Case Study: FFI Thread Safety

[kassane]

This case to be studied does not affect or influence the progress of the Zig project and its features.
Everything that is or will be tested has an experimental purpose.

References

• Proposal: Event loop redesign ziglang/zig#8224
• Evented I/O introduces multithreading into the control plane, not only the data plane ziglang/zig#5962
• Proposal: Custom event loops ziglang/zig#9922
• Discussing an alternative std.event.Loop architecture ziglang/zig#5311
• A Fix for Aliasing Caused by Result Location Semantics ziglang/zig#12251
• Proposal: Pinned Structs ziglang/zig#7769

Related

• Case Study: C++ Thread-Safety Concurrency tigerbeetle-cpp#4

asio - FFI

• https://github.com/Heerozh/LuaAsio
• Emilua - internal - @vinipsmaker project

Others

• Vale & C - FFI safe
• Asio and Threads
• Strand
• How strands work and why you should use them
• asio on linux with multiple threads io_service->run : performance? chriskohlhoff/asio#886
• [Discussion] Stackful vs stackless coroutines chriskohlhoff/asio#1408 (comment)

[kassane]

There are some threadpool solutions written in Zig, one of which is officially supported in the standard library, but requires memory allocation and another heapless proposal.

Allocator - std

• https://github.com/ziglang/zig/blob/master/lib/std/Thread/Pool.zig

Note: When Observing the std version git history, it can be seen that it is based on the refactored heapless version.

Heapless

• https://github.com/mitchellh/libxev/blob/main/src/ThreadPool.zig
• https://zig.news/kprotty/resource-efficient-thread-pools-with-zig-3291
• https://github.com/mitchellh/libxev/blob/main/docs/xev_threadpool.3.scd

[kassane]

Based on the reference below.

• strand post probabilistic non-response chriskohlhoff/asio#1054

Something similar occurs in the current CI test. The zig test prints Hello task using posted strand however it is not displayed along with threadpool.