Yes, that you have absolutely no idea how the data is stored internally (there could be padding or not) and even changing compiler settings could override that (which is why some compilers have #pragmas to override this).

That example still makes a mistake though as it completely ignores the endianess of the system. Save a file in a little endian system then try to load it in a big endian system... you won't like it (this only matters for portability reasons, but there's not much of an excuse to avoid it). Also you the size of the variable types may change (e.g. a common recent case would be 32-bit vs 64-bit). You'll need to write individual bytes to work around that (then at least you can be guaranteed in which order they'll be stored in the file and with what size).

EDIT: oh, also even if none of the above was an issue, what if you change the object for whatever reason, e.g. to add a new variable or something? If you save the object as-is, suddenly old saves won't be usable anymore.

Let me give you an example Save load for a simple Pet Game.

Let's assume you have a Pet class. In this class it has the pet's age (based on days), it's health, and it's happiness. let's also assume it has an inventory of items which are unique based on some unique int ID. Let's also add a Save() and Load() function. At it's definition, it may look like this:

 
#include "TItem.h"
 
enum THappiness {
  SAD_PET,
  CONTENT_PET,
  HAPPY_PET
};
 
class TPet {
private:
  int Age;
  int Health;
  std::string Name;
 
  THappiness Happiness;
  std::vector<TItems> Inventory;
 
public:
  TPet(std::string petName);
  ~TPet();
 
  void AddItem(TItem item);
  std::vector<TItems> GetInventory();
  std::string GetName();
 
  int GetHealth();
  void Sethealth(int Health);
 
  int GetAge();  
  void IncreamentAge();
 
  THappiness GetHappiness();
  void SetHappiness(THappiness happiness);
  
  void Load(std::string saveFileName);
  void Save(std::string saveFileName);  
};

I'm not going to worry about the other files, and no go into detail about reading and writing to files (nor loops), but your save and load would do something like this (in Pseudo-code):

void TPet::Save(std::string saveFileName)
{
  FileType saveFile = FileOpen(saveFileName);
 
  // We know Age Health and happiness will always be the same length, so we can write them in without
  // having to write in the length of the data
  WriteFile(saveFile, &Age, sizeof(Age));
  WriteFile(saveFile, &Health, sizeof(Health));
  WriteFile(saveFile, &Happiness, sizeof(Happiness));
 
  // Store the Name, but give it the name length 1st
  int length = Name.size();
  WriteFile(saveFile, &length, sizeof(length));
  // loop through and write each character
 
  // Now write the number of inventory items
  length = Inventory.size();
  WriteFile(saveFile, &length, sizeof(length));
  // now loop through and store each item id, assuming you can access TItem::GetId()
 
  CloseFile(saveFile);
}
 
// In load you do the same thing in the same order

I got a related question to this. Can't you just write a whole instance of a object to the binary file, and then load a full object with a load function? Or am I missing something?

No, you can't. In the class, there are instances of std::string, and a Vector Of Items being stored. What is actually stored for those values in the class instance is only an address. So, you need to explicitly store each of those items (the character's in the string, and the inventory items stores in the vector).

Yes, that you have absolutely no idea how the data is stored internally (there could be padding or not) and even changing compiler settings could override that (which is why some compilers have #pragmas to override this).

That example still makes a mistake though as it completely ignores the endianess of the system. Save a file in a little endian system then try to load it in a big endian system... you won't like it (this only matters for portability reasons, but there's not much of an excuse to avoid it). Also you the size of the variable types may change (e.g. a common recent case would be 32-bit vs 64-bit). You'll need to write individual bytes to work around that (then at least you can be guaranteed in which order they'll be stored in the file and with what size).

EDIT: oh, also even if none of the above was an issue, what if you change the object for whatever reason, e.g. to add a new variable or something? If you save the object as-is, suddenly old saves won't be usable anymore.

This is a ridiculous thing to be concerned about for a beginner's project.

This is a "For Beginner's" forum, so explaining endianness to beginner's is like explaining calculus to a Pre-Algebra class; calculus is good to know, but not for Pre-Algebra students.

You're assuming #1, the user will make this game for different platforms (and different OS's), and #2, he's planning on supporting saving on one machine and loading on another. I'll guarantee you he's not.

As far as having different version of saved files, that's something, but my example was just that; an example, to give this guy an idea of how to save. Heck, it's obvious not complete, I didn't even fill out the loops. Leave the advanced details to other forums looking for advanced help.

[quote name='Nausea' timestamp='1357503230' post='5018290']
Ok, thank you very much for explaining that
So I would have to save everything inside a class separate and then load it in the right order later?
[/quote]

If you have items like strings or a list of items, then you will have to manually write and then read that data out.

If you only have native types, like int's, then you can store that inside a struct (or possibly a class), and just write the struct data to a file, and then read it back just as you wrote it, like this:

Serialization sounds scary but it's actually really easy to do once you give it a try. For every class whose instances will need to be serialized just write a pair of load/dump functions for it that convert to and from 'raw' data. If it has members that aren't raw data then those member object classes also need load/dump functions written, then you just call those from the owner and chain it on to the serialization. By the time you reach the end you can just load/dump the top level object and the whole chain should respond correctly. It's really a lot easier still if you make use of istream and ostream for loading and dumping, since you can just pass the stream down the chain to whatever is getting serialized next. Just remember that for containers you'll want to dump a member count prior to dumping the actual members so that when you go to load the data you know how much data to use for populating the container before moving on to populating the other members.

Seriously, though, it sounds more complex than it is. When you start doing it yourself it should come out sort of naturally. Just start by writing the dump functions and setting everything up for that, then start writing the load functions and adjust the dump functions as needed until everything matches up. If you get stuck then just post showing where you're at and ask for help.

In terms of security, which was mentioned by OP, if I want to cheat in a game I typically just modify the values in memory while it's running, so I wouldn't worry too much about the savefile format until it becomes an issue in competitive environments or something. Basically the more powerful you make your security the more hackers get off on breaking it. Even giant companies like Sony or Nintendo can't engineer foolproof systems with millions of dollars of research, so....

Serialization is enough to stop people that don't care. Nothing is enough to stop people that do care.