Although it is very early in the process, higher-level parallelism is slated to be a key theme of the next version of C++, says Bjarne Stroustrup, the founder of the language.
Interviewed recently, Stroustrup says the 35-year-old language has to exploit new hardware, including transactional memory capabilities promoted by IBM and Intel. To take advantage of these features, C++17 will need to handle parallelism and concurrency. In addition it will have to deal with multicore processors and vector parallelism, as well as accommodate parallel algorithms, transactional memory, and task parallelism and SIMD (single instruction, multiple data) vectors, he says. “C++ as a systems language has to be able to do that well.”
Parallelism is too hard for developers now, Stroustrup says, and the use of threads and locks in dealing with parallelism can be error-prone. “We have to get to the next level of abstraction and that’s where we’re going.”
Currently, C++ has object-oriented and functional call syntaxes. “It’s a bit confusing,” and the syntaxes don’t have exactly the same meaning, Stroustrup says. “My idea is to unify them.” A unified call syntax is particularly important for writing generic code, he says.
“We hope that C++17 will be a major new version, like C++11, rather than a minor version, like C++14,” Stroustrup said. “Major versions change the way people think.” But he cautioned that some features proposed so far probably would not make it into the release.
Other major proposals for C++17 include faster compilation (championed mostly by Apple, Google, and Microsoft), contracts, and better type-checking. A meeting about the future of C++ is to be held by the standards committee at the University of Illinois, Urbana-Champagne, next week, Stroustrup says.
A draft of C++14 was approved in August, bringing minor but numerous changes and expanding on the previous version of the language, C++11. That version focused on capabilities such as lambdas. C++17 could arrive in 2017.