Abstract

Conventionally, a safety property [1, 6] for a concurrent implementation I informally requires that no” bad” thing happens at any point in any execution of I. If a” bad” thing happens, there is no way to” fix” in the future, which implies that a safety property must be prefix-closed: every prefix of a safe execution of I must also be safe. Moreover, a safety property should also be limit-closed [6]: the infinite limit of any sequence of ever extending safe executions must also be safe.

Date

2012

Authors

PK Hagit Attiya, Sandeep Hans, S Ravi

Journal

4th Workshop on the Theory of Transactional Memory, Madeira, Portugal