Red Hat at the ISO C++ Standards Meeting (November 2017): Parallelism and Concurrency
Several Red Hat engineers attended the JTC1/SC22/WG21 C++ Standards Committee meetings in November 2017. This post focuses on the sessions of SG1, the study group on parallelism and concurrency. SG1 had a full schedule as usual, with Executors, Futures, and deferred reclamation mechanisms (e.g., RCU) being major discussion topics. We also started to track the state of proposals and topics we will need to discuss in a publicly accessible bug tracker.
I have argued in the past that SG1 should try to do what good Open Source projects do to better serve its community. The Open Source approach goes beyond just licensing, and good Open Source projects know how to use open collaboration processes to work efficiently. In the context of standardization, users can, for example, benefit from knowing why a certain feature was designed the way it is, or why some other feature was not yet considered ready for the standard; similarly, SG1 can benefit from user feedback that is better informed (e.g., when users evaluate features from Technical Specifications through the intended use cases).
Therefore, I am glad to see that we have started to make more of the internal state of SG1 visible. This is in an early stage so far and consists of bugs for proposals that SG1 has discussed or topics it intends to discuss. Most of the bugs might not be really useful yet for people not deeply involved in C++ standardization, but if you are enthusiastic about a certain proposal or topic, the information should be valuable.
For example, look at this bug about semaphores and Futex-like abstraction: It states that this feature is in a stage where we work on the wording (i.e., the text in the standard that specifies the semantics) and that it targets the standard directly (i.e., the International Standard (IS), instead of going through a Technical Specification first). There are a few open questions (“OQ”) and a few action items (“AI”) the C++ committee needs to take care of.
Please note that these bugs are not intended for general discussion, so please do not add your feedback there. As I have said previously, this is just a start, and I hope that SG1 will do more to enable collaboration that is more open with its community (including providing ways to provide feedback to Technical Specifications).
Switching back to technical proposals, the Template Library for Parallel for Loops was merged into Parallelism TS v2. The RAII Interface for Deferred Reclamation is ready for a new Concurrency TS once it has gone through review by the Library Working Group. RCU and Hazard Pointers are also making progress towards this TS.
We spent much time discussing Executors, as in previous meetings. It seems that the design of the proposal is slowly but steadily getting better, and we were able to resolve a few design questions or at least improve our understanding of the different opinions. The Library Evolution Working Group also provided the first round of feedback on the proposal. There are still several open questions, though. We also discussed the Networking TS and how its different execution abstractions could fit into SG1 Executors proposal.
Another big topic worth mentioning is Futures. There seems to be some consensus in SG1 that the current design is not quite what we need. There are several ideas how to improve the design, but I think we are still a few iterations away from having a new design that is good enough for gathering user feedback through a TS.
More information can be found in the Bugzilla. If you have any feedback on how SG1 could help users to contribute to the standardization process, please leave a comment.
Join the Red Hat Developer Program (it’s free) and get access to related cheat sheets, books, and product downloads.