Making Concurrency Mainstream

Опубликовано 7 сентября 2016, 16:48

The most widely used concurrent software techniques, which are based on threads, monitors (or approximations to monitors), and semaphores, yield incomprehensible and untestable software. Bugs due to race conditions, timing unpredictability, and potential deadlocks can go undetected for a very long time. Unexpected interactions between even loosely coupled software components can destabilize systems. Yet increased parallelism in general-purpose computing (particularly multicore systems), multi-threaded languages such as Java and C#, increased networking in embedded computing, and a growing diversity of attached hardware requiring specialized device drivers mean that a much greater fraction of software is concurrent. Software designers are poorly equipped for this. They use threads because syntactically, threads change almost nothing. They only later discover that semantically, threads change everything. By then it is too late. Yet there is no shortage of theory; there is a mature community with sound, well-developed techniques that the mainstream largely ignores. How can we change that? In this talk, I will make a case that composition languages, which describe program structure only, can be coupled with concurrent models of computation and conventional imperative languages to form a powerful troika. Such heterogeneous combinations of languages do have a chance for acceptance, and in certain niche situations, have already achieved a measure of acceptance.