I use the Builder pattern for several classes of my project (multiple arguments, some mandatory, some optional, etc.). These classes are immutable (no setters, deep copy for collections getters).
I am now trying to store those objects in a database using a persistence framework that constructs objects with default constructor + setters. It doesn't like my Builders very much!
I don't want to degrade that setup to POJOs and lose the advantages of the current design (flexibility, immutability, construction security).
I would welcome any feedback on workarounds that can be used in this situation (I could wrap each of these classes but that would double the number of classes and I would rather avoid that).
One post actually points that as a specific disadvantage of the Builder pattern.
EDIT
One answer proposes to use private constructor / setters but that only works if the fields of the class are not final, which is not my case.
FINAL EDIT
Thanks to all.
What I think will be my final solution looks like this and works fine (for the record, I'm using MongoDB + Morphia):
class AClass {
private final String aField;
private final AClass() {
aField = "";
}
//Standard builder pattern after that - no setters (private or public)
}
As I said in my comment: you can include a default constructor and all required setters but make them private. This way you maintain the immutability of your Objects but an ORM such as Hibernate will be able to access the methods/constructor when it needs to.
Anybody else would be able to access these methods using reflection too, but then they can access the private member variables using reflection too. So there is no real downside to adding the private methods.
Related
This question already has answers here:
Are getters and setters poor design? Contradictory advice seen [duplicate]
(16 answers)
Closed 9 years ago.
I have been going through clean code book which states that the class should not expose the internal state of its data and only should be exposing the behavior. In case of a very simpl and dumb java bean exposing the internal state which getter's and setters, is it not worth just removing them and make the private members public? Or just treat the class as a data structure?
I don't think so. It depends of the lifetime of your Object and its "exposure" (external modification).
If you're only using it as a data structure, exposing fields in secure way (final) sounds enough:
public class Person {
public final String firstName;
public final String lastName;
public Person(String firstName, String lastName) {
this.firstName = firstName;
this.lastName = lastName;
}
}
The term POJO was intended to distinguish classes from JavaBeans or any other convention. As such a POJO is by definition NOT required to do anything.
I have been going through clean code book which states that the class should not expose the internal state of its data and only should be exposing the behavior.
This is called encapsulation and a good principle.
In case of a very simpl and dumb java bean exposing the internal state which getter's and setters, is it not worth just removing them and make the private members public?
That is an alternative approach. Some projects may forbid this approach while others may encourage it. Personally, I would favour this approach for classes which are encapsulated in some way already e.g. they are package local.
There is a view that some day in some way your class might have additional requirements and changing the "API" will be impossible. This goes against the YAGNI principle and very rarely proves to be the case and when it does has a much lower cost than adding lots of methods which don't do anything.
However, this is not always the case and if you don't use accessor methods you should consider what the impact will be on the project if you have to change it later. Using accessor methods every where means you never need to worry about this.
In summary, if you are pretty sure accessor methods are pointless and it won't be a problem to add them later, I would say you should use your judgement. However if you are not sure if it could be a problem in the future or you don't want to have to worry about it, use accessor methods.
The definition of POJO doesn't mandate getter/setter.
Experimentally, I am not using getter and setter in my current project.
The approach I am taking is this one:
unless necessary, don't provide getter/setter.
So far, I didn't find a case where I really needed get/set.
Some friend told me: "having get/set is helpful if in the future you need xyz"; my reply has been: when -in the future- I need to do so, I will provide the getter and setter; I don't want to anticipate anything.
The objection about incapsulation, that some may raise, is not really a valid one: providing getter and setter breaks incapsulation in the same manner, plus you have additional lines of (useless) code. Bugs may also lay in getter and setters.
This is an example of one of a non-trivial domain class:
public class SSHKey implements IsSerializable {
public Long id;
public Long userId;
public String type;
public String bits;
public String fingerprint;
public String comment;
#SuppressWarnings("unused")
private SSHKey() { // required by gwt-rpc
}
public SSHKey(String text) throws InvalidSSHKeyException {
Ensure.that(text != null, new InvalidSSHKeyException("Invalid Key"));
text = text.trim();
String[] parts = text.split(" ", 3);
Ensure.that(parts.length >= 2,
new InvalidSSHKeyException("Invalid Key"));
type = getType(parts);
Ensure.that(type.equals("ssh-rsa") || type.equals("ssh-dss"),
new InvalidSSHKeyException(
"Key must start with 'ssh-rsa' or 'ssh-dss'"));
bits = getBits(parts);
comment = getComment(parts);
}
private String getBits(String[] parts) {
return parts[1];
}
private String getComment(String[] parts) {
if (parts.length == 3)
return parts[2];
return type + " " + bits.substring(0, min(15, bits.length())) + "...";
}
private String getType(String[] parts) {
return parts[0];
}
}
The constructor takes the responsibility to validate and prepare the data to be manageable. Thus this logic doesn't need to be in a setter/getter.
If I was shown object with public members some years ago, I would probably not like them; maybe I am doing something wrong now, but I am experimenting and so far it is ok.
Also, you need to consider if your class is designed to be extended or not (so, foresee the future is part of the requirements), and if you want your object to be immutable. Those things you can only do with get/set.
If your object must be immutable, and you can avoid the empty constructor, you can just add 'final' to the member instances, btw.
Unfortunately I had to add IsSerializable (similar to java.io.Serializable) and an empty constructor since this is required by gwt. So, you could tell me then "you see? you need the getter an setter"; well not so sure.
There are some jdbc frameworks which promote the use of public fields btw, like http://iciql.com
This doesn't imply that this project is correct, but that some people are thinking about it.
I suppose that the need of getter/setter is mostly cultural.
The issue with making the members accessible is that you no longer control them from inside the class.
Let's say that you make Car.speed accessible. Now, everywhere in you program there can be some reference to it. Now, if you want to make sure that speed is never set a negative value (or to make the change synchronized because you need to make it thread safe), you have to either:
in all the points where speed is accessible, rewrite the program to add the control. And hope that everybody that changes the program in the future remembers to do so.
make the member private again, create the getter and setter methods, and rewrite the program to use them.
Better get used to write getter and setter from the beginning. Nowadays, most IDEs do it automatically for you, anyway.
The canonical answer to this is: You don't know whether your simple data structure will stay so simple in the future. It might evolve more than you expect now. It might be also possible, that anytime soon you want some "value changed" observer in that bean. With getter and setter methods you can do this very simply later without changing you existing codebase.
Another pro point for getter/setter is: If in Rome, do like the Romans... Which means in this case: Many generic frameworks expect getter/setter. If you don't want to rule all these usefulls frameworks out right from the start then do you and your colleagues a favour and simply implement standard getter/and setters.
Only if you expose a class in a library that's used beyond your control.
If you do release such a library, the Uniform Access Principle dictates that you should use getters and setters in order to be able to change the underlying implementation later without requiring clients to change their code. Java doesn't give you other mechanisms to do this.
If you use this class in your own system, there's no need: your IDE can easily encapsulate a public field and update all its usages in one safe step. In this case, brevity wins, and you lose nothing for the time where you need encapsulation.
I think it's a good idea to use getters and setters, unless you have very specific speed/memory/efficiency requirements or very simple objects.
A good example is a Point, where it is probably both nicer and more efficient to expose it's .x and .y variables.
That said, it will actually not be a big effort to change the visibility of a few member variables and introduce getters and setters even for a large codebase, if you suddenly require some logic in a setter.
JavaBeans require getters and setters. POJOs do not, anyway this has its benefits
The objetive of the getters and setters is to achieve encapsulation, which manages the internal state of object. This allows you to add or change business rules in your application after the application has been implemented only change the getter or setter code, example, if you have a text field that only allows for more than 3 characters can check before assigning it to an attribute and throw an exception, other reason for not doing this is if it's possible you'll want to change the implementation or change variable names or something like. This cannot be enforced if the field is publicly accessible and modifyable
anyway you can use your IDE to generate setters and getters.
If you are developing a simple application can be recommended, if your application is complex and must give maintenance is not recommend.
for the data-type objects, like POJO / PODS / JavaBean, at python you have only public members
you can set those and get those easily, without generating boilerplate setter and getter code(in java these boilerplate code usually(98%) exposes the inner private tag as noted in the question)
and at python in the case you would need to interact with a getter, then you just define extra code only for that purpose
clean and effective at the language level
at java they chose the IDE development instead of changing base java, see JavaBean e.g. how old that is and java 1.0.2 is how old...
JDK 1.0 (January 23, 1996)
The EJB specification was originally developed in 1997 by IBM and later adopted by Sun Microsystems (EJB 1.0 and 1.1) in 1999
so just live with it, use the setter getter because those are enforced by java surroundings
That's the true what #Peter Lawrey explains about encapsulation.
Only one note: it's more important, when you are working with complex objects (for example in the domain model in a ORM project), when you have attributes that aren't simple Java types. For example:
public class Father {
private List childs = new ArrayList();
public Father() {
// ...
}
private List getChilds() {
return this.childs;
}
public void setChilds(List newChilds) {
this.childs = newChilds;
}
}
public class Child {
private String name;
// ...
private String getName() {
return this.name;
}
public void setName(String newName) {
this.name = newName;
}
}
If you expose one attribute (like the childs attribute in the Father class) as a public, you won't be able to identify what part of your code are setting or changing one property of your exposed attribute (in the case, for example, adding new Child to a Father or even changing the name of a existing Child). In the example, only a Father object can retrieve the childs content and all the rest of the classes can change it, using its setter.
This isn't about a recent problem, more like one from history which I think I just found at least one answer to. I was working in a shop wherein the Architect insisted on having many fields of our entities read-only, i.e. without setters of any sort. I mean we were using Hibernate and obviously you can annotate the fields as read only...but his requirement was to eliminate the setters so they couldn't ever be set without using Reflection. He even wrote a suite of utilities that used Reflection to set those fields for testing and you had to extend them if there were changes to your schema. My problem with this was that it was dog slow and a big bunch of seemingly unnecessary code that was always changing. Mocks could be used of course but those can get painful as well if your data is non-trivial. Now I just found something that looks like a good answer but I would like to hear from the community as it may not be the best answer. For one thing, I question if we should have even been doing that.
You have several ways:
mocks - the simplest, e.g. Mockito library;
builders - setters only + validation on build;
package-local setters + same-package factories in tests scope.
Do NOT use:
100500 constructor arguments - this way leads to extremely hard-to-support code.
Builders example:
interface Model (
A getA();
...
)
public final class ModelBuilder {
private ModelBuilder() {...}
public static ModelBuilder newBuilder() {...}
public ModelBuilder setA(A a) (...) // keep 'a', return 'this'
...
public Model build() {...}
}
I have a class that has many settable/gettable attributes. I'd like to use reflection to set these attributes, but I have 2 questions about my implementation
Here is some stripped down code from my class
class Q {
public String question_1;
public String question_2;
public String question_3;
public String answer_1;
public String answer_2;
public String answer_3;
//etc. etc. Many String attributes
// … constructor and other stuff are omitted
// here is my method for "dynamically" setting each attribute
public void set_attribute(String a_raw_string, String my_field) {
try {
Class cls = Class.forName("com.xyz.models.Q");
Field fld = cls.getField(my_field);
fld.set(this, a_raw_string);
}
catch (Throwable e) {
System.err.println(e);
}
}
I then set various fields like this:
Q q = new Q();
q.set_attribute("abcde", "question_1");
q.set_attribute("defgh", "question_2");
// etc.
This works (i.e., the instance variables are set when I call set_attribute.
However, they only work when the instance variables are declared public. When they are declared private I get a NoSuchFieldException
QUESTION 1: Why do I get that error when the fields are private? My naive assumption is that since the set_attribute function is part of the class, it should have unfettered access to the instance variables.
QUESTION 2: I think I may be overthinking this problem (i.e., I shouldn't be using reflection to set variables in this way). Is there a more recommended approach?
The reason that I want to use reflection is because it's a pain in the ass to declare a ton of setter methods…so I'm wondering if someone has solved this annoyance in a better way.
Thanks!
I think I may be overthinking this problem (i.e., I shouldn't be using reflection to set variables in this way)
Yep. Reflection is fairly slow and should only be used as a last resort. If this is simply to avoid having so much redundant code, consider using automatic code generation. For pure data objects, I would strongly recommend using protocol buffers; it will generate the getters / setters (you only need to declare the fields). Plus it allows for easy communication of the data between C++, Java, and Python.
If you have a class that has a lot of fields but isn't a pure data object... well
You should consider whether all the fields should be mutable. (Do you really need setters?)
Whether the fields should even be visible. (Do you need any accessors at all?)
It is often a good idea to make fields "final", initialize them in the constructor(s), and provide no access or provide limited access through an implemented interface.
Using setter methods is the accepted way to set values for class member variables, reflection should definitely not be used for that as the code will be harder to understand and run much more slowly.
Most IDEs (eg Eclipse or NetBeans) include tools for automatically creating getter and setter methods for a class's fields.
When they are private you need to call fld.setAccessible(true);
Yes, why don't you just set the fields directly and avoid reflection? It doesn't look like you're doing anything dynamic. It's just that they are private -- why? Perhaps you mean to expose getters/setters and make the fields private? If so, then you should just invoke the public setters.
I want to force future users of a class to access a private member also from future code written in that class only through an accessor method (even through junit or anything like that).
is there a way to do it in java? can someone show an example if possible?
You cannot force to do that, but you can create a method and document that enforcement in the javadoc.
private int myMember;
/**
* ATTENTION: use this method instead of setting the member directly.
*/
public void setMyMember(int value) {
this.myMember = value;
}
Also, there is an alternative solution which might work. Use ThreadLocal, like this:
private final ThreadLocal<String> member = new ThreadLocal<String>();
public void setMember(final String value) {
member.set(value);
}
The member field is final and cannot be changed. Therefore, clients will be forced to call the setter directly.
As long as the field is part of the class, anyone can access it directly. This can be a problem when we try to force all (co-)authors to go through the getters/setters because those method do some conversion, checking or bookkeeping stuff. Like incrementing internal counters.
A general solution, that comes to mind: it could be possible by using annotations. You'd have to create an annotation (and the annotation processor code) to ensure, that it is a compile time error if the field is used outside of it's getter/setter method:
#GetterSetterAccessOnly
private int value;
If you want to prevent reflection you can use a SecurityManager. If this is not an option you can get the call stack with Thread.currentThread().getStackTrace() and check the caller is from your class.
Two problems with this are; the performance won't be great. Anything you can do in the method you can do externally so the simplest work around is to copy its contents without the check. ;)
I suggest you document your reason for not allowing access this way.
Make use of inheritance to hide the field:
Do your class with all the fields and getter/setters that you need. (You may make it abstract)
Do a child class, that inherits from the previews one, and since the field isn't accessible, you force the use of the getter/setter pair.
As you are talking about the accessing in the same class, they have all the freedom to access the private member directly as well as through accessor method if that member has. So basically you cannot prevent them using the member directly in the same class.
I am afraid there is no standard way to do that. If a user has access to a class instance, although a private member is declared private, permissions can be changed at runtime and accessed anyway.
You need a classloader that enforces permissions. You can make an OSGi Bundle and enforce a control policy over the instance of your objects exported as services through interfaces. However this will tie you to an OSGi container to run your application.
Well, if they have access to your code, they can do anything they want. In the worst case, they remove your getters and setters and just put in a public field instead ;)
But of course you can motivate them to not access the variable directly with an according design:
you should check whether the others should rather implement subclasses instead of changing the class itself. Then private fields are, of course, only accessed via setters and getters.
you could move the data into a different class, and use your getters and setters to access the data in the other class. Doing this just for the sake of not having the data directly in your class is maybe a bit counter-intuitive, but since you probably have a good reason why they shouldn't access that very data, it indicates a different responsibility. So refactoring to meet the SRP is a good idea anyways.
I have a class that has many settable/gettable attributes. I'd like to use reflection to set these attributes, but I have 2 questions about my implementation
Here is some stripped down code from my class
class Q {
public String question_1;
public String question_2;
public String question_3;
public String answer_1;
public String answer_2;
public String answer_3;
//etc. etc. Many String attributes
// … constructor and other stuff are omitted
// here is my method for "dynamically" setting each attribute
public void set_attribute(String a_raw_string, String my_field) {
try {
Class cls = Class.forName("com.xyz.models.Q");
Field fld = cls.getField(my_field);
fld.set(this, a_raw_string);
}
catch (Throwable e) {
System.err.println(e);
}
}
I then set various fields like this:
Q q = new Q();
q.set_attribute("abcde", "question_1");
q.set_attribute("defgh", "question_2");
// etc.
This works (i.e., the instance variables are set when I call set_attribute.
However, they only work when the instance variables are declared public. When they are declared private I get a NoSuchFieldException
QUESTION 1: Why do I get that error when the fields are private? My naive assumption is that since the set_attribute function is part of the class, it should have unfettered access to the instance variables.
QUESTION 2: I think I may be overthinking this problem (i.e., I shouldn't be using reflection to set variables in this way). Is there a more recommended approach?
The reason that I want to use reflection is because it's a pain in the ass to declare a ton of setter methods…so I'm wondering if someone has solved this annoyance in a better way.
Thanks!
I think I may be overthinking this problem (i.e., I shouldn't be using reflection to set variables in this way)
Yep. Reflection is fairly slow and should only be used as a last resort. If this is simply to avoid having so much redundant code, consider using automatic code generation. For pure data objects, I would strongly recommend using protocol buffers; it will generate the getters / setters (you only need to declare the fields). Plus it allows for easy communication of the data between C++, Java, and Python.
If you have a class that has a lot of fields but isn't a pure data object... well
You should consider whether all the fields should be mutable. (Do you really need setters?)
Whether the fields should even be visible. (Do you need any accessors at all?)
It is often a good idea to make fields "final", initialize them in the constructor(s), and provide no access or provide limited access through an implemented interface.
Using setter methods is the accepted way to set values for class member variables, reflection should definitely not be used for that as the code will be harder to understand and run much more slowly.
Most IDEs (eg Eclipse or NetBeans) include tools for automatically creating getter and setter methods for a class's fields.
When they are private you need to call fld.setAccessible(true);
Yes, why don't you just set the fields directly and avoid reflection? It doesn't look like you're doing anything dynamic. It's just that they are private -- why? Perhaps you mean to expose getters/setters and make the fields private? If so, then you should just invoke the public setters.