessible to all potential users. Therefore implicit persistence, called retention elsewhere, is followed. The paper also observes that such implicit persistence is expensive in general because it can require garbage collection. Thus, most systems abandon implicit persistence in favor of program-ordered deallocation, which can cause references that are still available to point to nothing. By deciding on a particular set of object and persistent store manipulation operations, however, it should be possible to limit the complexity of determining when an object is inaccessible and thus deallocatable. PGRAPHITE implements these operations. Rather than get bogged down in the already well-understood details of implementing a complete typing system, the authors decided to build the tool as an extension of the existing system GRAPHITE, which handles nonpersistent graph objects, that is, data whose types are arcs, nodes, collections of nodes, etc. The object and persistent store manipulation operations mentioned in the previous paragraph are added to GRAPHITE to get PGRAPHITE. The rapidly developed prototype tool has allowed experimentation and exploration of alternative models and implementations of persistence. I found the paper clear, lively, and informative. It showed me a new way to view my old work on retention and got me all excited. A proof of the liveliness of the paper and my excitement is the abundance of argumentative comments that the paper provoked me to write in the margin of my copy. The paper was fun to read.