3 ms·
Being able to write the same program in both paradigms, perhaps even with the same parallel speedup, isn't that difficult (as soon as an appropriately rigorous
by hansvm 4y ago
Being able to write the same program in both paradigms, perhaps even with the same parallel speedup, isn't that difficult (as soon as an appropriately rigorous definition is available a proof probably won't take long). The difficulty is in gaining the same performance from structured concurrency as non-structured -- a difficulty that plagues modern languages with respect to goto as well.
- ghoward 4y agoThank you for your confidence! I agree with everything you have said, but I think it won't matter, for three reasons: 1) The switch to structured programming still happened, even with the performance hit. The advantages were so large that the performance hit didn't matter. I think the same will happen will happen with structured concurrency because I think the advantages are also that much larger. 2) The performance of structured concurrency can be improved with many techniques, including thread pools (to reuse threads). 3) If structured concurrency starts to take off (and I argue that it is), then system software will be optimized for that case. For example, Linux's clone() syscall might be optimized for the "create thread" use case. I think other OS's would do the same. Nevertheless, you are completely right.