Related
I see in C++ there are multiple ways to allocate and free data and I understand that when you call malloc you should call free and when you use the new operator you should pair with delete and it is a mistake to mix the two (e.g. Calling free() on something that was created with the new operator), but I'm not clear on when I should use malloc/ free and when I should use new/ delete in my real world programs.
If you're a C++ expert, please let me know any rules of thumb or conventions you follow in this regard.
Unless you are forced to use C, you should never use malloc. Always use new.
If you need a big chunk of data just do something like:
char *pBuffer = new char[1024];
Be careful though this is not correct:
//This is incorrect - may delete only one element, may corrupt the heap, or worse...
delete pBuffer;
Instead you should do this when deleting an array of data:
//This deletes all items in the array
delete[] pBuffer;
The new keyword is the C++ way of doing it, and it will ensure that your type will have its constructor called. The new keyword is also more type-safe whereas malloc is not type-safe at all.
The only way I could think that would be beneficial to use malloc would be if you needed to change the size of your buffer of data. The new keyword does not have an analogous way like realloc. The realloc function might be able to extend the size of a chunk of memory for you more efficiently.
It is worth mentioning that you cannot mix new/free and malloc/delete.
Note: Some answers in this question are invalid.
int* p_scalar = new int(5); // Does not create 5 elements, but initializes to 5
int* p_array = new int[5]; // Creates 5 elements
The short answer is: don't use malloc for C++ without a really good reason for doing so. malloc has a number of deficiencies when used with C++, which new was defined to overcome.
Deficiencies fixed by new for C++ code
malloc is not typesafe in any meaningful way. In C++ you are required to cast the return from void*. This potentially introduces a lot of problems:
#include <stdlib.h>
struct foo {
double d[5];
};
int main() {
foo *f1 = malloc(1); // error, no cast
foo *f2 = static_cast<foo*>(malloc(sizeof(foo)));
foo *f3 = static_cast<foo*>(malloc(1)); // No error, bad
}
It's worse than that though. If the type in question is POD (plain old data) then you can semi-sensibly use malloc to allocate memory for it, as f2 does in the first example.
It's not so obvious though if a type is POD. The fact that it's possible for a given type to change from POD to non-POD with no resulting compiler error and potentially very hard to debug problems is a significant factor. For example if someone (possibly another programmer, during maintenance, much later on were to make a change that caused foo to no longer be POD then no obvious error would appear at compile time as you'd hope, e.g.:
struct foo {
double d[5];
virtual ~foo() { }
};
would make the malloc of f2 also become bad, without any obvious diagnostics. The example here is trivial, but it's possible to accidentally introduce non-PODness much further away (e.g. in a base class, by adding a non-POD member). If you have C++11/boost you can use is_pod to check that this assumption is correct and produce an error if it's not:
#include <type_traits>
#include <stdlib.h>
foo *safe_foo_malloc() {
static_assert(std::is_pod<foo>::value, "foo must be POD");
return static_cast<foo*>(malloc(sizeof(foo)));
}
Although boost is unable to determine if a type is POD without C++11 or some other compiler extensions.
malloc returns NULL if allocation fails. new will throw std::bad_alloc. The behaviour of later using a NULL pointer is undefined. An exception has clean semantics when it is thrown and it is thrown from the source of the error. Wrapping malloc with an appropriate test at every call seems tedious and error prone. (You only have to forget once to undo all that good work). An exception can be allowed to propagate to a level where a caller is able to sensibly process it, where as NULL is much harder to pass back meaningfully. We could extend our safe_foo_malloc function to throw an exception or exit the program or call some handler:
#include <type_traits>
#include <stdlib.h>
void my_malloc_failed_handler();
foo *safe_foo_malloc() {
static_assert(std::is_pod<foo>::value, "foo must be POD");
foo *mem = static_cast<foo*>(malloc(sizeof(foo)));
if (!mem) {
my_malloc_failed_handler();
// or throw ...
}
return mem;
}
Fundamentally malloc is a C feature and new is a C++ feature. As a result malloc does not play nicely with constructors, it only looks at allocating a chunk of bytes. We could extend our safe_foo_malloc further to use placement new:
#include <stdlib.h>
#include <new>
void my_malloc_failed_handler();
foo *safe_foo_malloc() {
void *mem = malloc(sizeof(foo));
if (!mem) {
my_malloc_failed_handler();
// or throw ...
}
return new (mem)foo();
}
Our safe_foo_malloc function isn't very generic - ideally we'd want something that can handle any type, not just foo. We can achieve this with templates and variadic templates for non-default constructors:
#include <functional>
#include <new>
#include <stdlib.h>
void my_malloc_failed_handler();
template <typename T>
struct alloc {
template <typename ...Args>
static T *safe_malloc(Args&&... args) {
void *mem = malloc(sizeof(T));
if (!mem) {
my_malloc_failed_handler();
// or throw ...
}
return new (mem)T(std::forward(args)...);
}
};
Now though in fixing all the issues we identified so far we've practically reinvented the default new operator. If you're going to use malloc and placement new then you might as well just use new to begin with!
From the C++ FQA Lite:
[16.4] Why should I use new instead of
trustworthy old malloc()?
FAQ: new/delete call the
constructor/destructor; new is type
safe, malloc is not; new can be
overridden by a class.
FQA: The virtues of new mentioned by
the FAQ are not virtues, because
constructors, destructors, and
operator overloading are garbage (see
what happens when you have no garbage
collection?), and the type safety
issue is really tiny here (normally
you have to cast the void* returned by
malloc to the right pointer type to
assign it to a typed pointer variable,
which may be annoying, but far from
"unsafe").
Oh, and using trustworthy old malloc
makes it possible to use the equally
trustworthy & old realloc. Too bad we
don't have a shiny new operator renew or something.
Still, new is not bad enough to
justify a deviation from the common
style used throughout a language, even
when the language is C++. In
particular, classes with non-trivial
constructors will misbehave in fatal
ways if you simply malloc the objects.
So why not use new throughout the
code? People rarely overload operator
new, so it probably won't get in your
way too much. And if they do overload
new, you can always ask them to stop.
Sorry, I just couldn't resist. :)
Always use new in C++. If you need a block of untyped memory, you can use operator new directly:
void *p = operator new(size);
...
operator delete(p);
new vs malloc()
1) new is an operator, while malloc() is a function.
2) new calls constructors, while malloc() does not.
3) new returns exact data type, while malloc() returns void *.
4) new never returns a NULL (will throw on failure) while malloc() returns NULL
5) Reallocation of memory not handled by new while malloc() can
To answer your question, you should know the difference between malloc and new. The difference is simple:
malloc allocates memory, while new allocates memory AND calls the constructor of the object you're allocating memory for.
So, unless you're restricted to C, you should never use malloc, especially when dealing with C++ objects. That would be a recipe for breaking your program.
Also the difference between free and delete is quite the same. The difference is that delete will call the destructor of your object in addition to freeing memory.
Use malloc and free only for allocating memory that is going to be managed by c-centric libraries and APIs. Use new and delete (and the [] variants) for everything that you control.
There is one big difference between malloc and new. malloc allocates memory. This is fine for C, because in C, a lump of memory is an object.
In C++, if you're not dealing with POD types (which are similar to C types) you must call a constructor on a memory location to actually have an object there. Non-POD types are very common in C++, as many C++ features make an object automatically non-POD.
new allocates memory and creates an object on that memory location. For non-POD types this means calling a constructor.
If you do something like this:
non_pod_type* p = (non_pod_type*) malloc(sizeof *p);
The pointer you obtain cannot be dereferenced because it does not point to an object. You'd need to call a constructor on it before you can use it (and this is done using placement new).
If, on the other hand, you do:
non_pod_type* p = new non_pod_type();
You get a pointer that is always valid, because new created an object.
Even for POD types, there's a significant difference between the two:
pod_type* p = (pod_type*) malloc(sizeof *p);
std::cout << p->foo;
This piece of code would print an unspecified value, because the POD objects created by malloc are not initialised.
With new, you could specify a constructor to call, and thus get a well defined value.
pod_type* p = new pod_type();
std::cout << p->foo; // prints 0
If you really want it, you can use use new to obtain uninitialised POD objects. See this other answer for more information on that.
Another difference is the behaviour upon failure. When it fails to allocate memory, malloc returns a null pointer, while new throws an exception.
The former requires you to test every pointer returned before using it, while the later will always produce valid pointers.
For these reasons, in C++ code you should use new, and not malloc. But even then, you should not use new "in the open", because it acquires resources you need to release later on. When you use new you should pass its result immediately into a resource managing class:
std::unique_ptr<T> p = std::unique_ptr<T>(new T()); // this won't leak
Dynamic allocation is only required when the life-time of the object should be different than the scope it gets created in (This holds as well for making the scope smaller as larger) and you have a specific reason where storing it by value doesn't work.
For example:
std::vector<int> *createVector(); // Bad
std::vector<int> createVector(); // Good
auto v = new std::vector<int>(); // Bad
auto result = calculate(/*optional output = */ v);
auto v = std::vector<int>(); // Good
auto result = calculate(/*optional output = */ &v);
From C++11 on, we have std::unique_ptr for dealing with allocated memory, which contains the ownership of the allocated memory. std::shared_ptr was created for when you have to share ownership. (you'll need this less than you would expect in a good program)
Creating an instance becomes really easy:
auto instance = std::make_unique<Class>(/*args*/); // C++14
auto instance = std::unique_ptr<Class>(new Class(/*args*/)); // C++11
auto instance = std::make_unique<Class[]>(42); // C++14
auto instance = std::unique_ptr<Class[]>(new Class[](42)); // C++11
C++17 also adds std::optional which can prevent you from requiring memory allocations
auto optInstance = std::optional<Class>{};
if (condition)
optInstance = Class{};
As soon as 'instance' goes out of scope, the memory gets cleaned up. Transferring ownership is also easy:
auto vector = std::vector<std::unique_ptr<Interface>>{};
auto instance = std::make_unique<Class>();
vector.push_back(std::move(instance)); // std::move -> transfer (most of the time)
So when do you still need new? Almost never from C++11 on. Most of the you use std::make_unique until you get to a point where you hit an API that transfers ownership via raw pointers.
auto instance = std::make_unique<Class>();
legacyFunction(instance.release()); // Ownership being transferred
auto instance = std::unique_ptr<Class>{legacyFunction()}; // Ownership being captured in unique_ptr
In C++98/03, you have to do manual memory management. If you are in this case, try upgrading to a more recent version of the standard. If you are stuck:
auto instance = new Class(); // Allocate memory
delete instance; // Deallocate
auto instances = new Class[42](); // Allocate memory
delete[] instances; // Deallocate
Make sure that you track the ownership correctly to not have any memory leaks! Move semantics don't work yet either.
So, when do we need malloc in C++? The only valid reason would be to allocate memory and initialize it later via placement new.
auto instanceBlob = std::malloc(sizeof(Class)); // Allocate memory
auto instance = new(instanceBlob)Class{}; // Initialize via constructor
instance.~Class(); // Destroy via destructor
std::free(instanceBlob); // Deallocate the memory
Even though, the above is valid, this can be done via a new-operator as well. std::vector is a good example for this.
Finally, we still have the elephant in the room: C. If you have to work with a C-library where memory gets allocated in the C++ code and freed in the C code (or the other way around), you are forced to use malloc/free.
If you are in this case, forget about virtual functions, member functions, classes ... Only structs with PODs in it are allowed.
Some exceptions to the rules:
You are writing a standard library with advanced data structures where malloc is appropriate
You have to allocate big amounts of memory (In memory copy of a 10GB file?)
You have tooling preventing you to use certain constructs
You need to store an incomplete type
There are a few things which new does that malloc doesn’t:
new constructs the object by calling the constructor of that object
new doesn’t require typecasting of allocated memory.
It doesn’t require an amount of memory to be allocated, rather it requires a number of
objects to be constructed.
So, if you use malloc, then you need to do above things explicitly, which is not always practical. Additionally, new can be overloaded but malloc can’t be.
If you work with data that doesn't need construction/destruction and requires reallocations (e.g., a large array of ints), then I believe malloc/free is a good choice as it gives you realloc, which is way faster than new-memcpy-delete (it is on my Linux box, but I guess this may be platform dependent). If you work with C++ objects that are not POD and require construction/destruction, then you must use the new and delete operators.
Anyway, I don't see why you shouldn't use both (provided that you free your malloced memory and delete objects allocated with new) if can take advantage of the speed boost (sometimes a significant one, if you're reallocing large arrays of POD) that realloc can give you.
Unless you need it though, you should stick to new/delete in C++.
If you are using C++, try to use new/delete instead of malloc/calloc as they are operators. For malloc/calloc, you need to include another header. Don't mix two different languages in the same code. Their work is similar in every manner, both allocates memory dynamically from heap segment in hash table.
new will initialise the default values of the struct and correctly links the references in it to itself.
E.g.
struct test_s {
int some_strange_name = 1;
int &easy = some_strange_name;
}
So new struct test_s will return an initialised structure with a working reference, while the malloc'ed version has no default values and the intern references aren't initialised.
If you have C code you want to port over to C++, you might leave any malloc() calls in it. For any new C++ code, I'd recommend using new instead.
From a lower perspective, new will initialize all the memory before giving the memory whereas malloc will keep the original content of the memory.
In the following scenario, we can't use new since it calls constructor.
class B {
private:
B *ptr;
int x;
public:
B(int n) {
cout<<"B: ctr"<<endl;
//ptr = new B; //keep calling ctr, result is segmentation fault
ptr = (B *)malloc(sizeof(B));
x = n;
ptr->x = n + 10;
}
~B() {
//delete ptr;
free(ptr);
cout<<"B: dtr"<<endl;
}
};
Rare case to consider using malloc/free instead of new/delete is when you're allocating and then reallocating (simple pod types, not objects) using realloc as there is no similar function to realloc in C++ (although this can be done using a more C++ approach).
I had played before with few C/C++ applications for computer graphics.
After so many time, some things are vanished and I missed them a lot.
The point is, that malloc and new, or free and delete, can work both,
especially for certain basic types, which are the most common.
For instance, a char array, can be allocated both with malloc, or new.
A main difference is, with new you can instantiate a fixed array size.
char* pWord = new char[5]; // allocation of char array of fixed size
You cannot use a variable for the size of the array in this case.
By the contrary, the malloc function could allow a variable size.
int size = 5;
char* pWord = (char*)malloc(size);
In this case, it might be required a conversion cast operator.
For the returned type from malloc it's a pointer to void, not char.
And sometimes the compiler could not know, how to convert this type.
After allocation the memory block, you can set the variable values.
the memset function can be indeed slower for some bigger arrays.
But all the bites must be set first to 0, before assigning a value.
Because the values of an array could have an arbitrary content.
Suppose, the array is assigned with another array of smaller size.
Part of the array element could still have arbitrary content.
And a call to a memset function would be recomended in this case.
memset((void*)pWord, 0, sizeof(pWord) / sizeof(char));
The allocation functions are available for all C packages.
So, these are general functions, that must work for more C types.
And the C++ libraries are extensions of the older C libraries.
Therefore the malloc function returns a generic void* pointer.
The sructures do not have defined a new, or a delete operator.
In this case, a custom variable can be allocated with malloc.
The new and delete keywords are actually some defined C operators.
Maybe a custom union, or class, can have defined these operators.
If new and delete are not defined in a class, these may not work.
But if a class is derived from another, which has these operators,
the new and delete keywords can have the basic class behavior.
About freeing an array, free can be only used in pair with malloc.
Cannot allocate a variable with malloc, and then free with delete.
The simple delete operator references just first item of an array.
Because the pWord array can be also written as:
pWord = &pWord[0]; // or *pWord = pWord[0];
When an array must be deleted, use the delete[] operator instead:
delete[] pWord;
Casts are not bad, they just don't work for all the variable types.
A conversion cast is also an operator function, that must be defined.
If this operator is not defined for a certain type, it may not work.
But not all the errors are because of this conversion cast operator.
Also a cast to a void pointer must be used when using a free call.
This is because the argument of the free function is a void pointer.
free((void*)pWord);
Some errors can arise, because the size of the array is too small.
But this is another story, it is not because of using the cast.
With kind regards, Adrian Brinas
The new and delete operators can operate on classes and structures, whereas malloc and free only work with blocks of memory that need to be cast.
Using new/delete will help to improve your code as you will not need to cast allocated memory to the required data structure.
malloc() is used to dynamically assign memory in C
while the same work is done by new() in c++.
So you cannot mix coding conventions of 2 languages.
It would be good if you asked for difference between calloc and malloc()
As the title suggests, is it possible to start an array when you have the address given. i.e. If my start address is 10002432,basically &array[0] = 10002432 and array[1] will be stored in the next sequential address. I am able to copy to the said address using Pointers but instead of pointers, I would like to use arrays which are dynamically allocated. My current code is as follows which is using Pointers. I have tried various solutions that are currently are online but none of them are able to provide a definite solution.
I was thinking to make an overloaded new operator, but i get stuck on it as well, since underlying the overloaded new operator we use malloc which also doesnot allow to specify the start address.
I am running this code in a simulator (gem5), where I have control over memory addresses as I can specify how the virtual addresses map on the physical addresses.
int main(int argc, char const *argv[])
{
uintptr_t address = 10002432;
int *Pointer = (int*) address;
int *Pointer2 = Pointer;
for (int i = 0; i < 32; ++i)
{
*Pointer = i;
Pointer = Pointer + 4;
}
for (int i = 0; i < 32; ++i)
{
printf("%d\n",Pointer2 );
Pointer2 = Pointer2 + 4;
}
}
Any assistance would be highly appreciated.
No, it is not possible to "allocate" memory from specific address in standard C++.
Within an operating system that uses virtual memory (i.e. all modern multi-tasking operating systems), there is typically no system specific way to do such allocation either. It is however common to use constant addresses documented by the system provider on embedded systems. You can typically consider such memory "always allocated". Perhaps this applies to your simulator.
You can create dynamic objects into uninitialised memory through a pointer. The syntax for doing this is called placement new. Ths is sometimes useful to reuse "generic" allocated memory for some in-place containers such as std::vector, as well as in the mentioned case where the address is constant.
Important note: The address is assumed to satisfy the alignment requirement of the created object. If it is not satisfied, then the behaviour of the program is undefined.
Although there is syntax for doing so, it is not possible to create an array this way in practice. Array placement new is not guaranteed to start the array from the provided address. Some implementation specific amount of memory at the beginning from the address may be used by the language implementation.
Instead of creating an array, you can simply create the elements individually, separately, and pretend that they are in an array. This is what the implementation of std::vector in your standard library probably does as well. There are helper functions for that in the standard library:
using T = int;
std::uintptr_t address = 10002432; // Note: using hex would be more conventional
const std::size_t size = 32;
T* p = std::launder(reinterpret_cast<T*>(address));
if (!std::align(alignof(T), sizeof(T), p, sizeof(T)))
throw std::invalid_argument("bad alignment");
std::uninitialized_fill_n(p, size, T{});
// use it
std::destroy_n(p, size);
// Instead of filling from the value initialised value,
// you can copy a range. Here is an example with C++20 ranges:
auto i = std::views::iota(T{});
std::uninitialized_copy_n(i.begin(), size, p);
// use it
std::destroy_n(p, size);
Note that the destruction part is unnecessary in case of int and other trivial types, but mandatory for non-trivially-destructible types. This is a trivial example, and is not exception safe. If exception is thrown after successful initialisation, the destruction won't be called. You can use a RAII idiom to make sure the destruction happens.
If the given pointer is derived from a valid pointer to previously existing objects such as when reusing memory of a trivial object (which is not the case in the example), then std::launder is necessary as used in the example. It may be unnecessary otherwise (which is the case in the example, so you may get rid of laundering if you know what you're doing; but keeping it doesn't hurt).
When reinterpreting integer as pointer, the language standard provides no guarantee what memory address the value will be mapped to. You have to rely on guarantees provided by the language implementation for that memory address to be in any way useful. Except in the case you converted from a valid pointer into an integer (of sufficient size), and back to same pointer type, then that is guaranteed to be the same pointer value. But this case does not apply in the example.
I am following Stefanus Du Toit's hourglass pattern, that is, implementing a C API in C++ and then wrapping it in C++ again. This is very similar to the pimpl idiom, and it is also transparent to the user, but prevents more ABI-related issues and allows for a wider range of foreign language bindings.
As in the pointer-to-implementation approach, the underlying object's size and layout is not known by the outsiders at compile-time, so the memory in which it resides has to be dynamically allocated (mostly). However, unlike in the pimpl case, in which the object has been fully defined at the allocation point, here its properties are completely hidden behind an opaque pointer.
Memory obtained with std::malloc is "suitably aligned for any scalar type", which makes it unsuitable for the task. I'm not sure about the new-expression. Quoted from the section Allocation of the linked page:
In addition, if the new-expression is used to allocate an array of
char or an array unsigned char, it may request additional memory from
the allocation function if necessary to guarantee correct alignment of
objects of all types no larger than the requested array size, if one
is later placed into the allocated array.
Does this mean that the following code is compliant?
C API
size_t object_size ( void ); // return sizeof(internal_object);
size_t object_alignment ( void ); // return alignof(internal_object);
void object_construct ( void * p ); // new (p) internal_object();
void object_destruct ( void * p ); // static_cast<internal_object *>(p)->~internal_object();
C++ wrapper
/* The memory block that p points to is correctly aligned
for objects of all types no larger than object_size() */
auto p = new char[ object_size() ];
object_construct( p );
object_destruct( p );
delete[] p;
If it is not, how to dynamically allocate properly-aligned memory?
I can't find where the standard guarantees your proposed code to work. First, I cannot find the part of the standard that backs what you've quoted from CppReference.com, but even if we take that claim on faith, it still only says that it may allocate additional space. If it doesn't, you're sunk.
The standard does speak to the alignment of memory returned by operator new[]: "The pointer returned shall be suitably aligned so that it can be converted to a pointer of any complete object type …." (C++03, §3.7.2.1/2; C++11, §3.7.4.1/2) However, in the context where you're planning to allocate the memory, the type you plan to store in it isn't a complete type. And besides, the result of operator new[] isn't necessarily the same as the result of the new-expression new char[…]; the latter is allowed to allocate additional space for its own array bookkeeping.
You could use C++11's std::align. To guarantee that you allocate space that can be aligned to the required amount, you'd have to allocate object_size() + object_alignment() - 1 bytes, but in practice, allocating only object_size() bytes will probably be fine. Thus, you might try using std::align something like this:
size_t buffer_size = object_size();
void* p = operator new(buffer_size);
void* original_p = p;
if (!std::align(object_alignment(), object_size(), p, buffer_size) {
// Allocating the minimum wasn't enough. Allocate more and try again.
operator delete(p);
buffer_size = object_size() + object_alignment() - 1;
p = operator new(buffer_size);
original_p = p;
if (!std::align(object_alignment(), object_size(), p, buffer_size)) {
// still failed. invoke error-handler
operator delete(p);
}
}
object_construct(p);
object_destruct(p);
operator delete(original_p);
The allocator described in another question accomplishes much the same thing. It's templated on the type of object being allocated, which you don't have access to, but it's not required to be that way. The only times it uses its template type argument are to evaluate sizeof and alignof, which you already have from your object_size and object_alignment functions.
That seems like a lot to require from consumers of your library. It would be much more convenient for them if you moved the allocation behind the API as well:
void* object_construct() {
return new internal_object();
}
Make sure to move the destruction, too, by calling delete, not just the destructor.
That makes any alignment questions go away because the only module that really needs to know it is the one that already knows everything else about the type being allocated.
I read in a C++ book that malloc() & free() are liabrary functions, and thus are outside the control of the compiler.
However, if you have an operator to perform the combined act of dynamic storage allocation & initialization (new)and another operator to perform the combined act of clean up & releasing storage (delete), the compiler can still guarantee that constructors & destructors will be called for all objects.
So, I want to know that How does this is carried out by compiler?
Any Example or demo will be appriciated.
Thanks in advance.
malloc function returns a chunk of contiguous memory, that's all. How you type-cast and use it (for your objects) is your problem.
While the new operator returns objects allocated in memory. Although both returns a pointer, in the end you get the constructed objects and not raw memory in C++. Here the focus shifts from low-level memory handling to object handling with which comes type-safety. That's the reason new doesn't return void*.
Also, if you've noticed, in the former case, C, it is a function which does the allocation i.e. the language itself has no notion of allocating objects or memory. While in C++ new is an operator and thus the language inherently understands what it means to create objects dynamically. The compiler implements what the language mandates and is thus in a position to flag an error when it spots one.
An example would be:
int *ptr = malloc(sizeof(char) * 4);
Here the programmer assumes an integer is of size 4, while this might be true in his platform, it is not certainly true for all platforms. Also conceptually char and int are different types, but the type mismatch is ignored by the C compiler. All it does is calls the malloc function and assigns the address it returns to ptr. It's domain ends there, it's up to the programmer to use (or misuse?) the allocated memory. This isn't an error or weakness on the compiler's part since the language standard doesn't give more powers to the compiler to enforce more.
While in C++
int *ptr = new char[4];
would be flagged as an error; the right way to do it would be int *ptr = new int; where the types match. C++ is more stricter, by allowing lesser freedom in places where errors are common, there by leading to cleaner code. Type safety is arguably the biggest safety feature of the C++ language. Type casting has an ugly syntax for the same reason: they show weak points of a design. With a more stricter language, the compiler is able to enforce more restrictions on the developer (since humans are more error-prone, this works out well in most cases).
The compiler doesn't really directly "help" allocate and deallocate memory; your code has to explicitly do this. The C++ language provides deterministic execution of code when a thread leaves scope (destructors). That in turn is often used to free heap-allocated memory.
new and delete are key words in c++.
C++ compilers will generate hidden codes for you to deal with memory allocation/deallocation stuff.
For example,
struct Test() { }
Test *a = new Test();
The compiler might do something like this, (the following is pseudo code)
Test *a = (Test *)malloc(sizeof(Test));
if (a == nullptr) { throw std::bad_alloc; }
try
{
a.Test(); //call constructor
}
catch (...)
{
//constructor exception, free the memory first, then re-throw
free(a);
throw;
}
If it is for array, things will be more complicated,
struct Test() { }
Test *a = new Test[10];
The compiler might do something like this,
Test *a = (Test *)malloc(sizeof(Test) * 10);
if (a == nullptr) { throw std::bad_alloc; }
int i;
try
{
for (i = 0; i < 10, i++)
a[i].Test(); //call constructor
}
catch (...)
{
//call destructor for all constructed objects
for (int j = 0; j < i; j++)
a[j].~Test();
free(a);
throw;
}
Similar logic works for delete.
Alright, I think we all agree that what happens with the following code is undefined, depending on what is passed,
void deleteForMe(int* pointer)
{
delete[] pointer;
}
The pointer could be all sorts of different things, and so performing an unconditional delete[] on it is undefined. However, let's assume that we are indeed passing an array pointer,
int main()
{
int* arr = new int[5];
deleteForMe(arr);
return 0;
}
My question is, in this case where the pointer is an array, who is it that knows this? I mean, from the language/compiler's point of view, it has no idea whether or not arr is an array pointer versus a pointer to a single int. Heck, it doesn't even know whether arr was dynamically created. Yet, if I do the following instead,
int main()
{
int* num = new int(1);
deleteForMe(num);
return 0;
}
The OS is smart enough to only delete one int and not go on some type of 'killing spree' by deleting the rest of the memory beyond that point (contrast that with strlen and a non-\0-terminated string -- it will keep going until it hits 0).
So whose job is it to remember these things? Does the OS keep some type of record in the background? (I mean, I realise that I started this post by saying that what happens is undefined, but the fact is, the 'killing spree' scenario doesn't happen, so therefore in the practical world someone is remembering.)
One question that the answers given so far don't seem to address: if the runtime libraries (not the OS, really) can keep track of the number of things in the array, then why do we need the delete[] syntax at all? Why can't a single delete form be used to handle all deletes?
The answer to this goes back to C++'s roots as a C-compatible language (which it no longer really strives to be.) Stroustrup's philosophy was that the programmer should not have to pay for any features that they aren't using. If they're not using arrays, then they should not have to carry the cost of object arrays for every allocated chunk of memory.
That is, if your code simply does
Foo* foo = new Foo;
then the memory space that's allocated for foo shouldn't include any extra overhead that would be needed to support arrays of Foo.
Since only array allocations are set up to carry the extra array size information, you then need to tell the runtime libraries to look for that information when you delete the objects. That's why we need to use
delete[] bar;
instead of just
delete bar;
if bar is a pointer to an array.
For most of us (myself included), that fussiness about a few extra bytes of memory seems quaint these days. But there are still some situations where saving a few bytes (from what could be a very high number of memory blocks) can be important.
The compiler doesn't know it's an array, it's trusting the programmer. Deleting a pointer to a single int with delete [] would result in undefined behavior. Your second main() example is unsafe, even if it doesn't immediately crash.
The compiler does have to keep track of how many objects need to be deleted somehow. It may do this by over-allocating enough to store the array size. For more details, see the C++ Super FAQ.
Yes, the OS keeps some things in the 'background.' For example, if you run
int* num = new int[5];
the OS can allocate 4 extra bytes, store the size of the allocation in the first 4 bytes of the allocated memory and return an offset pointer (ie, it allocates memory spaces 1000 to 1024 but the pointer returned points to 1004, with locations 1000-1003 storing the size of the allocation). Then, when delete is called, it can look at 4 bytes before the pointer passed to it to find the size of the allocation.
I am sure that there are other ways of tracking the size of an allocation, but that's one option.
This is very similar to this question and it has many of the details your are looking for.
But suffice to say, it is not the job of the OS to track any of this. It's actually the runtime libraries or the underlying memory manager that will track the size of the array. This is usually done by allocating extra memory up front and storing the size of the array in that location (most use a head node).
This is viewable on some implementations by executing the following code
int* pArray = new int[5];
int size = *(pArray-1);
delete or delete[] would probably both free the memory allocated (memory pointed), but the big difference is that delete on an array won't call the destructor of each element of the array.
Anyway, mixing new/new[] and delete/delete[] is probably UB.
It doesn't know it's an array, that's why you have to supply delete[] instead of regular old delete.
I had a similar question to this. In C, you allocate memory with malloc() (or another similar function), and delete it with free(). There is only one malloc(), which simply allocates a certain number of bytes. There is only one free(), which simply takes a pointer as it's parameter.
So why is it that in C you can just hand over the pointer to free, but in C++ you must tell it whether it's an array or a single variable?
The answer, I've learned, has to do with class destructors.
If you allocate an instance of a class MyClass...
classes = new MyClass[3];
And delete it with delete, you may only get the destructor for the first instance of MyClass called. If you use delete[], you can be assured that the destructor will be called for all instances in the array.
THIS is the important difference. If you're simply working with standard types (e.g. int) you won't really see this issue. Plus, you should remember that behavior for using delete on new[] and delete[] on new is undefined--it may not work the same way on every compiler/system.
It's up to the runtime which is responsible for the memory allocation, in the same way that you can delete an array created with malloc in standard C using free. I think each compiler implements it differently. One common way is to allocate an extra cell for the array size.
However, the runtime is not smart enough to detect whether or not it is an array or a pointer, you have to inform it, and if you are mistaken, you either don't delete correctly (E.g., ptr instead of array), or you end up taking an unrelated value for the size and cause significant damage.
ONE OF THE approaches for compilers is to allocate a little more memory and store count of elements in the head element.
Example how it could be done:
Here
int* i = new int[4];
compiler will allocate sizeof(int)*5 bytes.
int *temp = malloc(sizeof(int)*5)
Will store 4 in first sizeof(int) bytes
*temp = 4;
and set i
i = temp + 1;
So i points to array of 4 elements, not 5.
And
delete[] i;
will be processed following way
int *temp = i - 1;
int numbers_of_element = *temp; // = 4
... call destructor for numbers_of_element elements if needed
... that are stored in temp + 1, temp + 2, ... temp + 4
free (temp)
Semantically, both versions of delete operator in C++ can "eat" any pointer; however, if a pointer to a single object is given to delete[], then UB will result, meaning anything may happen, including a system crash or nothing at all.
C++ requires the programmer to choose the proper version of the delete operator depending on the subject of deallocation: array or single object.
If the compiler could automatically determine whether a pointer passed to the delete operator was a pointer array, then there would be only one delete operator in C++, which would suffice for both cases.
Agree that the compiler doesn't know if it is an array or not. It is up to the programmer.
The compiler sometimes keep track of how many objects need to be deleted by over-allocating enough to store the array size, but not always necessary.
For a complete specification when extra storage is allocated, please refer to C++ ABI (how compilers are implemented): Itanium C++ ABI: Array Operator new Cookies
"undefined behaviour" simply means the language spec makes no gaurantees as to what will happen. It doesn't nessacerally mean that something bad will happen.
So whose job is it to remember these things? Does the OS keep some type of record in the background? (I mean, I realise that I started this post by saying that what happens is undefined, but the fact is, the 'killing spree' scenario doesn't happen, so therefore in the practical world someone is remembering.)
There are typically two layers here. The underlying memory manager and the C++ implementation.
Most memory managers were designed to meet the needs of the C language. In C the "free" function does not require the user to specify the size of the block. Therefore the memory manager will remember (among other things) the size of the block of memory that was allocated. This may be larger than the block the C++ implementation asked for. Typically the memory manager will store it's metadata before the allocated block of memory.
C++ has a culture of "you only pay for what you use". Therefore the C++ implementation will generally only remember the size of the array if it needs to do so for it's own purposes, typically because the type has a non-trival destructor.
So for types with a trivial destructor the implementation of "delete" and "delete []" is typically the same. The C++ implementation simply passes the pointer to the underlying memory manager. Something like
free(p)
On the other hand for types with a non-trivial destructor "delete" and "delete []" are likely to be different. "delete" would be somthing like (where T is the type that the pointer points to)
p->~T();
free(p);
While "delete []" would be something like.
size_t * pcount = ((size_t *)p)-1;
size_t count = *count;
for (size_t i=0;i<count;i++) {
p[i].~T();
}
char * pmemblock = ((char *)p) - max(sizeof(size_t),alignof(T));
free(pmemblock);
You cannot use delete for an array, and you cannot use delete [] for a non-array.
Hey ho well it depends of what you allocating with new[] expression when you allocate array of build in types or class / structure and you don't provide your constructor and destructor the operator will treat it as a size "sizeof(object)*numObjects" rather than object array therefore in this case number of allocated objects will not be stored anywhere, however if you allocate object array and you provide constructor and destructor in your object than behavior change, new expression will allocate 4 bytes more and store number of objects in first 4 bytes so the destructor for each one of them can be called and therefore new[] expression will return pointer shifted by 4 bytes forward, than when the memory is returned the delete[] expression will call a function template first, iterate through array of objects and call destructor for each one of them. I've created this simple code witch overloads new[] and delete[] expressions and provides a template function to deallocate memory and call destructor for each object if needed:
// overloaded new expression
void* operator new[]( size_t size )
{
// allocate 4 bytes more see comment below
int* ptr = (int*)malloc( size + 4 );
// set value stored at address to 0
// and shift pointer by 4 bytes to avoid situation that
// might arise where two memory blocks
// are adjacent and non-zero
*ptr = 0;
++ptr;
return ptr;
}
//////////////////////////////////////////
// overloaded delete expression
void static operator delete[]( void* ptr )
{
// decrement value of pointer to get the
// "Real Pointer Value"
int* realPtr = (int*)ptr;
--realPtr;
free( realPtr );
}
//////////////////////////////////////////
// Template used to call destructor if needed
// and call appropriate delete
template<class T>
void Deallocate( T* ptr )
{
int* instanceCount = (int*)ptr;
--instanceCount;
if(*instanceCount > 0) // if larger than 0 array is being deleted
{
// call destructor for each object
for(int i = 0; i < *instanceCount; i++)
{
ptr[i].~T();
}
// call delete passing instance count witch points
// to begin of array memory
::operator delete[]( instanceCount );
}
else
{
// single instance deleted call destructor
// and delete passing ptr
ptr->~T();
::operator delete[]( ptr );
}
}
// Replace calls to new and delete
#define MyNew ::new
#define MyDelete(ptr) Deallocate(ptr)
// structure with constructor/ destructor
struct StructureOne
{
StructureOne():
someInt(0)
{}
~StructureOne()
{
someInt = 0;
}
int someInt;
};
//////////////////////////////
// structure without constructor/ destructor
struct StructureTwo
{
int someInt;
};
//////////////////////////////
void main(void)
{
const unsigned int numElements = 30;
StructureOne* structOne = nullptr;
StructureTwo* structTwo = nullptr;
int* basicType = nullptr;
size_t ArraySize = 0;
/**********************************************************************/
// basic type array
// place break point here and in new expression
// check size and compare it with size passed
// in to new expression size will be the same
ArraySize = sizeof( int ) * numElements;
// this will be treated as size rather than object array as there is no
// constructor and destructor. value assigned to basicType pointer
// will be the same as value of "++ptr" in new expression
basicType = MyNew int[numElements];
// Place break point in template function to see the behavior
// destructors will not be called and it will be treated as
// single instance of size equal to "sizeof( int ) * numElements"
MyDelete( basicType );
/**********************************************************************/
// structure without constructor and destructor array
// behavior will be the same as with basic type
// place break point here and in new expression
// check size and compare it with size passed
// in to new expression size will be the same
ArraySize = sizeof( StructureTwo ) * numElements;
// this will be treated as size rather than object array as there is no
// constructor and destructor value assigned to structTwo pointer
// will be the same as value of "++ptr" in new expression
structTwo = MyNew StructureTwo[numElements];
// Place break point in template function to see the behavior
// destructors will not be called and it will be treated as
// single instance of size equal to "sizeof( StructureTwo ) * numElements"
MyDelete( structTwo );
/**********************************************************************/
// structure with constructor and destructor array
// place break point check size and compare it with size passed in
// new expression size in expression will be larger by 4 bytes
ArraySize = sizeof( StructureOne ) * numElements;
// value assigned to "structOne pointer" will be different
// of "++ptr" in new expression "shifted by another 4 bytes"
structOne = MyNew StructureOne[numElements];
// Place break point in template function to see the behavior
// destructors will be called for each array object
MyDelete( structOne );
}
///////////////////////////////////////////
just define a destructor inside a class and execute your code with both syntax
delete pointer
delete [] pointer
according to the output u can find the solutions
The answer:
int* pArray = new int[5];
int size = *(pArray-1);
Posted above is not correct and produces invalid value.
The "-1"counts elements
On 64 bit Windows OS the correct buffer size resides in Ptr - 4 bytes address