In languages with automatic garbage collection like Haskell or Go, how can the garbage collector find out which values stored on the stack are pointers to memory and which are just numbers? If the garbage collector just scans the stack and assumes all addresses to be references to objects, a lot of objects might get incorrectly marked as reachable.
Obviously, one could add a value to the top of each stack frame that described how many of the next values are pointers, but wouldn't that cost a lot of performance?
How is it done in reality?
Some collectors assume everything on the stack is a potential pointer (like Boehm GC). This turns out to be not as bad as one might expect, but is clearly suboptimal. More often in managed languages, some extra tagging information is left with the stack to help the collector figure out where the pointers are.
Remember that in most compiled languages, the layout of a stack frame is the same every time you enter a function, therefore it is not that hard to ensure that you tag your data in the right way.
The "bitmap" approach is one way of doing this. Each bit of the bitmap corresponds to one word on the stack. If the bit is a 1 then the location on the stack is a pointer, and if it is a 0 then the location is just a number from the point of view of the collector (or something along those lines). The exceptionally well written GHC runtime and calling conventions use a one word layout for most functions, such that a few bits communicate the size of the stack frame, with the rest serving as the bitmap. Larger stack frames need a multi word structure, but the idea is the same.
The point is that the overhead is low, since the layout information is computed at compile time, and then included in the stack every time a function is called.
An even simpler approach is "pointer first", where all the pointers are located at the beginning of the stack. You only need to include a length prior to the pointers, or a special "end" word after them, to tell which words are pointers given this layout.
Interestingly, trying to get this management information on to the stack produces a host of problem related to interop with C. For example, it is sub optimal to compile high level languages to C, since even though C is portable, it is hard to carry this kind of information. Optimizing compilers designed for C like languages (GCC,LLVM) may restructure the stack frame, producing problems, so the GHC LLVM backend uses its own "stack" rather than the LLVM stack which costs it some optimizations. Similarly, the boundary between C code, and "managed" code needs to be constructed carefully to keep from confusing the GC.
For this reason, when you create a new thread on the JVM you actually create two stacks (one for Java, one for C).
The Haskell stack uses a single word of memory in each stack frame describing (with a bitmap) which of the values in that stack frame are pointers and which are not. For details, see the "Layout of the stack" article and the "Bitmap layout" article from the GHC Commentary.
To be fair, a single word of memory really isn't much cost, all things considered. You can think of it as just adding a single variable to each method; that's not all that bad.
There exist GCs that assume that every bit pattern that is the address of something the GC is managing is in fact a pointer (and so don't release the something). This can actually work pretty well, because calls pointers are usually bigger than small common integers, and usually have to be aligned. But yes, this can cause collection of some objects to be delayed. The Boehm collector for C works this way, because it's library-based and so don't get any specific help from the compiler.
There are also GCs that are more tightly coupled to the language they're used in, and actually know the structure of the objects in memory. I've never read up specifically in stack frame handling, but you could record information to help the GC if the compiler and GC are designed to work together. One trick would be putting all the pointer references together and using one word per stack frame to record how many there are, which is not such a huge overhead. If you can work out what function corresponds to each stack frame without adding a word saying so, then you could have a per-function "stack frame layout map" compiled in. Another option would be to use tagged words, where you set the low order bit of words that are not pointers to 1, which (due to address alignment) is never needed for pointers, so you can tell them apart. That means you have to shift unboxed values in order to use them though.
It's important to realize that GHC maintains its own stack and does not use the C stack (other than for FFI calls). There's no portable way to access all of the contents of the C stack (for instance, in a SPARC some of it is hidden away in register windows), so GHC maintains a stack where it has full control. Once you maintain your own stack you can pick any scheme to distinguish pointers from non-pointers on the stack (like a using a bitmap).
Related
According to Wikipedia, the translation from lambda calculus to combinatory logic is trivial. Concatenative programming languages can rely solely on a stack for memory allocation.
What's stopping GHC from translating Haskell into a concatenative programming language, such as combinatory logic, and then simply using stack allocation for everything?
Is it feasible to do this translation and thus eliminate garbage collection for languages such as Haskell and OCaml? Are there downsides to doing this?
Suppose I have a function that generates a linked list of some size. The size is the function parameter.
The question is: where do I have to allocate memory for the list?
I can't allocate it on the function's stack, since it's invalid after the function is out. And I can't allocate it on the caller's stack, since I don't know how many memory I need to allocate before the function call. So I need to allocate it on the heap.
I think there may be RAII with manual heap management usable, But I can't see how to eliminate heap allocation at all.
Edit
I can't fit all my thoughts in the comment, so I write them here.
There is no magic about stack-based allocation languages. You still need to know when your data is relevant and remove them when they're not.
Imagine you have a separate stack, and your function has control to push and pop data in it. First, there is no automatic memory management anymore, i.e. the function terminates but the data is not deallocated automatically. Second, if you function allocates some memory, needed to support e.g. the list calculation, then all that stuff will be shuffled with the list that you want to return. No chance you can free unused memory (other lists, trees or so) since you have just push and pop operations. If you have other operations, then what is the difference with the heap?
What about few stacks, not one?
You need to allocate them somewhere, manage their growth and sometimes get them back. That stacks are separate constructions that you need manage by hands. No automatic memory management.
Stack-based languages are ok, but forget about the huge amount of algorithms, that was invented with the concept "get memory from somewhere" and "put the memory back", like hash maps, red-black trees, linked lists. Of course, we can allocate all of those structs on a stack, but we can't free their parts if they don't need anymore.
What about "trivial" lambda calculus translation to Turing machine?
Of course, it is trivial, if you resources are infinite. The math theory clarifies nothing about time and memory complexity of such translated constructions. It just approves that both of that models are equivalent, and all that we can say with Turing machine we can say with lambda calculus, and vice-versa. No guarantees that it can work with real-life limitations.
A concatenative programming language is every bit as capable of running out of memory as a functional programming language.
The fundamental challenge garbage collection addresses is freeing memory that is not, or is not known to be, used in a stack-like fashion. It is most especially useful when there is no clear place in the source code that can be pinpointed as the end of the object's lifetime.
If you simply translate a functional language into a concatenative one with only stack allocation, then you will end up overflowing the stack.
There have definitely been various efforts over the years to reduce the need for garbage collection. One interesting (but very complicated) attempt is the region inference system used in the ML Kit. Unfortunately, that's a bit much for most programmers, including myself, to understand. I believe others have worked on such systems since; I don't know the current state of the art.
The take-away is that some very heavy compiler machinery, along with careful programmer discipline and perhaps special annotations, can sometimes reduce or eliminate the need for garbage collection; no trivial transformation is going to do the trick.
Copying (generational) garbage collection offers the best performance of any form of automatic memory management, but requires pointers to relocated chunks of data be fixed up. This is enabled, in languages which support this memory management technique, by disallowing pointer arithmetic and making sure all pointers are to the beginning of identifiable objects.
If you're generating code at run time with a JIT compiler, things look a bit trickier because return addresses on the call stack will point to, not the beginning of code blocks, but random locations within them, so fixup is a problem.
How is this typically solved?
Quite often, you don't relocate code. This is both because it is indeed complicated to fix the stack and other addresses (think jumps across code fragments), and because you don't actually need garbage collection for such code (as it is only manipulated by code you write anyway, so you can do manual memory management). You also don't expect to create a whole lot of machine code (compared to application objects), so fragmentation etc. is not a concern.
If you insist on moving machine code and fixing up the stack, there is a way, I think: Similar to Mark-Compact, build a "break table" (I have no idea where this name comes from; "relocation table" might be clearer) that tells you the amount by which you should adjust pointers to moved objects. Now, walk the stack for return addresses (highly platform-specific, of course) and fix them if they refer to relocated code. Instead of looking for exact matches, search for the highest address lower than the return address you're currently replacing. You can check that this address indeed refers to some machine code that moved by looking at the object size (you have a pointer to the start of the object, after all). This approach isn't feasible for all objects, for various reasons.
There are other reasons to do something similar though. Some JIT compilers feature on-stack replacement, which means creating a new version (e.g. more optimized, or less optimized) of some machine code and replacing all occurrences of the old version with it. This is far more complicated than just fixing the return addresses though. You have to ensure the new version logically continue where the old one was left hanging. I am not familiar with how this is implemented, so I will not go into detail.
Tagged pointers are a common optimization when implementing dynamic languages: take advantage of alignment requirements that mean the low two or three bits of a pointer will always be zero, and use them to store type information.
Suppose you're using the Boehm garbage collector, which basically works by looking at active data for things that look like pointers. Tagged pointers don't look like pointers, in the sense that their low bits are nonzero.
Is this a showstopper, i.e. do you have to ditch tagged pointers if you're using Boehm? Or does it have a way around this problem?
AFAIK Boehm can handle this with the right options. It is capable, at a small price, of detecting interior pointers. It is also possible to write your own scanning code. Basically there are probably enough hooks to handle just about anything.
I have written my own collector, it is precise on the heap and conservative on the stack. It does not touch C made pointers. For some applications it will be faster because it knows a lot about my language allocated objects and doesn't care about other stuff which is managed, say, using traditional C++ destructors.
However it isn't incremental or generational, and it doesn't handle threads as well (it's not smart enough to stop threads with signals). On the plus side, however, it doesn't require magic linkage techniques which Boehm does (to capture mallocs, etc). On the seriously minus side you can't put managed objects into unmanaged ones.
When writing production-quality VC++ code, is the use of recursion acceptable? Why or why not?
Is there a way to determine at what point I would encounter a stack overflow?
Not really. A stack overflow happens when you exhaust the stack space - however...
The initial stack size can be changed programatically and may default to different amounts depending on your OS/compiler/etc
How much of it is already used up depends on what your app (and the libraries your app uses) has previously done - this is often impossible to predict
How much of the stack each call requires depends on what you do in your function. If you only allocate say 1 integer on the stack, you may be able to recurse an enourmous amount of times, but if you are allocating a 200k buffer on the stack, not so much.
The only times I've ever hit one is in an infinite loop, or using the aforementioned 200k buffer.
I find it far more prefereable for my app to just crash, than for it to loop forever using 100% CPU and have to be forcefully killed (this is a right PITA on a remote server over a bad connection as windows lacks SSH)
A rough guideline: Do you think your recursive function is likely to call itself more than say 10,000 times consecutively? Or are you doing something dumb like allocating 200k buffers on the stack?
If yes, worry about it.
If no, carry on with more important things.
Yes. But never in dead code. That would be silly.
Sure - e.g. if you want to traverse a tree structure what else would you use ?
Maybe you would like to have something like a maximum depth to be sure you're not writing an infinite loop. (if this makes sense in your example)
Is there a way to determine at what
point I would encounter a stack
overflow?
Depends how deep you go, and how large the actual recursion is. I take it you understand what recursion does?
Recursion is almost essential to traverse File structures like folder/directories.
Traversing a tree like structure is very easy if recursion is used.
I'm designing a language. First, I want to decide what code to generate. The language will have lexical closures and prototype based inheritance similar to javascript. But I'm not a fan of gc and try to avoid as much as possible. So the question: Is there an elegant way to implement closures without resorting to allocate the stack frame on the heap and leave it to garbage collector?
My first thoughts:
Use reference counting and garbage collect the cycles (I don't really like this)
Use spaghetti stack (looks very inefficient)
Limit forming of closures to some contexts such a way that, I can get away with a return address stack and a locals' stack.
I won't use a high level language or follow any call conventions, so I can smash the stack as much as I like.
(Edit: I know reference counting is a form of garbage collection but I am using gc in its more common meaning)
This would be a better question if you can explain what you're trying to avoid by not using GC. As I'm sure you're aware, most languages that provide lexical closures allocate them on the heap and allow them to retain references to variable bindings in the activation record that created them.
The only alternative to that approach that I'm aware of is what gcc uses for nested functions: create a trampoline for the function and allocate it on the stack. But as the gcc manual says:
If you try to call the nested function through its address after the containing function has exited, all hell will break loose. If you try to call it after a containing scope level has exited, and if it refers to some of the variables that are no longer in scope, you may be lucky, but it's not wise to take the risk. If, however, the nested function does not refer to anything that has gone out of scope, you should be safe.
Short version is, you have three main choices:
allocate closures on the stack, and don't allow their use after their containing function exits.
allocate closures on the heap, and use garbage collection of some kind.
do original research, maybe starting from the region stuff that ML, Cyclone, etc. have.
This thread might help, although some of the answers here reflect answers there already.
One poster makes a good point:
It seems that you want garbage collection for closures
"in the absence of true garbage collection". Note that
closures can be used to implement cons cells. So your question
seem to be about garbage collection "in the absence of true
garbage collection" -- there is rich related literature.
Restricting problem to closures does not really change it.
So the answer is: no, there is no elegant way to have closures and no real GC.
The best you can do is some hacking to restrict your closures to a particular type of closure. All this is needless if you have a proper GC.
So, my question reflects some of the other ones here - why do you not want to implement GC? A simple mark+sweep or stop+copy takes about 2-300 lines of (Scheme) code, and isn't really that bad in terms of programming effort. In terms of making your programs slower:
You can implement a more complex GC which has better performance.
Just think of all the memory leaks programs in your language won't suffer from.
Coding with a GC available is a blessing. (Think C#, Java, Python, Perl, etc... vs. C++ or C).
I understand that I'm very late, but I stumbled upon this question by accident.
I believe that full support of closures indeed requires GC, but in some special cases stack allocation is safe. Determining these special cases requires some escape analysis. I suggest that you take a look at the BitC language papers, such as Closure Implementation in BitC. (Although I doubt whether the papers reflect the current plans.) The designers of BitC had the same problem you do. They decided to implement a special non-collecting mode for the compiler, which denies all closures that might escape. If turned on, it will restrict the language significantly. However, the feature is not implemented yet.
I'd advise you to use a collector - it's the most elegant way. You should also consider that a well-built garbage collector allocates memory faster than malloc does. The BitC folks really do value performance and they still think that GC is fine even for the most parts of their operating system, Coyotos. You can migitate the downsides by simple means:
create only a minimal amount of garbage
let the programmer control the collector
optimize stack/heap use by escape analysis
use an incremental or concurrent collector
if somehow possible, divide the heap like Erlang does
Many fear garbage collectors because of their experiences with Java. Java has a fantastic collector, but applications written in Java have performance problems because of the sheer amount of garbage generated. In addition, a bloated runtime and fancy JIT compilation is not really a good idea for desktop applications because of the longer startup and response times.
The C++ 0x spec defines lambdas without garbage collection. In short, the spec allows non-deterministic behavior in cases where the lambda closure contains references which are no longer valid. For example (pseudo-syntax):
(int)=>int create_lambda(int a)
{
return { (int x) => x + a }
}
create_lambda(5)(4) // undefined result
The lambda in this example refers to a variable (a) which is allocated on the stack. However, that stack frame has been popped and is not necessarily available once the function returns. In this case, it would probably work and return 9 as a result (assuming sane compiler semantics), but there is no way to guarantee it.
If you are avoiding garbage collection, then I'm assuming that you also allow explicit heap vs. stack allocation and (probably) pointers. If that is the case, then you can do like C++ and just assume that developers using your language will be smart enough to spot the problem cases with lambdas and copy to the heap explicitly (just like you would if you were returning a value synthesized within a function).
Use reference counting and garbage collect the cycles (I don't really like this)
It's possible to design your language so there are no cycles: if you can only make new objects and not mutate old ones, and if making an object can't make a cycle, then cycles never appear. Erlang works essentially this way, though in practice it does use GC.
If you have the machinery for a precise copying GC, you could allocate on the stack initially and copy to the heap and update pointers if you discover at exit that a pointer to this stack frame has escaped. That way you only pay if you actually do capture a closure that includes this stack frame. Whether this helps or hurts depends on how often you use closures and how much they capture.
You might also look into C++0x's approach (N1968), though as one might expect from C++ it consists of counting on the programmer to specify what gets copied and what gets referenced, and if you get it wrong you just get invalid accesses.
Or just don't do GC at all. There can be situations where it's better to just forget the memory leak and let the process clean up after it when it's done.
Depending on your qualms about GC, you might be afraid of the periodic GC sweeps. In this case you could do a selective GC when an item falls out of scope or the pointer changes. I'm not sure how expensive this would be though.
#Allen
What good is a closure if you can't use them when the containing function exits? From what I understand that's the whole point of closures.
You could work with the assumption that all closures will be called eventually and exactly one time. Now, when the closure is called you can do the cleanup at the closure return.
How do you plan on dealing with returning objects? They have to be cleaned up at some point, which is the exact same problem with closures.
So the question: Is there an elegant way to implement closures without resorting to allocate the stack frame on the heap and leave it to garbage collector?
GC is the only solution for the general case.
Better late than never?
You might find this interesting: Differential Execution.
It's a little-known control stucture, and its primary use is in programming user interfaces, including ones that can change dynamically while in use. It is a significant alternative to the Model-View-Controller paradigm.
I mention it because one might think that such code would rely heavily on closures and garbage-collection, but a side effect of the control structure is that it eliminates both of those, at least in the UI code.
Create multiple stacks?
I've read that the last versions of ML use GC only sparingly
I guess if the process is very short, which means it cannot use much memory, then GC is unnecessary. The situation is analogous to worrying about stack overflow. Don't nest too deeply, and you cannot overflow; don't run too long, and you cannot need the GC. Cleaning up becomes a matter of simply reclaiming the large region that you pre-allocated. Even a longer process can be divided into smaller processes that have their own heaps pre-allocated. This would work well with event handlers, for example. It does not work well, if you are writing compiler; in that case, a GC is surely not much of a handicap.