Let me link to a (very longwinded, sorry) post I made a while back explaining how a particular type of memory pool allocator works: https://news.ycombinator.com/item?id=6176091

tl;dr if you think "manual memory management" means "malloc thing, use thing, free thing" you're kind of missing the point.

Also, tricks like Duff's device and XOR swap have long since become pessimizations, not optimizations, on desktop and mobile-class hardware. Duff's device, for instance, wastes instruction cache when any decent CPU's branch predictor can eliminate most of a simple loop's overhead anyway. With the exception of manually vectorizing tight loops to take advantage of SIMD hardware (which is more of assembly's domain anyway), I think the "optimized C" of today looks pretty straightforward and unobfuscated, just with more attention paid to the details that higher level languages mostly gloss over for you.

> ...and possibly my generation has missed out on something wonderful...

I'm pretty certain I'm younger than you, I just grew up really wanting to make games and know how computers work. Maybe you don't need or even want to think about this kind of stuff, but it's not like the problems just magically disappeared one day. Or, more bluntly, "if C and C++ are so dumb then why isn't your browser written in Python?"

I think the HN crowd honestly believes that the only reason the next browser hasn't been written in a GC language is momentum and time. That the next one will very obviously be written in Ruby or Python. I've seen discussions here to that effect.

I find it hard to read HN discussions about C. One faction is totally amazed by the C equivalent of being able to tie their shoes. The other faction believes shoelaces are obsolete and everyone should wear velcro.

This "my generation has missed out" thing is really odd... If you are honestly interested in it, go learn how it works! The commenter you're replying to should quit making excuses.

Sorry for the poor phrasing, maybe you've misunderstood: I've mentioned that I'm interested, am fond of it, and have learned how it works. Not sure what excuses you're talking about.

OTOH, for when I'm not writing a browser, a HFTs, or the next-gen video game engine, the trade offs are increasingly not worth it. On top of that we don't have Amigas to have fun with anymore and stay close to the hardware on (I guess maybe Raspberry Pi qualifies as some sort of substitute) and so that particular brand of hacker, a small group to begin with, keeps on shrinking. That's what I meant by missing out.

> Or, more bluntly, "if C and C++ are so dumb then why isn't your browser written in Python?"

Because Python's standard implementation, CPython is an interpreter?

I grew up on assembler, cycle counting, and computer graphics. You're right, a lot of the old-timey optimizations are not applicable on today's hardware, but this is just the march of technology. I didn't have to worry about generating graphics scanline-by-scanline like those poor Atari 2600 programmers. They didn't have to worry about instruction pairing on the Pentium though. Each generation has their tricks.

As for 'manual' management, consider script engines that need to temporarily store the AST while compiling/interpreting a piece of script. These typically allocate each tree node by pointer-bumping from a large pre-allocated block of memory, and free the entire thing once done by resetting the pointer. Basically this is the same idea as a nursery in a generational collector, with the difference that you already know beforehand that you're not going to promote any of the elements.

What GC algorithm is best (mark/sweep, generational, soft real-time, hard real-time) depends on your needs I guess. In that light, it's too bad MSR canned Singularity: http://research.microsoft.com/pubs/69431/osr2007_rethinkings...

> In that light, it's too bad MSR canned Singularity:

Or that few people are aware of Oberon and Spin projects outside the respective universities.

As for Singularity at least parts of the native C# (Sing#) compiler landed on the Windows Phone 8 .NET native compiler (MDIL).

I would really like to know how project Midori relates to Singularity, but it is still kept internal.

Depends. Do you want, for example, a browser that pauses for upwards of 50 ms at a time? How about a game? Go's garbage collector can very much result in 50 ms pauses.

When The Verge reviews new phones, they definitely dock points for a laggy OS, and dropping three frames in a row regularly is considered laggy these days...

There are certainly GCs that don't pause for 50ms at a time -- Go's GC is not very good, and they even admit it.

Felix Klock's PhD is an interesting exploration of this: http://www.ccs.neu.edu/home/pnkfelix/thesis/ although his collector does have some long pauses on particularly evil programs (those programs cause much longer pauses on conventional collectors).

There are research prototypes for such GCs and usually the performance decreases if you want time guarantees. As far as I know there is no GC, which could be used in a AAA FPS game. Convincing would be CryEngine, UnrealEngine, idTech, SourceEngine, etc starting to use a GC in their rendering, physics, sound parts.

John Carmack suggest that someone should write a engine in more functional way (with GC).

So you would reconstruct all the objects given access to the last frame's memory, (which would immutable). Then run the GC between frames. This may cause problems if a slower computer never has time between frames to run GC. (Running at 27fps trying to reach 30fps, my old Laptop would do this most of the time)

The Witcher 2 for the XBox 360 has a GC on their engine, although they did face some issues.

http://www.makinggames.de/index.php/magazin/2155_porting_the...

> There are certainly GCs that don't pause for 50ms at a time -- Go's GC is not very good, and they even admit it.

Of course. I'm just refuting the specific claim of "ballpark C" performance with an immature GC.

So? The fact that there are good garbage collectors doesn't save us from popular languages and runtimes with bad GCs. As long as there are people who expect to be taken seriously when they release a GC'd language with a bad GC, manual memory management should not be treated as an obsolete, obscure, or archaic skill.

You act like most of those phones aren't Java based. Android's GC is actually pretty nice these days.

But all Android phones are still strictly laggier than the iPhone, and part of that is choice of runtime implementation.

Someone has to implement those garbage-collected languages. If you ever want to be one of those someones, you need to learn manual memory management and raw-memory manipulation.

The optimization tricks have merely changed, instead of the whole concept being redundant.

Nowadays you should aspire to reduce branches as much as possible to keep branch prediction happy. Dependent memory accesses should be kept to a minimum. Loops should be written in a way where autovectorizer can easily convert it to a bunch of SIMD instructions. One should understand what the compiler can optimize and write code that allows it to do it's job as efficiently as possible.

The same principles apply as they did in the ages gone by. There is a place for scripting languages and there is a place for heavily optimized library. It's just that the way to write those heavily optimized libraries has changed a bit. Instead of hand writing assembler one just has to write C in a way that allows the compiler to do it for you.

From a web development point of view this means that in general it's good to use <insert the framework that is currently fashionable> and for the parts that require some oomph it's good to have something really optimized like ImageMagick on background.

From mobile perspective it doesn't really matter in what way <It's like Yelp but for old shoes with location awareness> is written but you better hope that the basic system is optimized as much as possible. Because that transfers immediately into battery life.

I was introduced to programming through python. Now that I'm in school, I've been doing most of my work in C++. I have to say, as much as I like programming in python, my knowledge of programming is made so much better because I've had to write C++. From an educational perspective, I think non-GC languages will always have a place (if only for teaching about computers).

In production environments, I don't think that any programming practice should ever need to be "defended." Non-GC languages have their ups and downs, just as GC'ed languages have their ups and downs. Saying "it's indefensible to use one over the other" feels like saying "it's indefensible to fly somewhere instead of driving there." It doesn't make sense, since each is useful for some things and not others.

Once again, when asked the question "which tool is better" my answer is "the one that's best for the job."

I think the parent comment was pointing out that less and less, there's no real reason to need C++. The performance in safer languages is usually far more than adequate, while requiring less code (less bugs) and blocking entire classes of bugs.

Even Tim Sweeny of Epic (Unreal Engine) said they'd gladly switch languages to improve productivity. And game engines are one of the few places that still do need to eek out as much perf as possible.

In short, something like C++ increasingly has more downs than ups. I like C, I just have seen it used in so many places where it makes the code far more verbose, introduces security holes, and has a near-negligible performance benefit.)

"Best tool for the job" is misleading - some tools are nearly completely better than others. Just like globals and goto: they have uses, but less than more.

I once read a comment that said that C++ is like a very sharp knife. If you know how to use it, it's safer than a dull knife; if you don't, then you need to have your totin' chip taken away.

I like to compare it to the F4U Corsair: anyone but a talented and experienced pilot at the yoke will lead to certain disaster. But the thing is just so powerful and so absurdly versatile that until something decisively and significantly better comes along it remains a mainstay of any halfway decent air force.

A lot of people confuse being the Corsair of programming languages with being the Corvair of programming languages -- "unsafe at any speed". They're full of it. Their ignorance and lack of experience prevents them from seeing the scenarios in which the utility of C++ is absolutely indispensable.

That argument seems rather incorrectly biased. The gist is "C++ is awesome, and if you have any problems, it's because you're not awesome enough".

Tim Sweeny's talk[1] indicates that a majority of their bugs are due to low-level "unsafe" code issues. If the people writing Unreal have those problems, I think it's safe to say that it's a problem with the language. Unless you want to make the argument that Epic doesn't hired "talent and experienced" pilots.

C++ sure has places where it might be needed, but those places are growing less every day. And it's more due to the momentum, not the inherent merits of C++. If, say, Microsoft threw their weight behind a language like, say, Rust, the landscape could change. Or if they really pushed on the CLR to make the type system encompass more things and really beef up codegen, we might get somewhere. (The CLR can already easily slide into low level code, even hand-written assembly; the general codegen and limited memory management options are what kills.) Instead, we'll see more things grafted on to C++ (like recent type inference and lambdas) that make C++ somewhat less painful and further reduce the pressure to switch.

1: http://www.st.cs.uni-saarland.de/edu/seminare/2005/advanced-...

CLR seems to be getting some new toys past Visual Studio 2013.

There is the new JIT compiler based on Visual C++ backend and more targets besides Windows Phone 8 will be supported for native code deployment.

With C++ you trade a bit better performance for a truckload of complexity, busy work and unsafety. Whether that trade-off is worth it depends on the application, but the area in which this is a good trade-off is getting smaller and smaller. The analogy with the knife doesn't make much sense. A sharp knife is safer because you have to apply less pressure, and muscles are more precise and controlled when applying less pressure. C++ is the opposite.

No all GC enabled languages are like Python.

Have a look at Oberon for example. A systems programming language with GC, used to program a fully working single user desktop system in the mid 90's.

Used at many universities around the world.

Yes it required the use of Assembly for hardware communication, but so do C and C++.

Given that calls to malloc aren't permitted in big swaths of the code in my current (non-embedded) job, I think it's safe to say that there are still places where GC is not appropriate.

I never claimed otherwise, I enjoy this sort of coding but increasingly GC languages seem to be the pragmatic choice and that only makes usage like yours more precious and interesting.

As that is what I was asking about I'd love for you to elaborate. :)

Sorry for the confusion, I guess I could have worded my post better.

Heh, fair enough. My day job is in HFT.

What you are asking for is essentially areas of computing where you want to get the most out of the hardware. Non-casual games is one such area. Some parts of finance is another.

I'm not sure why you are referring to it as "games". The high-level programming you can enjoy today rely on efficient low-level implementations. You might not see them, but they are still there and still being developed.

"manual malloc/free are increasingly futile unless you're targeting embedded devices"

Or have any algorithm that uses a lot of memory, or if you want control over how much memory the algorithm uses at a time.

In the article (which is great, by the way), you can see that mark-and-sweep means scanning all of the active objects. This does not have great scalability properties. CPU gets faster, but memory gets larger, so matters won't improve over time.

GCs are great for processes that are tiny. As soon as you have a memory-hungry algorithm, it starts looking worse.

Also, as others have pointed out, manual memory management is not synonymous with malloc/free. There are many other choices available, and region-oriented allocators are often nice for algorithms that use a lot of memory. PostgreSQL internally uses "memory contexts" where you allocate objects with the same lifetime (e.g. transaction-local state) in the same context, and wipe it all at once.

For example: It's nice to be able to write a big hunk of state to persistent store and read it back in again with a single operation (like for app pause/resume) vs. iterating through a huge object graph and serializing. Of course it'd be rad if languages could help you not screw this up (maybe Rust does?)

GC doesn't solve all problems, for instance some of the weak reference bugs crop up just as often in Java as they do in Objective-C w/ ARC and ref counting.

No, but it does simplify coding a lot.

In the presence of a GC many algorithms and code refactorings become much easier if you don't have to track all the time who is responsible for releasing a chunk of memory.

Specially in the presence of multicore algorithms and big code bases.

Added to that the fact that in today's industry developers tend to know just a little bit of the code base, given the size of some projects and team's attrition.