Hacker News Reader

Comment by WildUtah | Hacker News Reader

a lot of implementation leverage (e.g. garbage collection)

Putting garbage collection in that basket is controversial and lately the evidence is piling up against it.

Notice how new important growing languages like Swift and Rust don't have garbage collection and programmers don't miss it. GC is more and more restricted to low performance languages like Python, Javascript, and Ruby and a few specialized highly tuned and expensively engineered exceptions like the JVM and obscure LISP engines.

And if you've ever tried to program responsive, interactive, predictable, or time-critical systems, you know that GC is ruinous and awful for those applications. Meanwhile languages like Rust, Swift, and C++ can be used for everything from systems programming to games to low power mobile apps to scripting.

View on HN

Swift does have garbage collection! It uses reference counting, and AFAIK it doesn't have cycle detection, which means it's up to the developer to ensure that no cycles occur.

Regarding responsive interfaces: pure reference counting is particularly problematic for this, because an unbounded amount of work may happen every time an object's reference is decremented (because the object might be freed, which causes other objects' references to become zero, etc). If this happens inside a critical section Your Life Is Pain. At least you can control when this happens by taking and then dropping an additional reference. (See Obj-C's allocation pools.)

C++'s shared_ptr<>, BTW, is also reference counting, and suffers from the same problem. So it's technically correct (which is the best kind of correct) to say that C++ has garbage collection too.

Reference counting isn't garbage collection. It's the opposite of garbage collection. With reference counting, which is optional in C++ and can be opted out of in Swift, you keep track of your allocations and release them when you're done with them. With garbage collection an outside program decides when you're done with memory and deals with it for you on its terms.

It's true that there can be unlimited work done in reference counting. Of course, it's true there can be unlimited work in any function of any program. [0] With reference counting a programmer can fix the amount of work to be done and when in deterministic fashion. With garbage collection, there is no such power. The gc decides when it will work and you cannot control or predict the interruption.

You cannot produce serious interactive or time sensitive control software like that. You certainly can't write system software like that.

[0] https://en.wikipedia.org/wiki/Halting_problem

> Reference counting isn't garbage collection.

"Reference counting is a form of garbage collection"[1]

[1] https://en.wikipedia.org/wiki/Garbage_collection_(computer_s...

If the runtime mechanically detects that an object will never be used again, without you having to explicitly say so (and woe unto you if you're mistaken), that's what's meant by garbage collection. Reference counting does have a deterministic cost, but it's extremely high and applies to everything you do. Copying collection benefits from deferring and batching the work when most objects die quickly and can be ignored with zero effort (just recycle the arena containing all of them).

You can write a reference counter in three short lines of code that works the same as shared_ptr<T> or Swift. A modern automatic garbage collector requires a threading library plus tens of thousands of lines of intricate code. We're not comparing like with like here.

The three line version is really easy to write. And every time you assign to a pointer, it requires uncached atomic updates to the refcounts of the old and new objects, which is staggeringly expensive compared to a block of code that can go unused for minutes at a time. And all it gets you is learning that a small piece of memory is available slightly sooner, but you'll probably want to coalesce your free memory anyway for better locality.

I hope that the 100-man years of top engineers version has some benefits over the 3-line newbie approach.

People tend to think that refcounting must be faster than a tracing GC because you don't have to scan the memory. The second half of the previous sentence is true, but what they forget is that refcounting needs to do a lot of things that a GC doesn't have to:

Like you said, need to update the reference every time it's moved.

malloc becomes much slower, compared to a GC language (with a compacting GC) where malloc is effectively just a single add instruction.

free also needs to do a lot of bookkeeping while the GC language doesn't need to do anything.

Multithreading introduces another set of issues when dealing with refcounting that is not a problem with a GC.

Multithreading also introduces issues for GC. It needs to stop all threads. If you look at how much work has been delivered towards the GC's of for example JVM, Go and V8. It is a really tough problem to get right for all application types.

Reference counting is very predictable. And if you see a bottle neck you can start fixing it.

And with ARC (Swift/ObjC) you get even compiler support to optimize all ref counting administration.

Further malloc doesn't have anything to do with ref counting as ref counting sits on top if it.

Further, GC languages also need to do book keeping: Generations and Compaction.

That is true, but only during GC itself. Refcounting causes these issues every time the reference is copied.

For malloc, the benefit with a compacting GC is that the free memory will always be in a contiguous segment. Therefore, a malloc operating is similar to a stack allocation in that all you need to do is to move a single pointer. Without a compacting GC, you need to search a tree whenever you want to allocate memory.

In order to know whether an arena is empty, you need to have walked, at a minimum, each object that was in the arena. That applies for copying collections too as during the copy phase you've walked and inspected each object that might be copied into the arena at least once. Arenas do not reduce the cost of garbage collection to O(1). And in any event arenas are not solely the preserve of sweeping or sweeping+copying collectors.

The algorithmic complexity of a sweeping collector is identical to that of a reference counting collector that can also break cycles. See A Unified Theory of Garbage Collection at http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.146....

A reference counting collector that _cannot_ break cycles (like in Swift) might have lower algorithmic complexity. I'm not sure.

The biggest bone of contention with reference counting is the constant adjustment of the reference count of objects on the stack. OTOH, the bone of contention with sweeping collectors is constantly walking all extant objects, including ones not directly on the stack.

There are various optimizations you can make to one or the other or both; many arguments about how real-world scenarios tax one or the other model; and many arguments about which is easier to optimize. But at the end of the day reference counting is no worse in terms of algorithmic complexity than sweeping. What matters most will be the quality of implementation. Is the compiler smart enough to know that an object passed 3 routines deep doesn't really escape the stack scope? If it can do that, it can elide all the cost of either approach.

What matters most will be the quality of implementation. Is the compiler smart enough to know that an object passed 3 routines deep doesn't really escape the stack scope? If it can do that, it can elide all the cost of either approach.

This is the plan with Swift and what makes it zippy.

bone of contention with sweeping collectors is constantly walking

The real problem is that you cannot determine when the program will pause. It makes interactive and soft realtime impossible with gc. (Hard realtime should not run a generic allocator at all.) You can't write anything but servers and unpleasantly unresponsive apps.

Android developers have to carefully jump through hoops to avoid allocating any RAM at all if they don't want to bug their users. For years when the chips were slow, ref counting was a major competitive advantage for iPhone. A hundred-billion dollar win.

And then there's system software. You can't afford a live, unpredictable runtime messing with you in systems code so anything that touches low level code can't use gc. There's a disincentive to really invest in a language that simply can't ever do a full stack, even if you're unlikely to need to write a device driver anytime soon. That's why no major browsers, performant ACID databases, or fast app servers depend on RAM gc.

Thanks for the paper touching on reference counting work queues, I hadn't paid much attention to that approach since proponents tend to focus on instantly reclaiming each dead object. I still find semi-space copying (move O(live) objects and ignore O(dead) objects where dead >> live) to be not just one alternative but an enormous win I'd almost never forego, since buying more memory is so much easier than buying more CPU and bus throughput and has been for years.

I have trouble believing that behind-the-scenes escape analysis is very effective on real-world code. Look at how much painfully explicit source annotation Rust found they needed.

> Reference counting isn't garbage collection.

Reference books in computer science, acknowledged by the experts in the field, happen to disagree.

http://gchandbook.org/

How about: Reference counting is one form of garbage collection, but "Garbage Collection" without any additional qualifiers or context usually implies "tracing garbage collection", with or without the aid of other GC algorithms.

In common usage, I don't hear of languages like C++ described as a "garbage collecting language". That usually describes Java, Python, etc.

C++ has a GC API and a few dialects that support GC like C++/CLI.

Having RC library types doesn't make a language GC, because their correct use is not enforced by language semantics.

I could say this in response to a dozen posts here, but arguing over the strict definition of the term is largely beside the point, which is the actual capabilities of the various languages. In fact, the argument itself shows that the distinction is no longer either as straightforward or useful as it once was, making the dichotomy moot.

Well, those that review CS papers about GC algorithms to ACM, OOPSLA and SIGPLAN, think differently.

Elsewhere in this discussion:

hellofunk: Then C++ is also garbage collected.

yourself: Kind of, but only because C++11 defined a pluggable GC ABI.

This is the sort of discussion that is far more fruitful than a pedantic argument over the precise meaning of a term of art in a field that is rapidly advancing. And if the academic community really is obsessed with this, then that is unfortunate.

They're duals, actually, and most practical GC algorithms used in production are hybrids of the two:

http://www.cs.virginia.edu/~cs415/reading/bacon-garbage.pdf

They're duals, actually

No, actually, they're not. Just try allocating a lot of looped structure RAM and see.

I'm surprised by this heated debate regarding GC in this thread. Are you arguing that scripting languages that use reference counting are not garbage collected? PHP used reference counting for a long time (cycle detection only made it into the language in version 5.3 if I trust Wikipedia), it feels odd to consider PHP as not being garbage collected.

For me the underlying technology is not very relevant, what matters is "do I have to worry about exactly when the resource is going to be dropped". I use the resource and when I stop using it it'll probably be deleted soon-ish. At any rate it's not really my problem (at least in a perfect world).

You say "With reference counting [...] you keep track of your allocations and release them when you're done with them" but that's not really true, I don't keep track of anything, the language or library does. I can't really tell by looking at some code in isolation when or how the resource is destroyed. The GC takes care of that.

Some languages like python, Java, Go or PHP have 1st party garbage collection that's active by default and other languages like C, C++, Rust and friends have optional 3rd party GC implemented as libraries (reference counting or otherwise).

I really don't see what's so controversial about this.

In C, C++, Rust, Swift, &c I can determine how and when references go into or out of scope and get freed.

The defining characteristic of a garbage collected system is an active outside process with priority and control over my code that chooses when to inject bugs (to wit: arbitrary pauses) into my program outside my control.

PHP is a bad example, it was designed around the idea of the OS doing the cleanup for you when the process exited at the end of the request.

When people say garbage collection they usually mean a tracing GC.

You can twist words and say that shared_ptr means C++ has 'garbage collection' but I doubt anyone would agree with you if were were to word it like:

"C++ has a garbage collector"

Ref counting is a form of 'garbage collection' in so far as it is a way to release memory, that's about it. I certainly wouldn't consider Rust's Arc<T> a 'opt in garbage collector'.

> When people say garbage collection they usually mean a tracing GC.

Wikipedia lists both tracing and reference counting as forms of GC... hell, when interviewing at Microsoft I was asked to describe the two primary families of GC, expecting tracing and ref counting.

Anyone can define a bat as a bird because it's an animal with wings that flies. But you'll soon find that isn't a useful conceptual distinction.

Well, you are free to disagree with the computer science experts in the field of language design, but it won't make it less true.

More like, you're free to ignore the common-sense suggestion to examine the algorithms used as they pertain to your use case, just because someone once said "In theory these two things are equivalent". But it won't help you once you hit the inevitable edge case.

Terminology is useful right up until it isn't. Resting on definitions rather than seeking proper descriptions is a great way to fail spectacularly.

Not someone, the main language researchers in the field of memory management.

Not a random dude on Internet.

You're doing exactly what I'm talking about, you're trying to show how the authority you consider authoritative is right, not trying to understand how in a very real sense, reference counting also is not garbage collection because it lacks some critical feature from someone else's point of view.

If you're right, you're just some guy referencing an authority. Yay you. If you're wrong, you're totally ... all over the point of what someone else is saying. Instead of trying to force us to see how you're right, why not try to see how someone else isn't wrong?

It is not me that is right or wrong, it is the top researchers in the field of programming languages.

The papers of computer science, written since the field exists, speak for themselves. One just has to read them.

I agree with you, when people start trying to make these sorts of arguments I like to link the following video from Richard Feynman.

https://www.youtube.com/watch?v=45trtAfWhYg

The problem is that "garbage collection" as a term seems to have become synonymous with "tracing GC" (or, in other circles, "whatever the JVM does"). I've started using the term "dynamic lifetime determination" to refer to both tracing garbage collection and reference counting (with our without cycle detection, though the distinction matters). The question then is how pervasive a given language makes dynamic lifetime determination. In Java (GC) and Python (RC with CC), it's ubiquitous. In Go (GC with value types) and Swift (RC without CC), it's pervasive. In Rust and C++, it's selective (note that a large C++ codebase with shared_ptr everywhere would, effectively, be garbage-collected (to say nothing of e.g. Boehm)).

In contrast, C's automatic storage duration for stack-allocated items ( http://en.cppreference.com/w/cpp/language/storage_duration#S... ) and Rust's system of ownership for heap-allocated items are things that I'd consider "static lifetime determination".

> The problem is that "garbage collection" as a term seems to have become synonymous with "tracing GC"

The term has been used that way since I first heard it back in the '80s, so it's probably not a battle worth fighting at this point.

Yes, of course, there's a formal sense under which reference counting is just an outlying position within the broader world of garbage collection, but from an engineering point of view, it's a significantly different set of tradeoffs, and reference counting doesn't really qualify as the same thing when you're deciding how to structure your program.

> Notice how new important growing languages like Swift and Rust don't have garbage collection

Swift is using reference counting which is a form of garbage collection.

This is not what people are talking about when they say it doesn't have a garbage collector.

Python mostly uses reference counting, but has a garbage collector which periodically checks for cycles. It's even possible to disable the garbage collector.

Reply on news.ycombinator.com