Fundamentals of garbage collection (2023)

On the context of .NET runtime, as missing from the title.

Kindof unrelated to the article, but I was recently wondering if it would be possible to detect and deny pointer cycles in a language in an efficient way, so that you could then use simple reference counting instead of full-blown garbage collection.

It probably wouldn't be usable for a general-purpose programming language, but for a special-purpose scripting language I could see it making the language implementation easier.

On any one object you can just follow the references to see if you get back to the same object. Not super efficient as you’d have to do it for each reference as it is set.

But if it was a simple scripting language and you needed that constraint, it’s relativity easy to implement.

One solution is to forbid recursive data types - e.g., require every struct type to only reference types that have already been defined. I can't think of any languages that do this.

Another solution is to make things immutable (like Erlang), or "as-if" immutable (like Koka), which guarantees that data can only point to things that have already been defined, preventing cycles.* Erlang uses this to simplify generational collection - because old data can't point to young data, it doesn't need a card table or anything like that.

I think it's perfectly possible to have a general purpose language without cycles: you can just use integer indices into an array instead of pointers if you want cyclic data structures. This is common in Rust, when people want to avoid the overhead of reference counting, but don't want to use unsafe code.

* A hidden assumption here is that the language is eagerly evaluated. There are languages like Haskell that have immutability and cyclic data structures.

Even with cyclic relationships between types, immutability makes cycles within instances difficult (without laziness anyway). A syntax tree would be a good example.

Yes, either is sufficient, I think.

Edit: I think the common idea with both solutions is that our objects have some weak order (the order in which their types were defined, and the time at which the object was created, respectively), and objects are only allowed to point to objects strictly less than them in this order.

Question: does anyone run "Server GC" for the ASP.NET applications?

There is bunch of people copy pasting documentation to SO "explaining" server GC. I am running bunch of .NET stuff in VMs and never set "Server GC" and never ran into issues with default but also not sure if it is worth testing out.

I guess it does not matter much if you are running in containers but I am running on VMs in IIS.

Yes, and the nice thing about doing it with immutability is you can still have recursive types to build linked lists, trees, and/or dags. From there you can build hash-array-mapped-tries, finger-trees, and so on, giving you friendly dict/list or JSON style data structures.

When I play around with changing various GCs for Java (via Clojure), then I always setup benchmarks measuring what kind of thing I want to improve, run all GCs via that benchmark to chose which to use for that service/project and call it a day. There is a lot of theorizing and navel-gazing around GCs it seems to me, and in the end it is the results that count so setup some way to measure, find the differences then move on from there :)

Server GC is a tradeoff between latency and throughput. It makes a ton of sense for a web server where a small additional overhead of a few milliseconds on some responses won't matter.

Workstation GC is what you want when latency is critical. This is what you'd use if you were developing a UI or game engine.

I've seen workstation GC stay in the microsecond region when strategically executing GC.Collect at allocation batch boundaries.

Hello, I'm writing an implementation of the Common Lisp language that uses an enhanced reference counting algorithm (that I've taken from literature) that detects and handles cycles. Performance seems okay, though I still haven't tried large programs.

https://savannah.nongnu.org/p/alisp

Like Rust if it has no Rc?