Better to allocate on the stack or heap within a class

Question

This was more of a conceptual question, but I'll provide a particular instance where I'm wondering about this. If I have a class that has several objects as properties, is it better to allocate them statically within the class, or dynamically during construction? For example, I have the following class (unecessary code emitted)

class OutlinedText
{
protected:
    sf::Text BaseText;
    sf::Text BackgroundText;
}

sf::Text are both objects as well. My question is, is it better to have them as declared above, and initialize them as follows

OutlinedText::OutlinedText(std::string &text, sf::Color MainColor, sf::Color OffsetColor, sf::Font &font, sf::Vector2f Pos, unsigned int BaseFontSize, unsigned int BackFontSize, float Offset)
{
    BaseText = sf::Text(text, font, BaseFontSize);
    BaseText.SetColor(MainColor);
    BackgroundText = sf::Text(text, font, BackFontSize);
    BackgroundText.SetColor(OffsetColor);
}

or, should i have them as pointers and allocate them with new as follows:

BaseText = new sf::Text(text, font, BaseFontSize);
BaseText->SetColor(MainColor);

and deallocate them with delete in the destructor? I know the stack allocates memory alot faster, but I think I'm double initializing the way I have now. SO is it a case by case thing or is one always better then the other? I'm still used to the C# way of doing things, so I was curious what the proper way to do this for C++ is. If I have a fundamental misunderstanding, please correct me.

Thanks in advance

On a side note, take a look at using an initialization list. — Peter Huene, Apr 02 '11 at 04:04

score 12 · Accepted Answer · edited May 23 '17 at 10:24

Wherever possible, you should avoid explicit dynamic allocation. Dynamic allocation is relatively expensive and makes object lifetime management more difficult.

Note that your question is somewhat misleading: in the first case, BaseText and BackgroundText aren't necessarily allocated on the stack. If the OutlinedText object of which they are a member is allocated on the heap or is a static variable, then they won't exist on the stack at all.

I think I'm double initializing the way I have it now.

You are: each of the member variables is initialized before the constructor body is entered, then in the constructor body, you assign to the member variables, so effectively they are getting "double initialized."

You can (and in most cases should) use the constructor's initialization list to initialize member variables:

OutlinedText::OutlinedText(std::string& text, 
                           sf::Color MainColor, 
                           sf::Color OffsetColor,
                           sf::Font& font, 
                           sf::Vector2f Pos, 
                           unsigned int BaseFontSize, 
                           unsigned int BackFontSize, 
                           float Offset)
    : BaseText(text, font, BaseFontSize),         // This is the constructor's
      BackgroundText(text, font, BackFontSize)    // initializer list
{
    BaseText.SetColor(MainColor);
    BackgroundText.SetColor(OffsetColor);
}

This way, the member variables only get initialized once (via the initializers in the initializer list), not twice.

Should I have them as pointers and allocate them with new as follows and deallocate them with delete in the destructor?

No: you should not have to write delete in your C++ program: you should use a smart pointer (e.g., auto_ptr, shared_ptr, or unique_ptr, depending on what kind of object lifetime you need) to ensure that dynamically allocated objects are automatically destroyed. If you need to store a collection of objects, you should use one of the Standard Library containers, like vector, map, or set, which also automatically clean up the objects that you store in them.

Manually managing the lifetime of a dynamically allocated object is tedious and difficult to get right and should be avoided. You don't just need to "clean up in the destructor," you also need to correctly implement (or suppress automatic generation of) a copy constructor and a copy assignment operator.

C++ is fundamentally different from C#, especially with respect to object lifetimes and how objects come into existence, how they are copied, and when they are destroyed. Definitely make sure that you have a good introductory C++ book if you really want to learn the language.

@Martin: Good idea; I missed the OP's remark about him thinking he was double-initializing things. — James McNellis, Apr 02 '11 at 15:12

score 2 · Answer 2 · answered Apr 02 '11 at 07:22

As James answered your primary question, I'll touch on something else you said:

I know the stack allocates memory alot faster, but I think I'm double initializing the way I have now.

You're not technically double-initializing, but what you're doing is indeed inefficient. As @Peter Huene commented, you should be using an initialization list:

OutlinedText::OutlinedText(
    std::string& text,
    sf::Color MainColor,
    sf::Color OffsetColor,
    sf::Font& font,
    sf::Vector2f Pos,
    unsigned int BaseFontSize,
    unsigned int BackFontSize,
    float Offset
)
  : BaseText(sf::Text(text, font, BaseFontSize)),
    BackgroundText(sf::Text(text, font, BackFontSize))
{
    BaseText.SetColor(MainColor);
    BackgroundText.SetColor(OffsetColor);
}

This idiom, in combination with avoiding explicit dynamic allocation, should be nearly always preferred.

Also, it doesn't appear that you're modifying text or font, so you should probably be passing them by const& rather than by &. (Also consider passing MainColor, OffsetColor, and Pos by const& rather than by value if they're larger than sizeof(void*).)

score 0 · Answer 3 · answered Apr 02 '11 at 04:04

If they will fit on the stack and you can initialise them to a useful state in the ctor then stack variables save you constantly checking for them being valid.

If they can't be set until some later point then using new and having an unset variable as NULL might make sense, especially if the variable doens't have an obvious not-set value

score 0 · Answer 4 · answered Apr 02 '11 at 04:08

Both and neither, or the best engineering answer yet, "It depends!"

For small objects, allocation on the stack is great and convenient. The objects will be constructed for you automatically, and if you want to invoke an alternate ctor you should really use the initializer list to avoid the double-initializing issue you mention.

Larger objects, or objects that are passed out of the class, are appropriate for the heap instead.

score 0 · Answer 5 · answered Apr 02 '11 at 04:25

generally, it is best to use an instance variable which is not allocated dynamically.

you can be sure it exists when the object uses it
it will require less overall memory
it is less likely to fail
copy/move is simpler to implement (it's often automatic)
storage is automatic (smart pointer implementations are not necessarily thread safe)

sometimes, you'll need to allocate it dynamically. typically, you'll know when to do this:

lazy initialization (sometimes it's necessary... it's also a really bad idea in some cases)
polymorphism
decoupling (e.g. using PIMPL to reduce build dependencies)
shared memory (e.g. it's ref-counted)
its physical size is large. even when the physical size is large, it's often better to to force the clients to allocate the container on the heap by making the constructors private.

wrt your example: the text in the text objects is likely allocated dynamically - the text object itself likely consumes a few words. if it's only a few words and there is no need to allocate it dynamically, definitely avoid dynamic allocation.

when you choose dynamic, always use some form of smart pointer.

good luck!

Better to allocate on the stack or heap within a class

5 Answers5