Programs with Recursively Defined Data Structures Using Pointers

1 Introduction The theme of this seminar was to study programs that manipulate dynamically allocated data. The attendees were researchers from three different areas: • Design and implementation of combinatorial (e.g., graph) algorithms with a heavy usage of dynamically allocated memory. Many of these algorithms are currently implemented in C++, e.g., see [MN95]. • Verification of partial correctness of such implementations, e.g., proving that a program only refers to allocated memory cells. Proving that a program does not create memory leaks. By convention, we call these cleanness checks since they must hold for every reasonable program as opposed to correctness, which is program specific. The hope is that some of these tests can be carried out by future compilers. • Compiler optimization to speed up the execution time of such programs. For example prefetching of linked data structures can improve performance by 45%, see [LM96]. Three panels of open problems were held (summarized in Section 3). Few of the open problems already initiated research, e.g., in the area of improving locality of programs that manipulate dynamically allocated memory and in the area of automatically eliminating " checking code " that validates that certain invariants are maintained after a library operation, e.g., that the insertion of a new edge into a directed graph, maintains the Euler equation. 33 talks were given presenting the state of art techniques in these areas (the schedule is given in Section 2 and the abstract is summarized in Section 4). The talks were very interesting and accessible to all groups. Some of the talks presented solutions to open problems. One of the most interesting subject studied is the treatment of memory hierarchy in general and cache in particular. Finally, one talk by John L. Ross presented a solution to a problem posed by Luddy Harrison at the Dagstuhl seminar No. 9535, held in August 1995. The schedule of the workshop was not fixed on the usual daily basis. Instead, we did an experiment on Monday afternoon: the schedule was fixed in an additional " problem session. " 3 In this session speakers from different fields were encouraged to discover relations between their works (the preceding " five minutes madness " session helped a lot) and to build a session. Moreover, we included time for questions at the end of these and other sessions to review all talks in summary. In our opinion, some of the most …