Jan L.A. Van de Snepscheut, "Algorithms for On-the-fly Garbage Collection" revisited, Information Processing Letters 24, 1987, pp 211 -- 216.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....colours between two generations acting on the same physical bits may be further investigated. ffl The atomic section performed by the mutator can be refined into : Mutator is . left or right(k) i ; shadeG 1(i) As shown by Van de Snepscheut, the order of operations is important [Van de Snepscheut 87] As also shown in [Dijkstra 78] the proof must be revised since we now violate invariant (P1) Between the two operations, there exist an edge from a black G 1 to a possibly white G 1 cell. Invariant (P1) must be refined to support this temporary and local configuration. ffl Remark that in ....

Jan L.A. Van de Snepscheut, "Algorithms for On-the-fly Garbage Collection" revisited, Information Processing Letters 24, 1987, pp 211 -- 216.

....turned out to be wrong, and was discovered by the authors just before the proof reached publication. Ben Ari later proposed the same modification to his algorithm and argued for its correctness without discovering its flaw. Counterexamples were later given by Pixley [Pix88] and van de Snepscheut [dS87] Furthermore, although Ben Ari s algorithm (which is the one we verify in PVS) is correct, his proof of the safety property was found to be flawed. This flaw was essentially reproduced by Pixley[Pix88] where it again survived the review process, and was only discovered ten years later by ....

.... the review process, and was only discovered ten years later by Russinoff during the course of his mechanical verification [Rus94] Ben Ari also gave a flawed proof of a liveness property (every garbage node will eventually be collected) that was later observed and corrected by van de Snepscheut [dS87] The garbage collector has also been specified and verified in the UNITY framework [CM88] using a notion of refinement called superposition. This refinement notion differs from ours in the sense that the initial algorithm (specification) is not regarded as a specification of lower levels. Rather ....

J. L. A. Van de Snepscheut. "algorithms for on-the-fly garbage collection" revisited. Information Processing Letters, 24, March 1987.

.... like many others, is based on the concept of multi area collection which was pioneered by Bishop [Bis77] Distributed [Hug85, LL86, Rud86, Bev87, Gol89, Der90, Piq91] or fault tolerant [Ves87, Sch89, SGP90] or real time [Bak78, QBQ89, Yua90, Bak91] or concurrent GCs [KS77, DLM 78, HK82, vdS87] among others 8 ) have been studied for long. To mix increment and decrement messages through asynchronous communication links to maintain reference counters raises some difficulties. However elegant solutions [Bev87, Gol89, Piq91] have been proposed based on variants of reference counters ....

Jan L A van de Snepscheut. "algorithms for on-the-fly garbage collection" revisited. Information Processing Letters, 24:211--216, March 1987.

....it correct revealed a (sometimes deep) bug. Their solution involves three colours. Ben Ari s later solution is based on the same algorithm, but it only uses two colours, and the proof is therefore simpler. Alternative proofs of Ben Ari s algorithm were then later published by Van de Snepscheut [6] and Pixley [17] All of these proofs were informal pencil and paper proofs. Ben Ari defends this as follows: So as not to obscure the main ideas, the exposition is limited to the critical facets of the proof. A mechanically verifiable proof would need all sorts of trivial invariants . and ....

....before pointer redirection) This claim was, however, wrong, but was discovered by the authors before the proof reached publication. Ben Ari then later again proposed this modification and argued for its correctness without discovering its flaw. Counter examples were later given in [17] and [6]. Furthermore, although Ben Ari s algorithm (which is the one we verify in PVS) is correct, his proof of the safety property was flawed. This flaw was essentially repeated in [17] where it yet again survived the review process, and was only discovered 10 years after when Russinoff detected the ....

[Article contains additional citation context not shown here]

J. L. A. Van de Snepscheut. "Algorithms for On-the-Fly Garbage Collection" Revisited. Information Processing Letters, 24, March 1987.

....an algorithm satisfying G2, usually a supplementary algorithm (typically of the mark and sweep type) is used to collect cyclic structures of garbage. In our application, however, cyclic structures of garbage objects do not occur, and a supplementary algorithm is not necessary. Mark and sweep [1, 10, 29, 30]. Collectors of the second type mark all reachable objects as such, starting from the roots and recursively marking all descendants of marked objects. In this way all reachable objects become marked eventually. The design of the marking algorithm is complicated by the possibility that references ....

....objects are marked, and subsequently all unmarked objects are reclaimed. In this section it is shown how the garbage collection algorithm of Ben Ari [1] can be transformed into a termination detection algorithm. Actually, we use a variant of the algorithm described by Van de Snepscheut [29]) This algorithm was designed to run concurrently with a single processor mutating the references contained in memory cells. Thus the copying of a reference is a single atomic step, where we have assumed so far that it consists of the sending and receipt of a message. When using Ben Ari s ....

[Article contains additional citation context not shown here]

Van de Snepscheut, J.L.A, "Algorithms for On--the--fly Garbage Collection" Revisited, Inf. Proc. Lett. 24 (1987) 211--216.

....it correct revealed a (sometimes deep) bug. Their solution involves three colours. Ben Ari s later solution is based on the same algorithm, but it only uses two colours, and the proof is therefore simpler. Alternative proofs of Ben Ari s algorithm were then later published by Van de Snepscheut [4] and Pixley [10] All of these proofs were informal pencil and paper proofs. Ben Ari defends this as follows: So as not to obscure the main ideas, the exposition is limited to the critical facets of the proof. A mechanically verifiable proof would need all sorts of trivial invariants : and ....

....before pointer redirection) This claim was, however, wrong, but was discovered by the authors before the proof reached publication. Ben Ari then later again proposed this modification and argued for its correctness without discovering its flaw. Counter examples were later given in [10] and [4]. Furthermore, although Ben Ari s algorithm (which is the one we verify in PVS) is correct, his proof of the safety property was flawed. This flaw was essentially repeated in [10] where it yet again survived the review process, and was only discovered 10 years after when Russinoff detected the ....

[Article contains additional citation context not shown here]

J. L. A. Van de Snepscheut. "algorithms for on-the-fly garbage collection " revisited. Information Processing Letters, 24, March 1987.

....it correct revealed a (sometimes deep) bug. Their solution involves three colours. Ben Ari s later solution is based on the same algorithm, but it only uses two colours, and the proof is therefore simpler. Alternative proofs of Ben Ari s algorithm were then later published by Van de Snepscheut [5] and Pixley [11] All of these proofs were informal pencil and paper proofs. Ben Ari defends this as follows: So as not to obscure the main ideas, the exposition is limited to the critical facets of the proof. A mechanically verifiable proof would need all sorts of trivial invariants . and ....

....before pointer redirection) This claim was, however, wrong, but was discovered by the authors before the proof reached publication. Ben Ari then later again proposed this modification and argued for its correctness without discovering its flaw. Counter examples were later given in [11] and [5]. Furthermore, although Ben Ari s algorithm (which is the one we verify in PVS) is correct, his proof of the safety property was flawed. This flaw was essentially repeated in [11] where it yet again survived the review process, and was only discovered 10 years after when Russinoff detected the ....

[Article contains additional citation context not shown here]

J. L. A. Van de Snepscheut. "algorithms for on-the-fly garbage collection " revisited. Information Processing Letters, 24, March 1987.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC