CSR: Small: Beating Implementations of C++11 Concurrency Into Shape

点击翻译按钮获取中文摘要

The recently-ratified C and C++ standards, "C11" and "C++11"respectively, add a concurrency model that supports writinghigh-performance, portable code for machines with multiple processors.The concurrency model, however, is large and complicated; it is notparticularly easy for compiler and library developers to get all ofits corner cases right. Errors in implementing the new model canintroduce bugs into important pieces of software, such as operatingsystems, web browsers, web sites, database engines, and embeddedsystems, all of which are written, at least partially, in concurrent Cand C++.The PI's previous work on randomized testing for C compilers uncoveredmore than 450 bugs in production-quality compilers, most of which werefixed by compiler developers. The PI's current project extends thisresearch agenda to support stress testing of implementations of theC11 and C++11 concurrency model. The intellectual merit of this workstems from the need to generate random, but standards-conforming,concurrent code; the need to synthesize "test oracles" that canautomatically ascertain the success or failure of a test case; and,the need to develop "hostile" simulators for flushing out errors incompiled concurrent code.The expected impact of the PI's work is to significantly reduce theperiod during which implementations of the C11 / C++11 concurrencymodel are flaky and immature, and to reduce the lifetime of compilerbugs that are introduced during ongoing development.