By Andreas Brinck

2009-12-11 10:15:40 8 Comments

I haven't used C very much in the last few years. When I read this question today I came across some C syntax which I wasn't familiar with.

Apparently in C99 the following syntax is valid:

void foo(int n) {
    int values[n]; //Declare a variable length array

This seems like a pretty useful feature. Was there ever a discussion about adding it to the C++ standard, and if so, why it was omitted?

Some potential reasons:

  • Hairy for compiler vendors to implement
  • Incompatible with some other part of the standard
  • Functionality can be emulated with other C++ constructs

The C++ standard states that array size must be a constant expression (

Yes, of course I realize that in the toy example one could use std::vector<int> values(m);, but this allocates memory from the heap and not the stack. And if I want a multidimensional array like:

void foo(int x, int y, int z) {
    int values[x][y][z]; // Declare a variable length array

the vector version becomes pretty clumsy:

void foo(int x, int y, int z) {
    vector< vector< vector<int> > > values( /* Really painful expression here. */);

The slices, rows and columns will also potentially be spread all over memory.

Looking at the discussion at comp.std.c++ it's clear that this question is pretty controversial with some very heavyweight names on both sides of the argument. It's certainly not obvious that a std::vector is always a better solution.


@Quuxplusone 2014-02-03 03:01:16

(Background: I have some experience implementing C and C++ compilers.)

Variable-length arrays in C99 were basically a misstep. In order to support VLAs, C99 had to make the following concessions to common sense:

  • sizeof x is no longer always a compile-time constant; the compiler must sometimes generate code to evaluate a sizeof-expression at runtime.

  • Allowing two-dimensional VLAs (int A[x][y]) required a new syntax for declaring functions that take 2D VLAs as parameters: void foo(int n, int A[][*]).

  • Less importantly in the C++ world, but extremely important for C's target audience of embedded-systems programmers, declaring a VLA means chomping an arbitrarily large chunk of your stack. This is a guaranteed stack-overflow and crash. (Anytime you declare int A[n], you're implicitly asserting that you have 2GB of stack to spare. After all, if you know "n is definitely less than 1000 here", then you would just declare int A[1000]. Substituting the 32-bit integer n for 1000 is an admission that you have no idea what the behavior of your program ought to be.)

Okay, so let's move to talking about C++ now. In C++, we have the same strong distinction between "type system" and "value system" that C89 does… but we've really started to rely on it in ways that C has not. For example:

template<typename T> struct S { ... };
int A[n];
S<decltype(A)> s;  // equivalently, S<int[n]> s;

If n weren't a compile-time constant (i.e., if A were of variably modified type), then what on earth would be the type of S? Would S's type also be determined only at runtime?

What about this:

template<typename T> bool myfunc(T& t1, T& t2) { ... };
int A1[n1], A2[n2];
myfunc(A1, A2);

The compiler must generate code for some instantiation of myfunc. What should that code look like? How can we statically generate that code, if we don't know the type of A1 at compile time?

Worse, what if it turns out at runtime that n1 != n2, so that !std::is_same<decltype(A1), decltype(A2)>()? In that case, the call to myfunc shouldn't even compile, because template type deduction should fail! How could we possibly emulate that behavior at runtime?

Basically, C++ is moving in the direction of pushing more and more decisions into compile-time: template code generation, constexpr function evaluation, and so on. Meanwhile, C99 was busy pushing traditionally compile-time decisions (e.g. sizeof) into the runtime. With this in mind, does it really even make sense to expend any effort trying to integrate C99-style VLAs into C++?

As every other answerer has already pointed out, C++ provides lots of heap-allocation mechanisms (std::unique_ptr<int[]> A = new int[n]; or std::vector<int> A(n); being the obvious ones) when you really want to convey the idea "I have no idea how much RAM I might need." And C++ provides a nifty exception-handling model for dealing with the inevitable situation that the amount of RAM you need is greater than the amount of RAM you have. But hopefully this answer gives you a good idea of why C99-style VLAs were not a good fit for C++ — and not really even a good fit for C99. ;)

For more on the topic, see N3810 "Alternatives for Array Extensions", Bjarne Stroustrup's October 2013 paper on VLAs. Bjarne's POV is very different from mine; N3810 focuses more on finding a good C++ish syntax for the things, and on discouraging the use of raw arrays in C++, whereas I focused more on the implications for metaprogramming and the typesystem. I don't know if he considers the metaprogramming/typesystem implications solved, solvable, or merely uninteresting.

A good blog post that hits many of these same points is "Legitimate Use of Variable Length Arrays" (Chris Wellons, 2019-10-27).

@MadScientist 2014-03-25 11:21:02

I agree VLAs were just wrong. The much more widely implemented, and far more useful, alloca() should have been standardized in C99 instead. VLAs are what happens when a standards committee jumps out ahead of implementations, instead of the other way around.

@michaeljt 2014-09-26 12:56:44

I didn't realise that C allowed VLAs as function parameters. I agree (for what that is worth, as I don't write compilers) that that is rather frightening. VLAs as local variables on the other hand do seem sensible to me. Granted you can overflow your stack, but you have to think about that any time you use local variables, so there is nothing really new there. And the "sizeof" thing - isn't that just the price for a feature?

@M.M 2015-03-17 04:52:00

The variably-modified type system is a great addition IMO, and none of your bullet points violate common sense. (1) the C standard does not distinguish between "compile-time" and "run-time" so this is a non-issue; (2) The * is optional, you can (and should) write int A[][n]; (3) You can use the type system without actually declaring any VLAs. For example a function can accept array of variably modified type, and it can be called with non-VLA 2-D arrays of differing dimensions. However you make valid points in the latter part of your post.

@Arvid 2016-04-28 13:41:08

the type argument becomes especially obvious when transitioning to gsl::span<>. a VLA (since g++ supports those as extensions of c++) does not make sense when being passed through a template trying to deduce its size.

@einpoklum 2016-05-10 23:34:31

You know, all of that useful stuff you described would continue to exist even if VLAs were supported - it would just not compile for VLAs, just for fixed-length arrays. Not that I disagree with your VLA reservations, it's just that they're more about use than about having the feature. Doing reinterpret_casts of void * to a function pointer also means unexpected behavior and a possible stack overflow, but that doesn't mean it shouldn't be possible.

@Jeff 2016-05-31 20:54:45

"declaring a VLA means chomping an arbitrarily large chunk of your stack. This is a guaranteed stack-overflow and crash. (Anytime you declare int A[n], you're implicitly asserting that you have 2GB of stack to spare" is empirically false. I just ran a VLA program with a stack far less than 2GB without any stack overflow.

@Quuxplusone 2016-05-31 21:41:20

@Jeff: What was the maximum value of n in your test case, and what was the size of your stack? I suggest you try inputting a value for n at least as large as the size of your stack. (And if there's no way for the user to control the value of n in your program, then I suggest you just propagate the maximum value of n straight into the declaration: declare int A[1000] or whatever it is you need. VLAs are only necessary, and only dangerous, when the maximum value of n isn't bounded by any small compile-time constant.)

@dashesy 2016-07-10 01:45:48

In my experience the main reason I like VLA is that I can index 2 dimensional arrays like arrays a[2][3] and memcpy(&a[1][2], ptr). I never really cared if that memory is on the stack or on the heap. If there was a way to generate a view into some buffer as arbitrary shape matrix and access it like array that would be all that I would want from VLA. That means a class that gets a pointer, and provides operator [] and & just for the syntactic sugar.

@supercat 2017-01-12 16:13:31

@MadScientist: The standard should have included a LIFO-allocation function whose freeing intrinsic requires passing a pointer and size of the earlier allocation, a pair of mark/release intrinsics which may use something larger than a pointer [as with va_list], where the release would clean up any LIFO-allocated objects since the corresponding "mark", and a specification that returning from a function in which things have been LIFO-allocated may free the objects or not, at the implementation's convenience.

@supercat 2017-01-12 16:17:31

@MadScientist: Such semantics could be supported by any platform [unlike those of alloca], and separating out the mark/release semantics allow implementations that do such allocation on the heap to recover storage cleanly after a longjmp.

@MadScientist 2017-01-12 18:59:56

Since alloca() can be implemented using such intrinsics it's by definition true that alloca() could be implemented on any platform, as a compiler standard function. There's no reason that the compiler couldn't detect the first instance of alloca() and arrange for the types of marks and releases to be embedded in the code, and there's no reason a compiler can't implement alloca() using the heap if it can't be done with the stack. What is hard/non-portable is have alloca() implemented on top of a C compiler, so that it works across a wide array of compilers and operating systems.

@supercat 2018-09-13 21:40:07

@MadScientist: The semantics I describe could be supported on any hosted compiler that would allow the addition of a library and header file. While a hosted compiler for any platform could handle alloca() if it were acceptable to have longjmp abandon any stranded made by in functions that are exited thereby, I think it would be better to define semantics that can be supported by existing compilers, and that can coexist with longjmp.

@supercat 2018-09-13 21:58:38

Incidentally, if one wants to use the convention of passing array-size arguments after pointers to teh arrays themselves, that can be done using K&R1 function syntax, but not using "modern" syntax.

@SergeyA 2019-04-10 15:15:16

Well, somehow C++ compilers have solved those problems by allowing VLAs as an extension. You also miss the point regarding difference between VLAs of small size and fixed array of size equal to the maximum VLA size - you'd have to default construct all elements for static array, and you might not be able to!

@Quuxplusone 2019-04-11 13:06:12

@SergeyA: Interesting comment about "you might not be able to [default-construct A]." True: in the one special case where the length of A vla[n] is dynamically determined to be 0, we won't need to call A::A() at all. So, should the compiler require A::A() to exist and be callable? Clang's VLA extension says yes; GCC's extension says "sometimes." (And re "solved those problems," see .)

@Will 2019-04-19 06:01:14

'After all, if you know "n is definitely less than 1000 here", then you would just declare int A[1000].' is just nonsense. If, say, the VLA length is 10 on 99.99% of function invocations and only hits its upper bound of 1000 on 0.01% of invocations, you've just basically wasted 1000 bytes that will never be freed as long as the frame remains on the stack -- which could be almost all the time if the function is high up your control flow hierarchy. You may think 1000 bytes is not much, but then factor in all your cache misses every time your CPU has to move in and out of that function!

@Viktor Sehr 2013-08-13 10:40:51

Seems it will be available in C++14:

Update: It did not make it into C++14.

@Default 2013-08-13 11:12:34

interesting. Herb Sutter discusses it here under Dynamic Arrays:‌​g (this is the reference for the wikipedia information)

@strager 2014-02-24 03:26:19

"Run-time sized arrays and dynarray have been moved to the Array Extensions technical specification" wrote on Wikipedia on 18 January 2014‎:…

@M.M 2014-11-04 02:37:29

Wikipedia isn't a normative reference :) This proposal did not make it into C++14.

@einpoklum 2016-05-10 23:37:00

@ViktorSehr: What's the status of this w.r.t. C++17?

@Viktor Sehr 2017-03-02 15:44:11

@einpoklum No idea, use boost::container::static_vector

@Davis Herring 2019-11-03 21:41:27

@this: No: the Arrays TS was killed in 2016.

@Johannes Schaub - litb 2009-12-11 10:28:54

There recently was a discussion about this kicked off in usenet: Why no VLAs in C++0x.

I agree with those people that seem to agree that having to create a potential large array on the stack, which usually has only little space available, isn't good. The argument is, if you know the size beforehand, you can use a static array. And if you don't know the size beforehand, you will write unsafe code.

C99 VLAs could provide a small benefit of being able to create small arrays without wasting space or calling constructors for unused elements, but they will introduce rather large changes to the type system (you need to be able to specify types depending on runtime values - this does not yet exist in current C++, except for new operator type-specifiers, but they are treated specially, so that the runtime-ness doesn't escape the scope of the new operator).

You can use std::vector, but it is not quite the same, as it uses dynamic memory, and making it use one's own stack-allocator isn't exactly easy (alignment is an issue, too). It also doesn't solve the same problem, because a vector is a resizable container, whereas VLAs are fixed-size. The C++ Dynamic Array proposal is intended to introduce a library based solution, as alternative to a language based VLA. However, it's not going to be part of C++0x, as far as I know.

@Andreas Brinck 2009-12-11 10:46:31

+1 and accepted. One comment though, I think the safety argument is a little bit weak since there are so many other ways to cause stack overflows. The safety argument could be used to support the position that you should never use recursion and that you should allocate all objects from the heap.

@jalf 2009-12-11 10:57:52

So you're saying that because there are other ways to cause stack overflows, we might as well encourage more of them?

@Johannes Schaub - litb 2009-12-11 11:00:14

@Andreas, agreed about the weakness. But for recursion, it takes a huge number of calls until stack is eaten up, and if that can be the case, people would use iteration. As some people on the usenet thread say, though, this is not an argument against VLAs in all cases, since sometimes you definitely may know an upper bound. But in those cases, from what i see a static array can equally be sufficient, since it would not waste much space anyway (if it would, then you would actually have to ask whether the stack area is large enough again).

@Johannes Schaub - litb 2009-12-11 11:03:43

Also look at Matt Austern's answer in that thread: The language specification of VLAs would probably considerably more complex for C++, because of the stricter type matches in C++ (example: C allows assigning a T(*)[] to a T(*)[N] - in C++ this is not allowed, since C++ does not know about "type compatibility" - it requires exact matches), type parameters, exceptions, con- and destructors and stuffs. I'm not sure whether the benefits of VLAs would really pay off all that work. But then, i have never used VLAs in real life, so i probably don't know good use cases for them.

@Steve Jessop 2009-12-11 13:54:44

@Johannes: there are probably some cases where use of a VLA overcomes your objections. For instance a recursive function, at each step i requires N_i amount of space. The only upper limit I have, S, is on the sum of all N_i, not any individual N_i. So I might know that (a) S is small enough for my stack, so VLA's won't overflow, but (b) S * the max depth of recursion is too big for my stack. Of course this doesn't justify a tricky language feature: I should just allocate S on the stack at the top of the recursion, and pass down a pointer to what's left unused.

@Steve Jessop 2009-12-13 14:03:27

Ah, I now see that Kaz Kylheku already gave that example on the com.std.c++ thread. The Windows string example given by Alf afterwards actually seems wrong to me, in that for my taste it puts too much responsibility on the caller of a narrow string function, to ensure that there is enough stack left for the library to use a VLA. VLA's don't do what you actually want in that kind of string-conversion situation, which is to allocate on the stack if there's space and from the heap if there isn't. But I guess on Windows, there's "room" on the stack iff there is on the heap, pretty much.

@Viktor Sehr 2009-12-16 20:46:17

@Johannes: "benefit of being able to create small arrays without wasting space or calling constructors for not used elements". A boost::array-like container which takes a runtime parameter for how many element to initialize of the allocation would be excactly as useful (except for maybe in recursive algorithms). Any runtime sized array ought to have a max number of elements, and allocating this amount of stack space immediately will give you your un-avoidable crash sooner, which is good.

@Johannes Schaub - litb 2009-12-16 21:29:03

@Viktor, indeed. But a maximally sized array will also call all constructors, which might not be what you want. I think the array-like container is a good compromise between changing the language to introduce VLAs (which i believe is a large change), and having no support at all.

@Viktor Sehr 2009-12-16 22:47:41

@Johannes: No, what I mean is; if the array is a wrapper around a primitive type, (ie unsigned char[N * sizeof(T)]) the array can call the placement new/constructor for the number of elements in the runtime size variable. (if N denotes the static maximum size, T the type and n the runtime size, your constructor does something like for(i...n) {new (this + sizof(T)*i) T();}

@Johannes Schaub - litb 2009-12-16 23:52:19

@Viktor, i agree. Such a thing would be useful, and you could provide N as a template parameter as the maximal count of items. And writing it will be easier in C++0x with the various alignment tools available (alignof, etc).

@AHelps 2015-04-10 18:33:25

The main reason I want VLAs is that they serve very well for evaluating polynomials of arbitrary degree--I need the speed of stack allocation, and the value for n will not be very high--and it's usually 2-4, so setting an arbitrary maximum of 20 feels wasteful. I currently do this with a messy template solution, but VLAs would fit much more cleanly.

@supercat 2015-06-03 16:15:39

@AHelps: Perhaps what would be best for that would be a type that behaves somewhat like vector but requires a fixed LIFO usage pattern and maintains one or more per-thread statically-allocated buffers which are generally sized according to the largest total allocation the thread has ever used, but which could be explicitly trimmed. A normal "allocation" would in the common case require nothing more than a pointer copy, pointer-from-pointer subtraction, integer comparison, and pointer addition; de-allocation would simply require a pointer copy. Not much slower than a VLA.

@Johannes Schaub - litb 2015-06-03 17:27:34

BTW, in C++ I recommend folly's or boost's small_vector (…). Folly's version allows you to also forbid it to use the heap. That solves the issues with unneeded constructor calls with a "large enough stack array" approach.

@John Z. Li 2018-09-12 01:17:24

@AndreasBrinck Well said. Isn't c++' way that let programmers decide what to use and use it safely.

@Alan 2014-01-15 08:40:03

I have a solution that actually worked for me. I did not want to allocate memory because of fragmentation on a routine that needed to run many times. The answer is extremely dangerous, so use it at your own risk, but it takes advantage of assembly to reserve space on the stack. My example below uses a character array (obviously other sized variable would require more memory).

void varTest(int iSz)
    char *varArray;
    __asm {
        sub esp, iSz       // Create space on the stack for the variable array here
        mov varArray, esp  // save the end of it to our pointer

    // Use the array called varArray here...  

    __asm {
        add esp, iSz       // Variable array is no longer accessible after this point

The dangers here are many but I'll explain a few: 1. Changing the variable size half way through would kill the stack position 2. Overstepping the array bounds would destroy other variables and possible code 3. This does not work in a 64 bit build... need different assembly for that one (but a macro might solve that problem). 4. Compiler specific (may have trouble moving between compilers). I haven't tried so I really don't know.

@einpoklum 2016-05-10 23:39:45

... and if you want to roll this yourself, maybe use a RAII class?

@Viktor Sehr 2016-05-11 07:51:58

You could simply use boost::container::static_vector thou.

@Ruslan 2016-06-22 13:29:59

This doesn't have equivalents for other compilers which have more raw assembly than MSVC. VC will likely understand that esp changed and will adjust its accesses to stack, but in e.g. GCC you'll just break it completely — at least if you use optimizations and -fomit-frame-pointer in particular.

@philsquared 2009-12-11 10:26:33

This was considered for inclusion in C++/1x, but was dropped (this is a correction to what I said earlier).

It would be less useful in C++ anyway since we already have std::vector to fill this role.

@Kos 2011-08-10 16:34:30

No, we don't, std::vector doesn't allocate data on the stack. :)

@M.M 2014-03-16 21:20:07

"the stack" is an implementation detail; the compiler may allocate memory from anywhere so long as the guarantees about object lifetime are met.

@einpoklum 2016-05-10 23:36:09

@M.M: Fair enough, but in practice we still can't use std::vector instead of, say, alloca().

@M.M 2016-05-10 23:50:09

@einpoklum in terms of getting correct output for your program , you can. Performance is a quality-of-implementation issue

@pal 2019-08-09 13:49:30

@M.M quality-of-implementation is not portable. and if you don't need performance, you don't use c++ in the first place

@Eric 2013-03-21 17:05:05

In my own work, I've realized that every time I've wanted something like variable-length automatic arrays or alloca(), I didn't really care that the memory was physically located on the cpu stack, just that it came from some stack allocator that didn't incur slow trips to the general heap. So I have a per-thread object that owns some memory from which it can push/pop variable sized buffers. On some platforms I allow this to grow via mmu. Other platforms have a fixed size (usually accompanied by a fixed size cpu stack as well because no mmu). One platform I work with (a handheld game console) has precious little cpu stack anyway because it resides in scarce, fast memory.

I'm not saying that pushing variable-sized buffers onto the cpu stack is never needed. Honestly I was surprised back when I discovered this wasn't standard, as it certainly seems like the concept fits into the language well enough. For me though, the requirements "variable size" and "must be physically located on the cpu stack" have never come up together. It's been about speed, so I made my own sort of "parallel stack for data buffers".

@Jingguo Yao 2012-07-31 05:56:39

C99 allows VLA. And it puts some restrictions on how to declare VLA. For details, refer to of the standard. C++ disallows VLA. But g++ allows it.

@Vincent 2017-05-26 11:10:53

Can you provide a link to the standard paragraph that you are pointing ?

@Bengt Gustafsson 2011-01-22 19:33:36

There are situations where allocating heap memory is very expensive compared to the operations performed. An example is matrix math. If you work with smallish matrices say 5 to 10 elements and do a lot of arithmetics the malloc overhead will be really significant. At the same time making the size a compile time constant does seem very wasteful and inflexible.

I think that C++ is so unsafe in itself that the argument to "try to not add more unsafe features" is not very strong. On the other hand, as C++ is arguably the most runtime efficient programming language features which makes it more so are always useful: People who write performance critical programs will to a large extent use C++, and they need as much performance as possible. Moving stuff from heap to stack is one such possibility. Reducing the number of heap blocks is another. Allowing VLAs as object members would one way to achieve this. I'm working on such a suggestion. It is a bit complicated to implement, admittedly, but it seems quite doable.

@Dimitri C. 2009-12-11 10:22:57

Use std::vector for this. For example:

std::vector<int> values;

The memory will be allocated on the heap, but this holds only a small performance drawback. Furthermore, it is wise not to allocate large datablocks on the stack, as it is rather limited in size.

@AHelps 2015-04-10 18:26:44

A major application for variable length arrays is evaluation of arbitrary degree polynomials. In that case, your "small performance drawback" means "the code runs five times slower in typical cases." That's not small.

@L. F. 2019-10-06 13:57:40

Why don't you simply use std::vector<int> values(n);? By using resize after construction you are prohibiting non-moveable types.

@anon 2009-12-11 10:25:33

Arrays like this are part of C99, but not part of standard C++. as others have said, a vector is always a much better solution, which is probably why variable sized arrays are not in the C++ standatrd (or in the proposed C++0x standard).

BTW, for questions on "why" the C++ standard is the way it is, the moderated Usenet newsgroup comp.std.c++ is the place to go to.

@Patrick M 2013-07-11 01:14:42

-1 Vector is not always better. Often, yes. Always, no. If you only need a small array, are on a platform where heap space is slow, and your library's implementation of vector uses heap space, then this feature might very well be better if it existed.

@Edouard A. 2009-12-11 10:26:49

If you know the value at compile time you can do the following:

template <int X>
void foo(void)
   int values[X];


Edit: You can create an a vector that uses a stack allocator (alloca), since the allocator is a template parameter.

@Rob Kennedy 2009-12-11 15:43:26

If you know the value at compile time, you don't need a template at all. Just use X directly in your non-template function.

@Qwertie 2012-07-20 00:50:45

Sometimes the caller knows at compile-time and the callee does not, that's what templates are good for. Of course, in the general case, no one knows X until run-time.

@Oliver 2012-10-16 12:58:47

You can't use alloca in a STL allocator - allocated memory from alloca will be freed when the stack frame is destroyed - that's when the method who should allocate memory returns.

@PfhorSlayer 2009-12-11 10:31:23

You could always use alloca() to allocate memory on the stack at runtime, if you wished:

void foo (int n)
    int *values = (int *)alloca(sizeof(int) * n);

Being allocated on the stack implies that it will automatically be freed when the stack unwinds.

Quick note: As mentioned in the Mac OS X man page for alloca(3), "The alloca() function is machine and compiler dependent; its use is dis-couraged." Just so you know.

@sashoalm 2016-07-01 11:49:17

Also, the scope for alloca() is the entire function, not just the block of code containing the variable. So using it inside of a loop it will continuously increase the stack. A VLA does not have this problem.

@MadScientist 2017-01-12 16:35:05

However, VLAs having the scope of the enclosing block means they are significantly less useful than alloca() with the scope of the entire function. Consider: if (!p) { p = alloca(strlen(foo)+1); strcpy(p, foo); } This cannot be done with VLAs, precisely because of their block scope.

@Adrian W 2018-06-25 15:37:30

That does not answer OP's why question. Moreover, this is a C-like solution, and not really C++-ish.

Related Questions

Sponsored Content

26 Answered Questions

39 Answered Questions

27 Answered Questions

[SOLVED] Why is using "" with array iteration a bad idea?

25 Answered Questions

[SOLVED] Why do we need virtual functions in C++?

10 Answered Questions

18 Answered Questions

[SOLVED] With arrays, why is it the case that a[5] == 5[a]?

4 Answered Questions

[SOLVED] Does C++ support Variable Length Arrays?

3 Answered Questions

[SOLVED] variable-length array and segmentation fault

Sponsored Content