Here is an incomplete list of languages which use 8 bytes for their reference count on 64-bit computers:
1. Rust, in Rc<T> and Arc<T>
2. C++, in std::shared_ptr<T>
3. Objective-C, in NSObject's retainCount
4. Swift, because of Objective-C
5. Python, where reference count is Py_ssize_t
These were literally the first languages i thought of, they all use 64-bit types. I would argue since reference counting is much rarer than GC, these make up the bulk of reference counting use in the real world. "Almost everyone", huh? It's a bit rich accusing the author of being "almost incoherent" and then say this.
> Reference counting can have arbitrarily long pauses as well.
Deallocating can take arbitrarily long, but deallocation does not stop-the-world. It stops the current thread, which is a huge difference. Malloc can take arbitrarily long as well, it's not like it's wait-free.
In addition, the GC pauses in regular programming languages are frequently orders of magnitude longer than deallocation. You would have to deallocate an enormous tree of object at the root for this to be an issue. And GCs have to do that as well, in addition to their regular stop-the-world pauses. This argument is just irrelevant.
> The blog states that atomic writes are "basically free", but that ignores the fact that in multi-threaded code, atomic writes can actually be fairly expensive since each one requires communication between every thread (This is why python still has a GIL)
First off, this is not why Python has a GIL, but lets leave that aside.
Atomic writes are more expensive than non-atomic ones, but they are not slow operations in the grand scheme of things. If you properly implement acquire-release semantics, they are not even that slow under high contention. Compare this to a GC which literally STOPS ALL THREADS, it's nothing.
> you still need a GC anyway to deal with objects that get too many references.
This is just silly. In languages which has both reference counting and traditional garbage collection (e.g. Python), they do it to avoid reference cycles, not because objects get "too many references". That is a ridiculous statement.
In fact! I just realized we do have data for which is more performant: this article describes how Instagram turned of GC for Python and just relied on reference counting. They gained 10% increase in performance:
https://instagram-engineering.com/dismissing-python-garbage-...
I think it's still an open question if reference counting is always faster than GC, but I do not believe you have the technical expertise to evaluate such a claim. Your comment is four paragraphs long and riddled with factual errors. If you want to be convincing, show data that is better than that Instagram case study.