D happens to have a great deal of ways to do incompatible things especially when it comes to I/O and system interface in general. here is a zoo of incompatible approaches:
The talk is presenting experimental work being done to have C libraries respect D's Fibers, goes beyond I/O and finally discusses a unified view of D's concurrency primitives.
The talk is based on results of a joint work with Alexandru Caciulescu. The talks aims to present a simple coherent runtime via a model of light user-mode threads (UMT), where I/O and blocking system calls are handled efficiently by the runtime, which maintains an event-loop.$(P Then we see how can implement such a model where even 3rd party C code behaves well when called from UMT. A very different approaches lead to radically different internal designs for Windows and Linux, which however (with a few considerations) are nicely fitting in our model.
The performance of the prototype is benchmarked against the baseline on a few typical applications.
The second part of the talk is focused on D itself and things like:
Dmitry Olshansky is a young all-around researcher and software engineer. He's been a long-time D language contributor with his most notable contributions being the std.regex and std.uni modules of the standard library. Aside from everything D-related, his main interests are compilers, runtimes, text processing, parallel and concurrent programming, scalable network systems and AI.