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.
> I was recently wondering if it would be possible to detect and deny pointer cycles in a language in an efficient way
In general, I think that cannot be done, but if one restricts what programs can do, solutions exist.
A simple way to do it is by requiring all references “pointing out of” an object to be set the moment the object is created, and be immutable afterwards (that’s what Lisp cons (https://en.wikipedia.org/wiki/Cons) does. Without setf or similar, lisp code cannot create cycles)
That disallows quite a ome code that modifies structures without introducing cycles, but still allows for quite some code to work.
One could also store an ‘age’ field with each object and check, when a reference is updated in an object, that it points to an object that is older than the one being modified. That gives some more leeway, at the price of using more (a lot more, in code using small objects) memory.
Another idea is to add a bit to each object “there are no cycles containing this object”, and have the runtime clear that when it no longer can guarantee that (edit: unfortunately, maintaining that invariant can be very costly. Whenever code does foo.field = bar, with both foo and bar known to be not part of a cycle, you still have to do a search through all objects reachable from bar to check whether a cycle was created and, if so, clear that bit in all objects in the cycle(s). That makes this idea impractical)
If, as I suspect happens in programming languages which are “mostly immutable”, there are many objects for which that flag stays set, that can significantly speed up checking for the creation of cycles.
You can make a programming language where cycles are impossible. Erlang is a prime example.
Region inference is another strategy in this space. It can limit the need for full-blown garbage collection in many cases, but also comes with its own set of added trade-offs.
Reference counting is just a different kind of garbage collection, really. It acts like a dual construction to a tracing GC in many cases. If you start optimizing both, you tend to converge to the same ideas over time. Refcounting isn't void of e.g. latency problems either: if I have a long linked list and snip the last pointer, then we have to collect all of that list. That's going to take O(n) time in the size of the list. For that reason, you'd have to delay collecting the large list right away, which means you are converging toward a tracing GC that can work simultaneously with the mutator. See e.g., Go's garbage collector.
Lexical closures easily give rise to cycles, without the program doing any imperative pointer swizzling to make a cycle explicitly.
For instance, a named lexical function can have itself in scope so that it can call itself recursively. This means that, as an object, it has a pointer to an environment, and that environment has an entry which contains that function itself: cycle.
If you deny cycles, that blows up at the starting line.
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.
I recently re-read this article and can confirm that it's excellent—not just this specific page, but all the other sections under "Garbage Collection" as well.
10 comments
[ 2.9 ms ] story [ 29.9 ms ] threadIt 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.
In general, I think that cannot be done, but if one restricts what programs can do, solutions exist.
A simple way to do it is by requiring all references “pointing out of” an object to be set the moment the object is created, and be immutable afterwards (that’s what Lisp cons (https://en.wikipedia.org/wiki/Cons) does. Without setf or similar, lisp code cannot create cycles)
That disallows quite a ome code that modifies structures without introducing cycles, but still allows for quite some code to work.
One could also store an ‘age’ field with each object and check, when a reference is updated in an object, that it points to an object that is older than the one being modified. That gives some more leeway, at the price of using more (a lot more, in code using small objects) memory.
Another idea is to add a bit to each object “there are no cycles containing this object”, and have the runtime clear that when it no longer can guarantee that (edit: unfortunately, maintaining that invariant can be very costly. Whenever code does foo.field = bar, with both foo and bar known to be not part of a cycle, you still have to do a search through all objects reachable from bar to check whether a cycle was created and, if so, clear that bit in all objects in the cycle(s). That makes this idea impractical)
If, as I suspect happens in programming languages which are “mostly immutable”, there are many objects for which that flag stays set, that can significantly speed up checking for the creation of cycles.
Region inference is another strategy in this space. It can limit the need for full-blown garbage collection in many cases, but also comes with its own set of added trade-offs.
Reference counting is just a different kind of garbage collection, really. It acts like a dual construction to a tracing GC in many cases. If you start optimizing both, you tend to converge to the same ideas over time. Refcounting isn't void of e.g. latency problems either: if I have a long linked list and snip the last pointer, then we have to collect all of that list. That's going to take O(n) time in the size of the list. For that reason, you'd have to delay collecting the large list right away, which means you are converging toward a tracing GC that can work simultaneously with the mutator. See e.g., Go's garbage collector.
For instance, a named lexical function can have itself in scope so that it can call itself recursively. This means that, as an object, it has a pointer to an environment, and that environment has an entry which contains that function itself: cycle.
If you deny cycles, that blows up at the starting line.
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.
> https://github.com/dotnet/AspNetCore.Docs/blob/main/aspnetco...
If you want to dive deeper into memory performance analysis in .NET, this is another must-read: https://github.com/Maoni0/mem-doc/blob/master/doc/.NETMemory...
It was written by Maoni Stephens, the architect of .NET's garbage collection.