摘要:This paper presents a new approach to exclude problems arising from dynamically switching between protected concurrent and unprotected single-threaded use of shared data when using software transactional memory in OO languages such as Java. The approach is based on a simple but effective programming model separating transactions from non-transactional operation. It prevents the application programmer from errors but does not force the software transactional memory library to observe non-transactional access and thereby preserves modularity of the software. A prototypical toolchain for validation and source code instrumentation was implemented as a proof of concept.