In [programming](programming.md) *memory management* is (unsurprisingly) the act and various techniques of managing the working [memory](memory.md) ([RAM](ram.md)) of a computer, i.e. for example dividing the total physically available memory among multiple memory users such as operating system processes and assuring they don't illegally access each other's part of memory. The scope of the term may differ depending on context, but tasks falling under memory management may include e.g. memory [allocation](allocation.md) (finding and assigning blocks of free memory) and deallocation (freeing such blocks), ensuring [memory safety](memory_safety.md), organizing blocks of memory and [optimizing](optimization.md) memory access (e.g. with [caches](cache.md) or data reorganization), [memory virtualization](virtual_memory.md) and related tasks such as address translation, handling out-of-memory [exceptions](exception.md) etc.
Memory management can be handled at different levels: hardware units such as the [MMU](mmu.md) and CPU [caches](cache.md) exist to perform certain time-critical memory-related tasks (such as address translation) quickly, [operating system](os.md) may help with memory management (e.g. implement virtual memory and offer [syscalls](syscall.md) for dynamic allocation and deallocation of memory), a [programming language](programming_language.md) may do some automatic memory management (e.g. [garbage collection](garbage_collection.md) or handling call stack) and programmer himself may do his own memory management (e.g. deciding between static and dynamic allocation or choosing the size of dynamic allocation chunk).
**Why all this fuzz?** As a newbie programmer who only works with simple variables and high level languages like [Python](python.md) that do everything for you you don't need to do much memory management yourself, but when working with data whose size may wildly differ and is not known in advance (e.g. files), someone has to handle e.g. the possibility of the data on disk not being able to fit to RAM currently allocated for your program, or -- if the data fits -- there may not be a big enough continuous chunk of memory for it. If we don't know how much memory a process will need, how much memory do we give it (too little and it may not be enough, too much and there will not be enough memory for others)? Someone has to prevent [memory leaks](memory_leak.md) so that your computer doesn't run out of memory due to [bugs](bug.md) in programs. With many [processes](process.md) running [simultaneously](multitasking.md) on a computer someone has to keep track of which process uses which part of memory and ensure [collisions](collision.md) (one process overwriting another processe's memory) don't happen, and someone needs to make sure that if bad things happen (such as process trying to write to a memory that doesn't belong to it), they don't have catastrophic consequences like [crashing](crash.md) or exploding the system.
## Memory Management In C
In [C](c.md) -- a [low level](low_level.md) language -- you need to do a lot of **manual** memory management and there is a **big danger of fucking up**, especially with dynamic allocation -- C won't hold your hand (but as a reward your program will be fast and efficient), there is no uber memory safety. There is no automatic [garbage collection](garbage_collection.md), i.e. if you allocate memory dynamically, YOU need to keep track of it and manually free it once you're done using it, or you'll end up with [memory leak](memory_leak.md).
For start let's see which kinds of allocation (and their associated parts of memory) there are in C:
- **static allocation (code/data memory)**: Simplest kind of allocation happening at compile time: if the compiler can do so (i.e. if it knows enough things such as the size of the data in advance), it allocates space of concrete size at some specific address in the part of memory reserved for code or static data (code and data may be in the same or separate parts depending on platform, see e.g. [Harvard architecture](harvard.md)) -- this is straightforward, simple, automatic and poses no real dangers, bloat or burden of dependencies. This kind of allocation applies to:
- **global variables** (variables declared outside any function, i.e. even outside `main`)
- **static variables** (variables inside functions declared with `static` keyword)
- **automatic allocation (stack memory)**: For local variables (variables inside functions) the memory is allocated in a special part of memory known as **call [stack](stack.md)** only at the time when the function is actually called and executed; i.e. this is similar to dynamic allocation (it happens at run time) but happens automatically, without needing any libraries or other explicit actions from the programmer. I.e. when a function is called at run time, a new *call frame* is created on stack which includes space for local variables of that function (along with e.g. return address from the function etc.). This is necessary e.g. to allow [recursion](recursion.md) (during which several instances of the same function may be active, each of which may have different values of its variables), and it also helps consume less RAM. This allows for creating variable sized arrays inside functions (e.g. `int array[x];` where `x` is variable) which is not possible to do with a global array (however variable size arrays aren't supported in old ANSI C!). The disadvantage over dynamic allocation is that stack memory is relatively small and overusing it may easily cause stack [overflow](overflow.md) (running out of memory). Still this kind of allocation is better than dynamic allocation as it doesn't need any libraries, it doesn't generate complex code and the only danger is that of stack overflow -- memory leaks can't happen (deallocation happens automatically when function is exited). Automatic allocation applies to:
- **local variables** (including function arguments and local **variable size arrays**)
- **dynamic allocation (heap memory)**: A kind of more complex manual allocation that happens at run time and is initiated by the programmer calling special functions such as `malloc` from the `stdlib` standard library, which return [pointers](pointer.md) to the allocated memory. This memory is taken from a special part of memory known as **[heap](heap.md)**. This allows to allocate, resize and deallocate potentially very big parts of memory, but requires caution as working with pointers is involved and there is a danger of **memory leaks** -- it is the responsibility of the programmer to free allocated memory with the `free` function once it is no longer needed, otherwise that memory will simply remain allocated and unusable by others (if this happens for example in a loop, the program may just start eating up more and more RAM and eventually run out of memory). Dynamic allocation is also pretty complex (it usually involves communicating with operating system and also keeping track of the structure of memory) and creates a [dependency](dependency.md) on the `stdlib` library. Some implementations of the allocation functions are also infamously slow (up to the point of some programmers resorting to program their own dynamic allocation systems). Therefore only use dynamic allocation when absolutely necessary! Dynamic allocation applies to:
- **memory allocated with special functions** (`malloc`, `calloc`, `realloc`)
Rule of the thumb: use the simplest thing possible, i.e. static allocation if you can, if not then automatic and only as the last option resort to dynamic allocation. The good news is that **you mostly won't need dynamic allocation** -- you basically only need it when working with data whose size can potentially be VERY big and is unknown at compile time (e.g. you need to load a WHOLE file AT ONCE which may potentially be VERY big). In other cases you can get away with static allocation (just reserving some reasonable amount of memory in advance and hope the data fits, e.g. a global array such as `int myData[DATA_MAX_SIZE]`) or automatic allocation if the data is reasonably small (i.e. you just create a variable sized array inside some function that processes the data). If you end up doing dynamic allocation, be careful, but it's not THAT hard to do it right (just pay more attention) and there are tools (e.g. [valgrind](valgrind.md)) to help you find memory leaks. However by the principles of [good design](lrs.md) **you should avoid dynamic allocation** if you can, not only because of the potential for errors and worse performance, but most importantly to avoid dependencies and complexity.
For [pros](pro.md): you can also create your own kind of pseudo dynamic allocation in pure C if you really want to avoid using stdlib or can't use it for some reason. The idea is to allocate a big chunk of memory statically (e.g. global `unsigned char myHeap[MY_HEAP_SIZE];`) and then create functions for allocating and freeing blocks of this static memory (e.g. `myAlloc` and `myFree` with same signatures as `malloc` and `free`). This allows you to use memory more efficiently than if you just dumbly (is it a word?) preallocate everything statically, i.e. you may need less total memory; this may be useful e.g. on [embedded](embedded.md). Yet another uber [hack](hacking.md) to "improve" this may be to allocate the "personal heap" on the stack instead of statically, i.e. you create something like a global pointer `unsigned char *myHeapPointer;` and a global variable `unsigned int myHeapSize;`, then somewhere at the beginning of `main` you compute the size `myHeapSize` and then create a local array `myHeap[myHeapSize]`, then finally set the global pointer to it as `myHeapPointer = myHeap`; the rest remains the same (your allocation function will access the heap via the global pointer). Just watch out for reinventing wheels, bugs and that you actually don't end up with a worse mess that if you took a more simple approach. Hell, you might even try to write your own garbage collection and array bound checking and whatnot, but then why just not fuck it and use an already existing abomination like [Java](java.md)? :)
