sobjectizer VS libunifex

Message Delivery Tracing aims to debug an application built on top of SObjectizer. In essence, it logs the primary stages of the message delivery process, allowing visibility into whether there is a suitable subscriber with the corresponding event handler.

An additional rationale for structuring cooperations in hierarchies is to facilitate the sharing and propagation of dispatchers. A recent update of SObjectizer includes new functionalities that allow access to both agent and cooperation dispatchers. This enhancement was prompted by feedback provided by a user and myself.

If a stop signal arrives, it will be enqueued at the end as a demand for image_producer_callback. This means, it will be processed after the other 6 demands currently in the queue. Maybe this is not an issue but in some cases it might be. At this point, another feature of SObjectizer is to consider: agent priorities. Essentially, this feature allows for the demands to be handled in different orders based on the priorities of agents. In this context, if we assign image_producer_callback a higher priority than others, the “stop signal” would be processed before the rest of the requests.

The real solution consists in using another slick feature of SObjectizer: mutable messages.

However, this kind of filtering is inefficient and might result in a significant run-time cost. Indeed, every empty cv::Mat follows all the message handling workflow, only to be thrown out. Although we expect that empty images will be sporadic, a more idiomatic approach exists: delivery filters.

SObjectizer’s agent states are quite sophisticated and provide some utilities that might be useful for developing a working solution. First of all, image_viewer can be modeled as a two-state agent:

An alternative way is using SObjectizer’s timers.

Those who came looking for actor model examples should check out sobjectizer

In sobjectizer the ownership is held by "environment" , while in rotor each thread must held appropriate context, when actor environment is running.

May be actor frameworks like caf, sobjectizer or rotor is something, that you are looking for.

I'm curious what led you to this conclusion. If you ran into scalability issues with its static_thread_pool, then that's a known issue. If it's something else, the authors (of which I'm one) would love to know.

Instead of calling add_executable you will call add_library. Here is a (only moderately complicated) production example of a library that can be built standalone (along with tests and example executables), or as a subproject, where it builds only the library

My original thought was to build my DDS implementation on top of libunifex in anticipation for standardization: https://github.com/facebookexperimental/libunifex

I'm trying to understand how to work with the proposed executors in a project, but after watching Eric Niebler's cppcon talks (https://youtu.be/xLboNIf7BTg) and looking at the libunifex examples (https://github.com/facebookexperimental/libunifex/tree/main/examples) I still have a hard time wrapping my head around how to employ the sender/receiver pattern in a larger project.

You have coroutines in C++20 but there is also the executives proposal that's making it's way into C++23 that is available as a library under the name unifex that only requires C++14

Asio-grpc makes exactly that possible by providing an Asio execution_context compatible interface to the CompletionQueue. It supports all types of RPCs (including generic ones), completion tokens, cancellation, as well as libunifex sender/receiver (if you want to try out what might become std::execution). The latest release (v1.7.0) also introduced a GrpcStream class for writing Rust/Golang select-style code.

P2300: they are trying to standardize facebookexperimental/libunifex

All the network handling for Instagram and all other Meta apps on all platforms is handled by their own C++ library https://github.com/facebookexperimental/libunifex.

This. I'd also like to mention libunifex. It's entirely based on tag_invoke and is a testament as to how much power it actually provides. On the other hand, it also proves how cumbersome it is to define CPOs with tag_invoke. But IMO it's a lot better than anything else anyone has ever created, and users usually don't need to define new CPOs, only library writers do, so there's that.

A work in progress implementation of the library: https://github.com/facebookexperimental/libunifex