Related
I know that virtual functions have an overhead of dereferencing to call a method. But I guess with modern architectural speed it is almost negligible.
Is there any particular reason why all functions in C++ are not virtual as in Java?
From my knowledge, defining a function virtual in a base class is sufficient/necessary. Now when I write a parent class, I might not know which methods would get over-ridden. So does that mean that while writing a child class someone would have to edit the parent class. This sounds like inconvenient and sometimes not possible?
Update:
Summarizing from Jon Skeet's answer below:
It's a trade-off between explicitly making someone realize that they are inheriting functionality [which has potential risks in themselves [(check Jon's response)] [and potential small performance gains] with a trade-off for less flexibility, more code changes, and a steeper learning curve.
Other reasons from different answers:
Virtual functions cannot be in-lined because inlining have to happen at runtime. This have performance impacts when you expect you functions benefits from inlining.
There might be potentially other reasons, and I would love to know and summarize them.
There are good reasons for controlling which methods are virtual beyond performance. While I don't actually make most of my methods final in Java, I probably should... unless a method is designed to be overridden, it probably shouldn't be virtual IMO.
Designing for inheritance can be tricky - in particular it means you need to document far more about what might call it and what it might call. Imagine if you have two virtual methods, and one calls the other - that must be documented, otherwise someone could override the "called" method with an implementation which calls the "calling" method, unwittingly creating a stack overflow (or infinite loop if there's tail call optimization). At that point you've then got less flexibility in your implementation - you can't switch it round at a later date.
Note that C# is a similar language to Java in various ways, but chose to make methods non-virtual by default. Some other people aren't keen on this, but I certainly welcome it - and I'd actually prefer that classes were uninheritable by default too.
Basically, it comes down to this advice from Josh Bloch: design for inheritance or prohibit it.
One of the main C++ principles is: you only pay for what you use ("zero overhead principle"). If you don't need the dynamic dispatch mechanism, you shouldn't pay for its overhead.
As the author of the base class, you should decide which methods should be allowed to be overridden. If you're writing both, go ahead and refactor what you need. But it works this way, because there has to be a way for the author of the base class to control its use.
But I guess with modern architectural speed it is almost negligible.
This assumption is wrong, and, I guess, the main reason for this decision.
Consider the case of inlining. C++’ sort function performs much faster than C’s otherwise similar qsort in some scenarios because it can inline its comparator argument, while C cannot (due to use of function pointers). In extreme cases, this can mean performance differences of as much as 700% (Scott Meyers, Effective STL).
The same would be true for virtual functions. We’ve had similar discussions before; for instance, Is there any reason to use C++ instead of C, Perl, Python, etc?
Most answers deal with the overhead of virtual functions, but there are other reasons not to make any function in a class virtual, as the fact that it will change the class from standard-layout to, well, non-standard-layout, and that can be a problem if you need to serialize binary data. That is solved differently in C#, for example, by having structs being a different family of types than classes.
From the design point of view, every public function establishes a contract between your type and the users of the type, and every virtual function (public or not) establishes a different contract with the classes that extend your type. The greater the number of such contracts that you sign the less room for changes that you have. As a matter of fact, there are quite a few people, including some well known writers, that defend that the public interface should never contain virtual functions, as your compromise to your clients might be different from the compromises you require from your extensions. That is, the public interfaces shows what you do for your clients, while the virtual interface shows how others might help you in doing it.
Another effect of virtual functions is that they always get dispatched to the final overrider (unless you explicitly qualify the call), and that means that any function that is needed to maintain your invariants (think the state of the private variables) should not be virtual: if a class extends it, it will have to either make an explicit qualified call back to the parent or else would break the invariants at your level.
This is similar to the example of the infinite loop/stack overflow that #Jon Skeet mentioned, just in a different way: you have to document in each function whether it accesses any private attributes so that extensions will ensure that the function is called at the right time. And that in turn means that you are breaking encapsulation and you have a leaking abstraction: Your internal details are now part of the interface (documentation + requirements on your extensions), and you cannot modify them as you wish.
Then there is performance... there will be an impact in performance, but in most cases that is overrated, and it could be argued that only in the few cases where performance is critical, you would fall back and declare the functions non-virtual. Then again, that might not be simple on a built product, since the two interfaces (public + extensions) are already bound.
You forget one thing. The overhead is also in memory, that is you add a virtual table and a pointer to that table for each object. Now if you have an object which has significant number of instances expected then it is not negligible. example, million instance equals 4 Mega byte. I agree that for simple application this is not much, but for real time devices such as routers this counts.
I'm rather late to the party here, so I'll add one thing that I haven't noticed covered in other answers, and summarise quickly...
Usability in shared memory: a typical implementation of virtual dispatch has a pointer to a class-specific virtual dispatch table in each object. The addresses in these pointers are specific to the process creating them, which means multi-process systems accessing objects in shared memory can't dispatch using another process's object! That's an unacceptable limitation given shared memory's importance in high-performance multi-process systems.
Encapsulation: the ability of a class designer to control the members accessed by client code, ensuring class semantics and invariants are maintained. For example, if you derive from std::string (I may get a few comments for daring to suggest that ;-P) then you can use all the normal insert / erase / append operations and be sure that - provided you don't do anything that's always undefined behaviour for std::string like pass bad position values to functions - the std::string data will be sound. Someone checking or maintaining your code doesn't have to check if you've changed the meaning of those operations. For a class, encapsulation ensures freedom to later modify the implementation without breaking client code. Another perspective on the same statement: client code can use the class any way it likes without being sensitive to the implementation details. If any function can be changed in a derived class, that whole encapsulation mechanism is simply blown away.
Hidden dependencies: when you know neither what other functions are dependent on the one you're overriding, nor that the function was designed to be overridden, then you can't reason about the impact of your change. For example, you think "I've always wanted this", and change std::string::operator[]() and at() to consider negative values (after a type-cast to signed) to be offsets backwards from the end of the string. But, perhaps some other function was using at() as a kind of assertion that an index was valid - knowing it'll throw otherwise - before attempting an insertion or deletion... that code might go from throwing in a Standard-specified way to having undefined (but likely lethal) behaviour.
Documentation: by making a function virtual, you're documenting that it is an intended point of customisation, and part of the API for client code to use.
Inlining - code side & CPU usage: virtual dispatch complicates the compiler's job of working out when to inline function calls, and could therefore provide worse code in terms of both space/bloat and CPU usage.
Indirection during calls: even if an out-of-line call is being made either way, there's a small performance cost for virtual dispatch that may be significant when calling trivially simple functions repeatedly in performance critical systems. (You have to read the per-object pointer to the virtual dispatch table, then the virtual dispatch table entry itself - means the VDT pages are consuming cache too.)
Memory usage: the per-object pointers to virtual dispatch tables may represent significant wasted memory, especially for arrays of small objects. This means less objects fit in cache, and can have a significant performance impact.
Memory layout: it's essential for performance, and highly convenient for interoperability, that C++ can define classes with the exact memory layout of member data specified by network or data standards of various libraries and protocols. That data often comes from outside your C++ program, and may be generated in another language. Such communications and storage protocols won't have "gaps" for pointers to virtual dispatch tables, and as discussed earlier - even if they did, and the compiler somehow let you efficiently inject the correct pointers for your process over incoming data, that would frustrate multi-process access to the data. Crude-but-practical pointer/size based serialisation/deserialisation/comms code would also be made more complicated and potentially slower.
Pay per use (in Bjarne Stroustrup words).
Seems like this question might have some answers Virtual functions should not be used excessively - Why ?. In my opinion the one thing that stands out is that it just add more complexity in terms of knowing what can be done with inheritance.
Yes, it's because of performance overhead. Virtual methods are called using virtual tables and indirection.
In Java all methods are virtual and the overhead is also present. But, contrary to C++, the JIT compiler profiles the code during run-time and can in-line those methods which don't use this property. So, JVM knows where it's really needed and where not thus freeing You from making the decision on your own.
The issues is that while Java compiles to code that runs on a virtual machine, that same guarantee can't be made for C++. It common to use C++ as a more organized replacement for C, and C has a 1:1 translation to assembly.
If you consider that 9 out of 10 microprocessors in the world are not in a personal computer or a smartphone, you'll see the issue when you further consider that there are a lot of processors that need this low level access.
C++ was designed to avoid that hidden deferencing if you didn't need it, thus keeping that 1:1 nature. Some of the first C++ code actually had an intermediate step of being translated to C before running through a C-to-assembly compiler.
Java method calls are far more efficient than C++ due to runtime optimization.
What we need is to compile C++ into bytecode and run it on JVM.
From here: http://www.oodesign.com/proxy-pattern.html
Applicability & Examples
The Proxy design pattern is applicable when there is a need to control
access to an Object, as well as when there is a need for a
sophisticated reference to an Object. Common Situations where the
proxy pattern is applicable are:
Virtual Proxies: delaying the creation and initialization of expensive
objects until needed, where the objects are created on demand (For
example creating the RealSubject object only when the doSomething
method is invoked).
Protection Proxies: where a proxy controls access to RealSubject
methods, by giving access to some objects while denying access to
others.
Smart References: providing a sophisticated access to certain objects
such as tracking the number of references to an object and denying
access if a certain number is reached, as well as loading an object
from database into memory on demand.
Well, can't Virtual Proxies be created by creating an individual function (other than the constructor) for a new object?
Can't Protection Proxies be created by simply making the function private, and letting only the derived classes get the accesses? OR through the friend class?
Can't Smart References be created by a static member variable which counts the number of objects created?
In which cases should the Proxy method be preferred on the access specifiers and inheritance?
What's the point that I am missing?
Your questions are a little bit abstract, and i'm not sure i can answer at all well, but here are my thoughts on each of them. Personally i dont agree that some of these things are the best designs for the job, but that isn't your question, and is a matter of opinion.
Virtual Proxies
I don't understand what you are trying to say here at all. The point of the pattern here is that you might have an object A that you know will take 100MB, and you don't know for sure that you will ever need to use this object.
To avoid allocating memory for this object until it is needed you create a dummy object B that implements the same interface as A, and if any of its methods are called B creates an instance of A, thus avoiding allocating memory until it is needed.
Protection Proxies
Here i think you have misunderstood the use of the pattern. The idea is to be able to dynamically control access to an object. For example you might want class A to be able to access class B's methods unless condition C is true. As i'm sure you can see this could not be achieved through the use of access specifiers.
Smart Referances
Here i think you misunderstand the need for smart pointers. As this is quite a complicated topic i will simply provide a link to a question about them: RAII and smart pointers in C++
If you have never programmed in a language like C where you manage your memory yourself then this might explain the confusion.
I hope this helps to answer some of your questions.
EDIT:
I didn't notice that this was tagged c++ so i assume you do in fact recognize the need to clean up dynamic memory. The single static reference count will only work if you only intend to ever have one instance of your object. If you create 2000 instances of an object, and then deleted 1999 of them none of them would have their memory freed until the last one left scope which is clearly not desirable (That is assuming you had kept track of the locations of all the allocated memory in order to be able to free it!).
EDIT 2:
Say you have a class as follows:
class A {
public:
static int refcount;
int* allocated_memory;
A() {
++refcount;
allocated_memory = new int[100000000];
}
~A() {
if(! --refcount) {
delete [] allocated_memory;
}
}
}
And some code that uses it:
int main() {
A problem_child; // After this line refcount == 1
while(true) {
A in_scope; // Here refcount == 2
} // We leave scope and refcount == 1.
// NOTE: in_scope.allocated_memory is not deleted
// and we now have no pointer to it. Leak!
return;
}
As you can see in the code refcount counts all references to all objects, and this results in a memory leak. I can explain further if you need, but this is really a seperate question in its own right.
I am no expert, but here are my thoughts on Virtual Proxies : If we control the initialization via a separate function say bool Create(); then the responsibility and control of Initialization lies with the client of the class. With virtual proxies , the goal is to retain creation control within the class, without client being aware of that.
Protection Proxies: The Subject being protected might have different kinds of clients, ones which need to get unprotected/unrestricted access to all Subject methods and the others which should be allowed access to a subset of methods hence need for Protection proxy.
A proxy is an object behaving as a different object to add some control/behavior. A smart pointer is a good example: it accesses the object as if you would use a raw pointer, but it also controls the lifetime of that object.
It's good to question whether there are alternatives to standard solutions to problems. Design Patterns represent solutions that have worked for many folks and have the advantage that there's a good chance that an experienced programmer coming to the code will recognize the pattern and so find maintaining the code easier. However, all designs represent trade-offs and patterns have costs. So you are right to challenge the use of patterns and consider alternatives.
In many cases design is not just about making code work, but considering how it is structured. Which piece of code "knows" what. You proposal for an alternative for Virtual Prozy moves (as fizzbuzz says) knowledge of creation from the proxy to the client - the client has to "know" to call Create() and so is aware of the life-cycle of the class. Whereas with the proxy he just makes an object that he treats as the worker, then invisibly creation happens when the proxy decides it makes sense. This refactoring of responsibility into the proxy is found to valuable, it allows us in the future to change those life-cycle rules without changing any clients.
Protection proxy: your proposal requires that clients have an inheritance relationship so that they can use protected methods. Generally speaking that couples client and worker too tightly, so we introduce a proxy.
Smart reference: no, a single static count is no good. You need to count references to individual instances.
If you carefully work through each case you will find there are merits to the design patterns. If you try to implement an alternative you start with some code that sems simpler that the design pattern and then discover that as you start to refactor and improve the code, removing deuplicationand so on you end up reinventing the design pattern - and that's a really nice outcome.
I have to write a bunch of DTOs (Data Transfer Objects) - their sole purpose is to transfer data between client app(s) and the server app, so they have a bunch of properties, a serialize function and a deserialize function.
When I've seen DTOs they often have getters and setters, but is their any point for these types of class? I did wonder if I'd ever put validation or do calculations in the methods, but I'm thinking probably not as that seems to go beyond the scope of their purpose.
At the server end, the business layer deals with logic, and in the client the DTOs will just be used in view models (and to send data to the server).
Assuming I'm going about all of this correctly, what do people think?
Thanks!
EDIT: AND if so, would their be any issue with putting the get / set implementation in the class definition? Saves repeating everything in the cpp file...
If you have a class whose explicit purpose is just to store it's member variables in one place, you may as well just make them all public.
The object would likely not require destructor (you only need a destructor if you need to cleanup resources, e.g. pointers, but if you're serializing a pointer, you're just asking for trouble). It's probably nice to have some syntax sugars constructors, but nothing really necessary.
If the data is just a Plain Old Data (POD) object for carrying data, then it's a candidate for being a struct (fully public class).
However, depending on your design, you might want to consider adding some behavior, e.g. an .action() method, that knows how to integrate the data it is carrying to your actual Model object; as opposed to having the actual Model integrating those changes itself. In effect, the DTO can be considered part of the Controller (input) instead of part of Model (data).
In any case, in any language, a getter/setter is a sign of poor encapsulation. It is not OOP to have a getter/setter for each instance fields. Objects should be Rich, not Anemic. If you really want an Anemic Object, then skip the getter/setter and go directly to POD full-public struct; there is almost no benefit of using getter/setter over fully public struct, except that it complicates code so it might give you a higher rating if your workplace uses lines of code as a productivity metric.
This question already has answers here:
Closed 12 years ago.
Possible Duplicate:
How much work should be done in a constructor?
I'm strugging with some advice I have in the back of my mind but for which I can't remember the reasoning.
I seem to remember at some point reading some advice (can't remember the source) that C++ constructors should not do real work. Rather, they should initialize variables only. The advice went on to explain that real work should be done in some sort of init() method, to be called separately after the instance was created.
The situation is I have a class that represents a hardware device. It makes logical sense to me for the constructor to call the routines that query the device in order to build up the instance variables that describe the device. In other words, once new instantiates the object, the developer receives an object which is ready to be used, no separate call to object->init() required.
Is there a good reason why constructors shouldn't do real work? Obviously it could slow allocation time, but that wouldn't be any different if calling a separate method immediately after allocation.
Just trying to figure out what gotchas I not currently considering that might have lead to such advice.
I remember that Scott Meyers in More Effective C++ recommends against having a superfluous default constructor. In that article, he also touched on using methods liked Init() to 'create' the objects. Basically, you have introduced an extra step which places the responsibility on the client of the class. Also, if you want to create an array of said objects, each of them would have to manually call Init(). You can have an Init function which the constructor can call inside for keeping the code tidy, or for the object to call if you implement a Reset(), but from experiences it is better to delete an object and recreate it rather than try to reset its values to default, unless the objects is created and destroyed many times real-time (say, particle effects).
Also, note that constructors can perform initialization lists which normal functions could not.
One reasons why one may caution against using constructors to do heavy allocation of resources is because it can be hard to catch exceptions in constructors. However, there are ways around it. Otherwise, I think constructors are meant to do what they are supposed to do - prepare an object for its initial state of execution (important for object creation is resource allocation).
The one reason not to do "work" in the constructor is that if an exception is thrown from there, the class destructor won't get called. But if you use RAII principles and don't rely on your destructor to do clean up work, then I feel it's better not to introduce a method which isn't required.
Depends on what you mean by real work. The constructor should put the object into a usable state, even if that state is a flag meaning it hasn't yet been initialised :-)
The only rationale I've ever come across for not doing real work would be the fact that the only way a constructor can fail is with an exception (and the destructor won't be called in that case). There is no opportunity to return a nice error code.
The question you have to ask yourself is:
Is the object usable without calling the init method?
If the answer to that is "No", I would be doing all that work in the constructor. Otherwise you'll have to catch the situation when a user has instantiated but not yet initialised and return some sort of error.
Of course, if you can re-initialise the device, you should provide some sort of init method but, in that case, I would still call that method from the constructor if the condition above is met.
In addition to the other suggestions regarding exception handling, one thing to consider when connecting to a hardware device is how your class will handle the situation where a device is not present or communication fails.
In the situation where you can't communicate with the device, you may need to provide some methods on your class to perform later initialization anyway. In that case, it may make more sense to just instantiate the object and then run through an initialization call. If the initialization fails, you can just keep the object around and try to initialize communication again at a later time. Or you may need to handle the situation where communication is lost after initialization. In either case, you will probably want to think about how you will design the class to handle communication problems in general and that may help you decide what you want to do in the constructor versus an initialization method.
When I've implemented classes that communicate with external hardware, I've found it easier to instantiate a "disconnected" object and provide methods for connecting and setting up initial status. This has generally provide more flexibility connecting/disconnecting/reconnecting with the device.
The only real reason is Testability. If your constructors are full of "real work", that usually means the objects can only be instantiated within a fully initialized, running application. It's a sign the object/class needs further decomposition.
When using a constructor and an Init() method you have a source of error. In my experience you will encounter situation where someone forgets to call it, and you might have a subtle bug in your hands. I would say you shouldn't do much work in your constructor but if any init method is needed, then you have a non-trivial construction scenario, and it is about time to look at the creational patterns. A builder function or a factory be wise to have a look at. With a private constructor making sure that no one except your factory or builder function actually build the objects, so you can be sure that it is always constructed correctly.
If your design allow for mistakes in implementation, someone will do those mistakes. My friend Murphy told me that ;)
In my field we work with loads of similar hardware related situations. Factories gives us both testability, security and better ways of failing construction.
It is worth considering lifetime issues and connecting/reconnecting, as Neal S. points out.
If you fail to connect to a device at the other end of a link then it is often the case that the 'device' at your end is usable and will be later if the other end gets its act together. Examples being network connections etc.
On the other hand if you try and access some local hardware device that does not exist and will never exist within the scope of your program (for example a graphics card that is not present) then I think this is a case where you want to know this in the constructor, so that the constructor can throw and the object can not exist. If you don't then you may end up with an object that is invalid and will always be so. Throwing in the constructor means that the object will not exist, and thus functions can't be called on that object. Obviously you need to be aware of cleanup issues if you throw in a constructor, but if you don't in cases like this then you typically end up with validation checks in all functions that may be called.
So I think that you should do enough in the constructor to ensure you have a valid, usable, object created.
I'd like to add my own experience there.
I won't say much about the traditional debate Constructor/Init... for example Google guidelines advise against anything the in the Constructor but that's because they advise against Exceptions and the 2 work together.
I can speak about a Connection class I use though.
When the Connection class is created, it will attempt to actually connect itself (at least, if not default constructed). If the Connection fails... the object is still constructed and you don't know about it.
When you try to use the Connection class you are thus in one of 3 cases:
no parameter has ever been precised > exception or error code
the object is actually connected > fine
the object is not connected, it will attempt to connect > this succeeds, fine, this fails, you get an exception or an error code
I think it's quite useful to have both. However, it means that in every single method actually using the connection, you need to test whether or not it works.
It's worth it though because of disconnection events. When you are connected, you may lose the connection without the object knowing about it. By encapsulating the connection self-check into a reconnect method that is called internally by all methods needing a working connection, you really isolate the developers from dealing with the issues... or at least as much as you can since when everything fails you have no other solution that letting them know :)
Doing "real work" in a constructor is best avoided.
If I setup database connections, open files etc inside a constructor and if in doing so one of them raise an exception then it would lead to a memory leak. This will compromise your application's exception safety.
Another reason to avoid doing work in a constructor is that it would make your application less testable. Suppose you are writing a credit-card payment processor. If say in CreditCardProcessor class's constructor you do all the work of connecting to a payment gateway, authentate and bill the credit card how do I ever write unit tests for CreditCardProcessor class?
Coming to your scenario, if the routines that query the device do not raise any exceptions and you are not going to test the class in isolation then there is its probably preferable to do work in the constructor and avoid calls to that extra init method.
There are a couple reasons I would use separate constructor/init():
Lazy/Delayed initialization. This allows you to create the object quickly, fast user response, and delay a more lengthy initialization for later or background processing. This is also a part of one or more design patterns concerning reusable object pools to avoid expensive allocation.
Not sure if this has a proper name, but perhaps when the object is created, the initialization information is unavailable or not understood by whoever is creating the object (for example, bulk generic object creation). Another section of code has the know-how to initialize it, but not create it.
As a personal reason, the destructor should be able to undo everything the constructor did. If that involves using internal init/deinit(), no problem, so long as they are mirror images of each other.
I have previously asked about the proper way of accessing member variables present in the project. In the project, I have CWinapp-derived class, CMainFrm class, a list of different view classes. However, currently, I have instances of different user-defined classes instantiated in the CWinApp-derived class, while the rest of the classes use a pointer obtained from AfxGetApp() function, and then access the different user-defined classes. I was told by some community members on the MFC newsgroup that this is a very bad design (i.e. the parent should not know anything about an app-class, view class, or document class). However, I'm not sure how otherwise I can access various user-defined classes without using this design. It would be great to hear some suggestions as I'm not familiar enough with MFC to come up with proper search terms.
"(i.e. the parent should not know anything about an app-class, view class, or document class)"
I'm not sure I understand this sentence, what do you mean with 'parent' here?
Anyway, in my opinion, the design you describe isn't really a problem. It's a trade off: do you either pass these classes to all functions that need them, complicating their use and API, or do you store them as a sort of global variables like you're doing? It depends on the data that is accessed, and how often. Data that is needed in many places can just as well be 'global'.
There are multiple ways of making data 'global': make it a member of CWinApp (that is, your CWinApp-derived class), or of CMainFrame, or do you make an actual 'global variable', or do you make a singleton, ...
The problem with global variables is that it becomes hard to figure out who accesses it when and from where. If you data as a member of CWinApp, you can access it through an accessor function and trace access from there (through log messages, break point, ...) This, in my opinion, mitigates most of the problems associated with global variables. What I usually do nowadays is use a Loki singleton.
The reason that is stated in your post for not making data a member of CWinApp, as a decoupling issue, is (in the context that you've presented it) a bit strange imo. If certain classes need access, they'll need to know of those data structures anyway, and their storage location is irrelevant. Maybe it's just because I don't know about the specifics of your design.