We Keep Coding

What Design Pattern to use when writing a PHP function which processes the output from another function?

I have two functions: myFunctionA() and myFunctionB().
myFunctionA() returns an Object which includes the key Page_Type which has a string value.
myFunctionB() processes a number of entries in the Object returned by myFunctionA(), including Page_Type and its string value.
Later, myFunctionA() is updated so it no longer returns an object including the key Page_Type but, instead, the key Page_Types - which has an array value.
Because of this, myFunctionB() will now also need to be updated - it will no longer be processing Page_Type which is a string, but Page_Types which is an array.
If I understand correctly (and I may not), the above is an example of Dependency Request and the extensive refactoring that it throws up can be avoided by (I think) deploying the Dependency Injection pattern (or possibly even the Service Locator pattern?) instead (??)
But, despite reading around this subject, I am still uncertain as to how Dependency Injection can work in PHP or Javascript functions (much of the explanation deals with programming languages like C++ and OOP concepts like Classes, whereas I am dealing with third party functions in PHP and javascript).
Is there any way to structure my code so that updating myFunctionA() (in any significant manner) will not then require me to also update myFunctionB() (and all other functions calling myFunctionA() - for instance myFunctionC(), myFunctionD(), myFunctionE() etc.) ?
And what if myFunctionH() requires myFunctionG() requires myFunctionF() which requires myFunctionA()? I don't want to be in a position where updating myFunctionA() now means that three more functions (F, G and H) all have to be updated.
Attempt at an answer:
The best answer I can think of at present - and this may not be a best practice answer because I don't know yet if there is a formal problem which corresponds with the problem I am describing in the question above - is the following restatement of how I presented the setup:
I have two (immutable) functions: myFunctionA__v1_0() and myFunctionB__v1_0().
myFunctionA__v1_0() returns an Object which includes the key
Page_Type which has a string value.
myFunctionB__v1_0() processes a number of entries in the Object
returned by myFunctionA__v1_0(), including Page_Type and its
string value.
Later, myFunctionA__v1_0() still exists but is also succeeded by myFunctionA__v2_0() which
returns an object including the key Page_Types - which has an array value.
In order for myFunctionB to access the object returned by myFunctionA__v2_0(),
there will now also need to be a myFunctionB__v1_1(), capable of processing
the array Page_Types.
This can be summarised as:
myFunctionB__v1_0() requires object returned by myFunctionA__v1_0()
myFunctionB__v1_1() requires object returned by myFunctionA__v2_0()
Since each function becomes immutable after being formally named, what never happens is that we end up with an example of myFunctionB__v1_0() requiring object returned by myFunctionA__v2_0().
I don't know if I am approaching this the right way, but this is the best approach I have come up with so far.

It is very common in programming for the provider - ie. myFunctionA() to know nothing about its consumer(s) myFunctionB(). The only correct way to handle this is to define an API up front and never change it ;)
I don't see the purpose of versioning the consumer - that reason would have to be "downstream" of myFunctionB() - ie. a consumer of myFunctionB(), that the author of myFunctionB() is not in control of... in which case myFunctionB() itself becomes a provider, and the author would have to deal with that (perhaps using the same pattern that you do)... But it's not your problem to deal with.
As for your provider myFunctionA(): If you cannot define an interface / API for the data itself up front - ie. you know that the structure of the data will have to change (in a non-backwards-compatible way), but you don't know how... then you will need to version something one way or another.
You are miles ahead of most since you see this coming and plan for it from the beginning.
The only way to avoid having to make changes to the consumer myFunctionB() at some point, is to make all changes to the provider myFunctionA() in a backwards-compatible way. The change you describe is not backwards compatible because myFunctionB() cannot possibly know what to do with the new output from myFunctionA() without being modified.
The solution you propose sounds like it should work. However, there are at least a couple of downsides:
It requires you to keep an ever-growing list of legacy functions around in case there are ever any consumers requesting their data. This will become very complicated to maintain, and likely impossible in the long run.
Depending what changes need to be made in the future it might no longer be possible to produce the output for myFunctionA__v1_0() at all - in your example you add the possibility of several page_types in your system - in this case you can probably just rewrite v1_0 to use the first and the legacy consumers will be happy. But if you decide to completely drop the concept of page_types from your system, you would have to plan for the complete removal of v1_0 one way or another. So you need to establish a way to communicate this to the consumers.
The only correct way to handle this still is to define an API up front and never change it.
Since we have established that:
you will have to make backwards incompatible changes
you don't know anything about the consumers and you don't have the power to change them when needed
I propose that instead of defining an immutable API for your data, you define an immutable API that allows you to communicate to consumers when they should or must upgrade.
This might sound complicated, but it doesn't have to be:
Accept a version parameter in the provider:
The idea is to let the consumer explicitly tell the provider which version to return.
The provider might look like this:
function myFunctionA(string $version) {
$page_types = ['TypeA', 'TypeB'];
$page = new stdClass();
$page->title = 'Page title';
switch ($version) {
case '1.0':
$page->error = 'Version 1.0 no longer available. Please upgrade!';
break;
case '1.1':
$page->page_type = $page_types[0];
$page->warning = 'Deprecated version. Please upgrade!';
break;
case '2.0':
$page->page_types = $page_types;
break;
default:
$page->error = 'Unknown version: ' . $version;
break;
}
return $page;
}
So the provider accepts a parameter which will contain the version that the consumer can understand - usually the one that was the newest when the consumer was last updated.
The provider makes a best-effort attempt to deliver the requested version
if it is not possible there is a "contract" in place to inform the consumer ($page->error will exist on the return value)
if it is possible, but there is a newer version available another "contract" is in place to inform the consumer of this ($page->warning will exist on the return value).
And handle a few cases in the consumer(s):
The consumer needs to send the version it expects as a parameter.
function myFunctionB() {
//The consumer tells the provider which version it wants:
$page = myFunctionA('2.0');
if ($page->error) {
//Notify developers and throw an error
pseudo_notify_devs($page->error);
throw new Exception($page->error);
} else if ($page->warning) {
//Notify developers
pseudo_notify_devs($page->warning);
}
do_stuff_with($page);
}
The second line of an older version of myFunctionB() - or a completely different consumer myFunctionC() might instead ask for an older version:
$page = myFunctionA('1.1');
This allows you to make backwards-compatible changes any time you want - without consumers having to do anything. You can do your best to still support old versions when at all possible, providing "graceful" degradation in legacy consumers.
When you do have to make breaking changes you can continue supporting the old version for a while before finally removing it completely.
Meta information
I'm not confident this would be useful... but you could add some meta information for consumers using an outdated version:
function myFunctionA(string $version) {
# [...]
if ($page->error || $page->warning) {
$page->meta = [
'current_version' => '3.0',
'API_docs' => 'http://some-url.fake'
]
}
return $page;
}
This could then be used in the consumer:
pseudo_notify_devs(
$page->error .
' - Newest version: ' . $page->meta['current_version'] .
' - Docs: ' . $page->meta['API_docs']
);
...if I were you I would be careful not to overcomplicate things though... Always KISS

Dependency injection is more relevant in a OOP context. However, the main thing I would do here is to stop thinking in terms of returning what you have available and start considering how the 2 methods work together and what's their contract.
Figure out what's the logical output for myFunctionA(), encode that contract into an object, and convert the data you have to that format. This way, even if the way you fetch data in myFunctionA() changes, you only have to update that one conversion.
As long as you adhere to that contract (which can represented through a custom object), myFunctionB() and other methods that expect to receive data as per the contract, you won't have to change those methods any longer.
So my main take away here would be to start thinking about the data you need and pass it around not in the structure you receive it, but in the way that it makes the most sense for your application.

It seems you have broken the interface between myFunctionA and myFunctionB by changing the return type from string to array.
I don't think DI can be of help.

Firstly, this is not Dependency Injection.
Your myFunctionA() could be call a Producer since it provide data, it should be proved a Data Structure.
Your myFunctionB() could be call a Consumer since it consume the data that provided by myFunctionA.
So in the order to make your Producers and your Consumer working independently, you need to add another layer between them, call Converter. The Converter layer will convert the Data Structure that provided by Producer to be a well know Data Structure that can understand by a Consumer
I really recommend you to read the book Clean Code chapter 6: Objects and data structures. So you will able to fully understand the concept above, about Data Structure
Example
Assume we had Data Structure call Hand, had property right hand and left hand.
class Hand {
private $rightHand;
private $leftHand
// Add Constructor, getter and setter
}
myFunctionA() will provide object Hand, Hand is a Data Structure
function myFunctionA() {
$hand = Hand::createHand(); //Assume a function to create new Hand object
return $hand;
}
let said we had another Data Structure, call Leg, the Leg will able to consume by myFunctionB();
class Leg {
private $rightLeg;
private $leftLeg
// Add Constructor, getter and setter
}
Then, we need to had a converter, in the middle , to convert from Hand to Leg, and use on myFunctionB()
class Converter {
public static function convertFromHandToLeg($hand) {
$leg = makeFromHand($hand); //Assume a method to convert from Hand to Leg
return $leg;
}
}
myFunctionB(Converter::convertFromHandToLeg($hand))
So, whenever you edit the response of myFunctionA(), mean you are going to edit Data Structure of Hand. You only need to edit the Converter to make sure it continue to convert from Hand to Leg correctly. You do not need to touch myFunctionB and vice versa.
This will very helpful when you had another Producer that will provide Hand like you mention on your question, myFunctionC(), myFunctionD()... And you also had many another Consumer that will consume Leg like myFunctionH(), myFunctionG()...
Hope this help

Is it memory inefficient to instantiate classes in a getter method? What OOP design prevents this?

I have a getter method that instantiates classes from deserialized data. I suspect this is a terribly inefficient way of doing things
class Container {
get content() {
let contentList = []
for (let c of this._content) { // this._content is an array of objects
contentList.push(new ContentTypes[c.type](c)) //ContentTypes is a mapping between type names and classes
}
return contentList
}
}
What's the appropriate design pattern in OOP land?

Whether the code is in a getter or a method, the efficiency doesn't change. But what changes is the expectation of the person writing code using your class.
I don't expect accessing a property on an object to incur much overhead. So when using an instance of Container, if I see a contents property, I'm not going to expect it to do a lot of work. Yours does (a bit).
I would either:
Keep a copy of what you return (making what you return immutable) and only discard it when the things it relies on change, or
Make it a method (explicitly).
(Or I might do both.)
I expect method calls to incur more overhead than property accesses. I don't know if there's a specific OOP principle one can cite about that, but that's my view from experience (for what it may or may not be worth).
But, note that there are counterexamples out in the wild. A famous and directly applicable one is the innerHTML property of DOM elements, which is an accessor property that does a fair bit of work when you use it.
Here's an example of returning an immutable thing you keep and re-return, but invalidate when its underlying data changes:
class Container {
constructor() {
this._content = [];
this._exposedContent = null;
}
addContent(content) {
this._content.push(content);
this._exposedContent = null; // Invalidating the cached copy if any
}
get content() {
// Do we have a cached copy?
if (!this._exposedContent) {
// No, create it
this._exposedContent = this._content.map(
c => Object.freeze(new ContentTypes[c.type](c))
);
}
return this._exposedContent;
}
}
(You might also expire the copy after a period of time.)
But it's a judgement call whether it's worth the memory overhead and complexity vs. what your code actually in your content accessor. My default position would be to use a method instead (getContent or similar) and not cache, but it's a judgement call.

Benefit of using createSelector when there is only one parent selector

Suppose we have a simple redux store:
{
user: {
age: 10
}
}
And we have a selector to get user:
const getUser = state => state.get('user') // suppose we used immutable.js
now we have two approaches to get user age:
use createSelector and use getUser as parent selector:
const getAge = createSelector(
getUser,
user => user.get('age')
)
use existing selector directly:
const getAge = state => getUser(state).get('age');
Question:
Which one is better, regarding performance, memorization, etc.
Another derived question:
The example I proposed here is very simple, let's consider when it comes to a very large and complicate redux store, will the difference of two approaches matter?

If the selector's properties change often, it would make sense to keep them on separate selectors. At least in the case of selectors where usually the object changes as a whole, I prefer to create one selector for the user.
As an example, let's say we have the object:
const profile = {
name : 'John',
email,
timezone,
...
locale : 'en',
}
If a <Translate> component is created that is tightly coupled to the locale, but not to other parts of the profile such as name, etc., then a selector for getLocale() makes sense (and is how we implement our locale component).
On the other side, if you have a Profile Page where users can edit various fields, then that page most likely needs to re-render frequently when changes are made, so it would make sense to just have a getProfile() selector on that page.
We handle a quite complex and large Redux state, and for example, because we use the currentUserId, or userName all over the place in our web application, we have dedicated selectors for these to take advantage of memoization.
But, for a quiz or assignment, we use a single selector getQuizContents() instead of its internal fields, as we know that most components that need information about a quiz, don't need just the name but many internal fields such as date, number of questions, etc.

As I understand for simple consistense and primitive results you shouldn't use createSelector at all, use it for expensive calculation only or when the new result is not shallow equal the old one.
example when you shouldn't use memoization:
export const getUserIds = state => state.ids; // always the same array
example when you should use memoization:
export const getUserIds = state => state.ids.map(id => id.toString) -- every time the new array because of map fn
Your code is returning simple raw values, so you don't need any memoization.
Fix me somebody if I am wrong

It is worth to use underscore/lodash in shouldComponentUpdate?

I have a situation in my project that I need to check if props are equal to prevent redundant invocation of render method and increase performance. The problem is that I can't use React.PureComponent because properties haven't got flat structure.
{
"header": "HeaderName",
"data": [{...}, {...}],
"configuration": {....}
}
I implemented shouldComponentUpdate method with lodash.
shouldComponentUpdate({data: newData, configuration: newConfiguration}) {
const {data, configuration} = this.props;
return !_.isEqual(data, newData) || !_.isEqual(newConfiguration, configuration);
}
It is worth to use lodash/underscore isEqual methods to check equality of properties in React components ? It is significant decrease of application performance ? What is the best solution of this problem (I can't change structure of properties) ?

underscore and lodash are going to be slow.
I try to keep state properties to be primitives as much as possible for performance and maintainability. Is there any way you can abstract out something from your data that would identify just the data you need to pass down to your child components? For example, in this sandbox, the data is not on state but rather one of the keys is a state property. The actual data is just an instance member, but the data to pass down still changes with user actions because the identifier is changing.
This might sound confusing - checking out the code in the sandbox will clarify.

Are getter/setters really useful? [duplicate]

What's the advantage of using getters and setters - that only get and set - instead of simply using public fields for those variables?
If getters and setters are ever doing more than just the simple get/set, I can figure this one out very quickly, but I'm not 100% clear on how:
public String foo;
is any worse than:
private String foo;
public void setFoo(String foo) { this.foo = foo; }
public String getFoo() { return foo; }
Whereas the former takes a lot less boilerplate code.

There are actually many good reasons to consider using accessors rather than directly exposing fields of a class - beyond just the argument of encapsulation and making future changes easier.
Here are the some of the reasons I am aware of:
Encapsulation of behavior associated with getting or setting the property - this allows additional functionality (like validation) to be added more easily later.
Hiding the internal representation of the property while exposing a property using an alternative representation.
Insulating your public interface from change - allowing the public interface to remain constant while the implementation changes without affecting existing consumers.
Controlling the lifetime and memory management (disposal) semantics of the property - particularly important in non-managed memory environments (like C++ or Objective-C).
Providing a debugging interception point for when a property changes at runtime - debugging when and where a property changed to a particular value can be quite difficult without this in some languages.
Improved interoperability with libraries that are designed to operate against property getter/setters - Mocking, Serialization, and WPF come to mind.
Allowing inheritors to change the semantics of how the property behaves and is exposed by overriding the getter/setter methods.
Allowing the getter/setter to be passed around as lambda expressions rather than values.
Getters and setters can allow different access levels - for example the get may be public, but the set could be protected.

Because 2 weeks (months, years) from now when you realize that your setter needs to do more than just set the value, you'll also realize that the property has been used directly in 238 other classes :-)

A public field is not worse than a getter/setter pair that does nothing except returning the field and assigning to it. First, it's clear that (in most languages) there is no functional difference. Any difference must be in other factors, like maintainability or readability.
An oft-mentioned advantage of getter/setter pairs, isn't. There's this claim that you can change the implementation and your clients don't have to be recompiled. Supposedly, setters let you add functionality like validation later on and your clients don't even need to know about it. However, adding validation to a setter is a change to its preconditions, a violation of the previous contract, which was, quite simply, "you can put anything in here, and you can get that same thing later from the getter".
So, now that you broke the contract, changing every file in the codebase is something you should want to do, not avoid. If you avoid it you're making the assumption that all the code assumed the contract for those methods was different.
If that should not have been the contract, then the interface was allowing clients to put the object in invalid states. That's the exact opposite of encapsulation If that field could not really be set to anything from the start, why wasn't the validation there from the start?
This same argument applies to other supposed advantages of these pass-through getter/setter pairs: if you later decide to change the value being set, you're breaking the contract. If you override the default functionality in a derived class, in a way beyond a few harmless modifications (like logging or other non-observable behaviour), you're breaking the contract of the base class. That is a violation of the Liskov Substitutability Principle, which is seen as one of the tenets of OO.
If a class has these dumb getters and setters for every field, then it is a class that has no invariants whatsoever, no contract. Is that really object-oriented design? If all the class has is those getters and setters, it's just a dumb data holder, and dumb data holders should look like dumb data holders:
class Foo {
public:
int DaysLeft;
int ContestantNumber;
};
Adding pass-through getter/setter pairs to such a class adds no value. Other classes should provide meaningful operations, not just operations that fields already provide. That's how you can define and maintain useful invariants.
Client: "What can I do with an object of this class?"
Designer: "You can read and write several variables."
Client: "Oh... cool, I guess?"
There are reasons to use getters and setters, but if those reasons don't exist, making getter/setter pairs in the name of false encapsulation gods is not a good thing. Valid reasons to make getters or setters include the things often mentioned as the potential changes you can make later, like validation or different internal representations. Or maybe the value should be readable by clients but not writable (for example, reading the size of a dictionary), so a simple getter is a nice choice. But those reasons should be there when you make the choice, and not just as a potential thing you may want later. This is an instance of YAGNI (You Ain't Gonna Need It).

Lots of people talk about the advantages of getters and setters but I want to play devil's advocate. Right now I'm debugging a very large program where the programmers decided to make everything getters and setters. That might seem nice, but its a reverse-engineering nightmare.
Say you're looking through hundreds of lines of code and you come across this:
person.name = "Joe";
It's a beautifully simply piece of code until you realize its a setter. Now, you follow that setter and find that it also sets person.firstName, person.lastName, person.isHuman, person.hasReallyCommonFirstName, and calls person.update(), which sends a query out to the database, etc. Oh, that's where your memory leak was occurring.
Understanding a local piece of code at first glance is an important property of good readability that getters and setters tend to break. That is why I try to avoid them when I can, and minimize what they do when I use them.

In a pure object-oriented world getters and setters is a terrible anti-pattern. Read this article: Getters/Setters. Evil. Period. In a nutshell, they encourage programmers to think about objects as of data structures, and this type of thinking is pure procedural (like in COBOL or C). In an object-oriented language there are no data structures, but only objects that expose behavior (not attributes/properties!)
You may find more about them in Section 3.5 of Elegant Objects (my book about object-oriented programming).

There are many reasons. My favorite one is when you need to change the behavior or regulate what you can set on a variable. For instance, lets say you had a setSpeed(int speed) method. But you want that you can only set a maximum speed of 100. You would do something like:
public void setSpeed(int speed) {
if ( speed > 100 ) {
this.speed = 100;
} else {
this.speed = speed;
}
}
Now what if EVERYWHERE in your code you were using the public field and then you realized you need the above requirement? Have fun hunting down every usage of the public field instead of just modifying your setter.
My 2 cents :)

One advantage of accessors and mutators is that you can perform validation.
For example, if foo was public, I could easily set it to null and then someone else could try to call a method on the object. But it's not there anymore! With a setFoo method, I could ensure that foo was never set to null.
Accessors and mutators also allow for encapsulation - if you aren't supposed to see the value once its set (perhaps it's set in the constructor and then used by methods, but never supposed to be changed), it will never been seen by anyone. But if you can allow other classes to see or change it, you can provide the proper accessor and/or mutator.

Thanks, that really clarified my thinking. Now here is (almost) 10 (almost) good reasons NOT to use getters and setters:
When you realize you need to do more than just set and get the value, you can just make the field private, which will instantly tell you where you've directly accessed it.
Any validation you perform in there can only be context free, which validation rarely is in practice.
You can change the value being set - this is an absolute nightmare when the caller passes you a value that they [shock horror] want you to store AS IS.
You can hide the internal representation - fantastic, so you're making sure that all these operations are symmetrical right?
You've insulated your public interface from changes under the sheets - if you were designing an interface and weren't sure whether direct access to something was OK, then you should have kept designing.
Some libraries expect this, but not many - reflection, serialization, mock objects all work just fine with public fields.
Inheriting this class, you can override default functionality - in other words you can REALLY confuse callers by not only hiding the implementation but making it inconsistent.
The last three I'm just leaving (N/A or D/C)...

Depends on your language. You've tagged this "object-oriented" rather than "Java", so I'd like to point out that ChssPly76's answer is language-dependent. In Python, for instance, there is no reason to use getters and setters. If you need to change the behavior, you can use a property, which wraps a getter and setter around basic attribute access. Something like this:
class Simple(object):
def _get_value(self):
return self._value -1
def _set_value(self, new_value):
self._value = new_value + 1
def _del_value(self):
self.old_values.append(self._value)
del self._value
value = property(_get_value, _set_value, _del_value)

Well i just want to add that even if sometimes they are necessary for the encapsulation and security of your variables/objects, if we want to code a real Object Oriented Program, then we need to STOP OVERUSING THE ACCESSORS, cause sometimes we depend a lot on them when is not really necessary and that makes almost the same as if we put the variables public.

EDIT: I answered this question because there are a bunch of people learning programming asking this, and most of the answers are very technically competent, but they're not as easy to understand if you're a newbie. We were all newbies, so I thought I'd try my hand at a more newbie friendly answer.
The two main ones are polymorphism, and validation. Even if it's just a stupid data structure.
Let's say we have this simple class:
public class Bottle {
public int amountOfWaterMl;
public int capacityMl;
}
A very simple class that holds how much liquid is in it, and what its capacity is (in milliliters).
What happens when I do:
Bottle bot = new Bottle();
bot.amountOfWaterMl = 1500;
bot.capacityMl = 1000;
Well, you wouldn't expect that to work, right?
You want there to be some kind of sanity check. And worse, what if I never specified the maximum capacity? Oh dear, we have a problem.
But there's another problem too. What if bottles were just one type of container? What if we had several containers, all with capacities and amounts of liquid filled? If we could just make an interface, we could let the rest of our program accept that interface, and bottles, jerrycans and all sorts of stuff would just work interchangably. Wouldn't that be better? Since interfaces demand methods, this is also a good thing.
We'd end up with something like:
public interface LiquidContainer {
public int getAmountMl();
public void setAmountMl(int amountMl);
public int getCapacityMl();
}
Great! And now we just change Bottle to this:
public class Bottle implements LiquidContainer {
private int capacityMl;
private int amountFilledMl;
public Bottle(int capacityMl, int amountFilledMl) {
this.capacityMl = capacityMl;
this.amountFilledMl = amountFilledMl;
checkNotOverFlow();
}
public int getAmountMl() {
return amountFilledMl;
}
public void setAmountMl(int amountMl) {
this.amountFilled = amountMl;
checkNotOverFlow();
}
public int getCapacityMl() {
return capacityMl;
}
private void checkNotOverFlow() {
if(amountOfWaterMl > capacityMl) {
throw new BottleOverflowException();
}
}
I'll leave the definition of the BottleOverflowException as an exercise to the reader.
Now notice how much more robust this is. We can deal with any type of container in our code now by accepting LiquidContainer instead of Bottle. And how these bottles deal with this sort of stuff can all differ. You can have bottles that write their state to disk when it changes, or bottles that save on SQL databases or GNU knows what else.
And all these can have different ways to handle various whoopsies. The Bottle just checks and if it's overflowing it throws a RuntimeException. But that might be the wrong thing to do.
(There is a useful discussion to be had about error handling, but I'm keeping it very simple here on purpose. People in comments will likely point out the flaws of this simplistic approach. ;) )
And yes, it seems like we go from a very simple idea to getting much better answers quickly.
Please note also that you can't change the capacity of a bottle. It's now set in stone. You could do this with an int by declaring it final. But if this was a list, you could empty it, add new things to it, and so on. You can't limit the access to touching the innards.
There's also the third thing that not everyone has addressed: getters and setters use method calls. That means that they look like normal methods everywhere else does. Instead of having weird specific syntax for DTOs and stuff, you have the same thing everywhere.

I know it's a bit late, but I think there are some people who are interested in performance.
I've done a little performance test. I wrote a class "NumberHolder" which, well, holds an Integer. You can either read that Integer by using the getter method
anInstance.getNumber() or by directly accessing the number by using anInstance.number. My programm reads the number 1,000,000,000 times, via both ways. That process is repeated five times and the time is printed. I've got the following result:
Time 1: 953ms, Time 2: 741ms
Time 1: 655ms, Time 2: 743ms
Time 1: 656ms, Time 2: 634ms
Time 1: 637ms, Time 2: 629ms
Time 1: 633ms, Time 2: 625ms
(Time 1 is the direct way, Time 2 is the getter)
You see, the getter is (almost) always a bit faster. Then I tried with different numbers of cycles. Instead of 1 million, I used 10 million and 0.1 million.
The results:
10 million cycles:
Time 1: 6382ms, Time 2: 6351ms
Time 1: 6363ms, Time 2: 6351ms
Time 1: 6350ms, Time 2: 6363ms
Time 1: 6353ms, Time 2: 6357ms
Time 1: 6348ms, Time 2: 6354ms
With 10 million cycles, the times are almost the same.
Here are 100 thousand (0.1 million) cycles:
Time 1: 77ms, Time 2: 73ms
Time 1: 94ms, Time 2: 65ms
Time 1: 67ms, Time 2: 63ms
Time 1: 65ms, Time 2: 65ms
Time 1: 66ms, Time 2: 63ms
Also with different amounts of cycles, the getter is a little bit faster than the regular way. I hope this helped you.

Don't use getters setters unless needed for your current delivery I.e. Don't think too much about what would happen in the future, if any thing to be changed its a change request in most of the production applications, systems.
Think simple, easy, add complexity when needed.
I would not take advantage of ignorance of business owners of deep technical know how just because I think it's correct or I like the approach.
I have massive system written without getters setters only with access modifiers and some methods to validate n perform biz logic. If you absolutely needed the. Use anything.

We use getters and setters:
for reusability
to perform validation in later stages of programming
Getter and setter methods are public interfaces to access private class members.
Encapsulation mantra
The encapsulation mantra is to make fields private and methods public.
Getter Methods: We can get access to private variables.
Setter Methods: We can modify private fields.
Even though the getter and setter methods do not add new functionality, we can change our mind come back later to make that method
better;
safer; and
faster.
Anywhere a value can be used, a method that returns that value can be added. Instead of:
int x = 1000 - 500
use
int x = 1000 - class_name.getValue();
In layman's terms
Suppose we need to store the details of this Person. This Person has the fields name, age and sex. Doing this involves creating methods for name, age and sex. Now if we need create another person, it becomes necessary to create the methods for name, age, sex all over again.
Instead of doing this, we can create a bean class(Person) with getter and setter methods. So tomorrow we can just create objects of this Bean class(Person class) whenever we need to add a new person (see the figure). Thus we are reusing the fields and methods of bean class, which is much better.

I spent quite a while thinking this over for the Java case, and I believe the real reasons are:
Code to the interface, not the implementation
Interfaces only specify methods, not fields
In other words, the only way you can specify a field in an interface is by providing a method for writing a new value and a method for reading the current value.
Those methods are the infamous getter and setter....

It can be useful for lazy-loading. Say the object in question is stored in a database, and you don't want to go get it unless you need it. If the object is retrieved by a getter, then the internal object can be null until somebody asks for it, then you can go get it on the first call to the getter.
I had a base page class in a project that was handed to me that was loading some data from a couple different web service calls, but the data in those web service calls wasn't always used in all child pages. Web services, for all of the benefits, pioneer new definitions of "slow", so you don't want to make a web service call if you don't have to.
I moved from public fields to getters, and now the getters check the cache, and if it's not there call the web service. So with a little wrapping, a lot of web service calls were prevented.
So the getter saves me from trying to figure out, on each child page, what I will need. If I need it, I call the getter, and it goes to find it for me if I don't already have it.
protected YourType _yourName = null;
public YourType YourName{
get
{
if (_yourName == null)
{
_yourName = new YourType();
return _yourName;
}
}
}

One aspect I missed in the answers so far, the access specification:
for members you have only one access specification for both setting and getting
for setters and getters you can fine tune it and define it separately

In languages which don't support "properties" (C++, Java) or require recompilation of clients when changing fields to properties (C#), using get/set methods is easier to modify. For example, adding validation logic to a setFoo method will not require changing the public interface of a class.
In languages which support "real" properties (Python, Ruby, maybe Smalltalk?) there is no point to get/set methods.

One of the basic principals of OO design: Encapsulation!
It gives you many benefits, one of which being that you can change the implementation of the getter/setter behind the scenes but any consumer of that value will continue to work as long as the data type remains the same.

You should use getters and setters when:
You're dealing with something that is conceptually an attribute, but:
Your language doesn't have properties (or some similar mechanism, like Tcl's variable traces), or
Your language's property support isn't sufficient for this use case, or
Your language's (or sometimes your framework's) idiomatic conventions encourage getters or setters for this use case.
So this is very rarely a general OO question; it's a language-specific question, with different answers for different languages (and different use cases).
From an OO theory point of view, getters and setters are useless. The interface of your class is what it does, not what its state is. (If not, you've written the wrong class.) In very simple cases, where what a class does is just, e.g., represent a point in rectangular coordinates,* the attributes are part of the interface; getters and setters just cloud that. But in anything but very simple cases, neither the attributes nor getters and setters are part of the interface.
Put another way: If you believe that consumers of your class shouldn't even know that you have a spam attribute, much less be able to change it willy-nilly, then giving them a set_spam method is the last thing you want to do.
* Even for that simple class, you may not necessarily want to allow setting the x and y values. If this is really a class, shouldn't it have methods like translate, rotate, etc.? If it's only a class because your language doesn't have records/structs/named tuples, then this isn't really a question of OO…
But nobody is ever doing general OO design. They're doing design, and implementation, in a specific language. And in some languages, getters and setters are far from useless.
If your language doesn't have properties, then the only way to represent something that's conceptually an attribute, but is actually computed, or validated, etc., is through getters and setters.
Even if your language does have properties, there may be cases where they're insufficient or inappropriate. For example, if you want to allow subclasses to control the semantics of an attribute, in languages without dynamic access, a subclass can't substitute a computed property for an attribute.
As for the "what if I want to change my implementation later?" question (which is repeated multiple times in different wording in both the OP's question and the accepted answer): If it really is a pure implementation change, and you started with an attribute, you can change it to a property without affecting the interface. Unless, of course, your language doesn't support that. So this is really just the same case again.
Also, it's important to follow the idioms of the language (or framework) you're using. If you write beautiful Ruby-style code in C#, any experienced C# developer other than you is going to have trouble reading it, and that's bad. Some languages have stronger cultures around their conventions than others.—and it may not be a coincidence that Java and Python, which are on opposite ends of the spectrum for how idiomatic getters are, happen to have two of the strongest cultures.
Beyond human readers, there will be libraries and tools that expect you to follow the conventions, and make your life harder if you don't. Hooking Interface Builder widgets to anything but ObjC properties, or using certain Java mocking libraries without getters, is just making your life more difficult. If the tools are important to you, don't fight them.

From a object orientation design standpoint both alternatives can be damaging to the maintenance of the code by weakening the encapsulation of the classes. For a discussion you can look into this excellent article: http://typicalprogrammer.com/?p=23

Code evolves. private is great for when you need data member protection. Eventually all classes should be sort of "miniprograms" that have a well-defined interface that you can't just screw with the internals of.
That said, software development isn't about setting down that final version of the class as if you're pressing some cast iron statue on the first try. While you're working with it, code is more like clay. It evolves as you develop it and learn more about the problem domain you are solving. During development classes may interact with each other than they should (dependency you plan to factor out), merge together, or split apart. So I think the debate boils down to people not wanting to religiously write
int getVar() const { return var ; }
So you have:
doSomething( obj->getVar() ) ;
Instead of
doSomething( obj->var ) ;
Not only is getVar() visually noisy, it gives this illusion that gettingVar() is somehow a more complex process than it really is. How you (as the class writer) regard the sanctity of var is particularly confusing to a user of your class if it has a passthru setter -- then it looks like you're putting up these gates to "protect" something you insist is valuable, (the sanctity of var) but yet even you concede var's protection isn't worth much by the ability for anyone to just come in and set var to whatever value they want, without you even peeking at what they are doing.
So I program as follows (assuming an "agile" type approach -- ie when I write code not knowing exactly what it will be doing/don't have time or experience to plan an elaborate waterfall style interface set):
1) Start with all public members for basic objects with data and behavior. This is why in all my C++ "example" code you'll notice me using struct instead of class everywhere.
2) When an object's internal behavior for a data member becomes complex enough, (for example, it likes to keep an internal std::list in some kind of order), accessor type functions are written. Because I'm programming by myself, I don't always set the member private right away, but somewhere down the evolution of the class the member will be "promoted" to either protected or private.
3) Classes that are fully fleshed out and have strict rules about their internals (ie they know exactly what they are doing, and you are not to "fuck" (technical term) with its internals) are given the class designation, default private members, and only a select few members are allowed to be public.
I find this approach allows me to avoid sitting there and religiously writing getter/setters when a lot of data members get migrated out, shifted around, etc. during the early stages of a class's evolution.

There is a good reason to consider using accessors is there is no property inheritance. See next example:
public class TestPropertyOverride {
public static class A {
public int i = 0;
public void add() {
i++;
}
public int getI() {
return i;
}
}
public static class B extends A {
public int i = 2;
#Override
public void add() {
i = i + 2;
}
#Override
public int getI() {
return i;
}
}
public static void main(String[] args) {
A a = new B();
System.out.println(a.i);
a.add();
System.out.println(a.i);
System.out.println(a.getI());
}
}
Output:
0
0
4

Getters and setters are used to implement two of the fundamental aspects of Object Oriented Programming which are:
Abstraction
Encapsulation
Suppose we have an Employee class:
package com.highmark.productConfig.types;
public class Employee {
private String firstName;
private String middleName;
private String lastName;
public String getFirstName() {
return firstName;
}
public void setFirstName(String firstName) {
this.firstName = firstName;
}
public String getMiddleName() {
return middleName;
}
public void setMiddleName(String middleName) {
this.middleName = middleName;
}
public String getLastName() {
return lastName;
}
public void setLastName(String lastName) {
this.lastName = lastName;
}
public String getFullName(){
return this.getFirstName() + this.getMiddleName() + this.getLastName();
}
}
Here the implementation details of Full Name is hidden from the user and is not accessible directly to the user, unlike a public attribute.

There is a difference between DataStructure and Object.
Datastructure should expose its innards and not behavior.
An Object should not expose its innards but it should expose its behavior, which is also known as the Law of Demeter
Mostly DTOs are considered more of a datastructure and not Object. They should only expose their data and not behavior. Having Setter/Getter in DataStructure will expose behavior instead of data inside it. This further increases the chance of violation of Law of Demeter.
Uncle Bob in his book Clean code explained the Law of Demeter.
There is a well-known heuristic called the Law of Demeter that says a
module should not know about the innards of the objects it
manipulates. As we saw in the last section, objects hide their data
and expose operations. This means that an object should not expose its
internal structure through accessors because to do so is to expose,
rather than to hide, its internal structure.
More precisely, the Law of Demeter says that a method f of a class C
should only call the methods of these:
C
An object created by f
An object passed as an argument to f
An object held in an instance variable of C
The method should not invoke methods on objects that are returned by any of the allowed functions.
In other words, talk to friends, not to strangers.
So according this, example of LoD violation is:
final String outputDir = ctxt.getOptions().getScratchDir().getAbsolutePath();
Here, the function should call the method of its immediate friend which is ctxt here, It should not call the method of its immediate friend's friend. but this rule doesn't apply to data structure. so here if ctxt, option, scratchDir are datastructure then why to wrap their internal data with some behavior and doing a violation of LoD.
Instead, we can do something like this.
final String outputDir = ctxt.options.scratchDir.absolutePath;
This fulfills our needs and doesn't even violate LoD.
Inspired by Clean Code by Robert C. Martin(Uncle Bob)

If you don't require any validations and not even need to maintain state i.e. one property depends on another so we need to maintain the state when one is change. You can keep it simple by making field public and not using getter and setters.
I think OOPs complicates things as the program grows it becomes nightmare for developer to scale.
A simple example; we generate c++ headers from xml. The header contains simple field which does not require any validations. But still as in OOPS accessor are fashion we generates them as following.
const Filed& getfield() const
Field& getField()
void setfield(const Field& field){...}
which is very verbose and is not required. a simple
struct
{
Field field;
};
is enough and readable.
Functional programming don't have the concept of data hiding they even don't require it as they do not mutate the data.

Additionally, this is to "future-proof" your class. In particular, changing from a field to a property is an ABI break, so if you do later decide that you need more logic than just "set/get the field", then you need to break ABI, which of course creates problems for anything else already compiled against your class.

One other use (in languages that support properties) is that setters and getters can imply that an operation is non-trivial. Typically, you want to avoid doing anything that's computationally expensive in a property.

One relatively modern advantage of getters/setters is that is makes it easier to browse code in tagged (indexed) code editors. E.g. If you want to see who sets a member, you can open the call hierarchy of the setter.
On the other hand, if the member is public, the tools don't make it possible to filter read/write access to the member. So you have to trudge though all uses of the member.

Getters and setters coming from data hiding. Data Hiding means We
are hiding data from outsiders or outside person/thing cannot access
our data.This is a useful feature in OOP.
As a example:
If you create a public variable, you can access that variable and change value in anywhere(any class). But if you create as private that variable cannot see/access in any class except declared class.
public and private are access modifiers.
So how can we access that variable outside:
This is the place getters and setters coming from. You can declare variable as private then you can implement getter and setter for that variable.
Example(Java):
private String name;
public String getName(){
return this.name;
}
public void setName(String name){
this.name= name;
}
Advantage:
When anyone want to access or change/set value to balance variable, he/she must have permision.
//assume we have person1 object
//to give permission to check balance
person1.getName()
//to give permission to set balance
person1.setName()
You can set value in constructor also but when later on when you want
to update/change value, you have to implement setter method.