Related
A web service returns a huge XML and I need to access deeply nested fields of it. For example:
return wsObject.getFoo().getBar().getBaz().getInt()
The problem is that getFoo(), getBar(), getBaz() may all return null.
However, if I check for null in all cases, the code becomes very verbose and hard to read. Moreover, I may miss the checks for some of the fields.
if (wsObject.getFoo() == null) return -1;
if (wsObject.getFoo().getBar() == null) return -1;
// maybe also do something with wsObject.getFoo().getBar()
if (wsObject.getFoo().getBar().getBaz() == null) return -1;
return wsObject.getFoo().getBar().getBaz().getInt();
Is it acceptable to write
try {
return wsObject.getFoo().getBar().getBaz().getInt();
} catch (NullPointerException ignored) {
return -1;
}
or would that be considered an antipattern?
Catching NullPointerException is a really problematic thing to do since they can happen almost anywhere. It's very easy to get one from a bug, catch it by accident and continue as if everything is normal, thus hiding a real problem. It's so tricky to deal with so it's best to avoid altogether. (For example, think about auto-unboxing of a null Integer.)
I suggest that you use the Optional class instead. This is often the best approach when you want to work with values that are either present or absent.
Using that you could write your code like this:
public Optional<Integer> m(Ws wsObject) {
return Optional.ofNullable(wsObject.getFoo()) // Here you get Optional.empty() if the Foo is null
.map(f -> f.getBar()) // Here you transform the optional or get empty if the Bar is null
.map(b -> b.getBaz())
.map(b -> b.getInt());
// Add this if you want to return null instead of an empty optional if any is null
// .orElse(null);
// Or this if you want to throw an exception instead
// .orElseThrow(SomeApplicationException::new);
}
Why optional?
Using Optionals instead of null for values that might be absent makes that fact very visible and clear to readers, and the type system will make sure you don't accidentally forget about it.
You also get access to methods for working with such values more conveniently, like map and orElse.
Is absence valid or error?
But also think about if it is a valid result for the intermediate methods to return null or if that is a sign of an error. If it is always an error then it's probably better throw an exception than to return a special value, or for the intermediate methods themselves to throw an exception.
Maybe more optionals?
If on the other hand absent values from the intermediate methods are valid, maybe you can switch to Optionals for them also?
Then you could use them like this:
public Optional<Integer> mo(Ws wsObject) {
return wsObject.getFoo()
.flatMap(f -> f.getBar())
.flatMap(b -> b.getBaz())
.flatMap(b -> b.getInt());
}
Why not optional?
The only reason I can think of for not using Optional is if this is in a really performance critical part of the code, and if garbage collection overhead turns out to be a problem. This is because a few Optional objects are allocated each time the code is executed, and the VM might not be able to optimize those away. In that case your original if-tests might be better.
I suggest considering Objects.requireNonNull(T obj, String message). You might build chains with a detailed message for each exception, like
requireNonNull(requireNonNull(requireNonNull(
wsObject, "wsObject is null")
.getFoo(), "getFoo() is null")
.getBar(), "getBar() is null");
I would suggest you not to use special return-values, like -1. That's not a Java style. Java has designed the mechanism of exceptions to avoid this old-fashioned way which came from the C language.
Throwing NullPointerException is not the best option too. You could provide your own exception (making it checked to guarantee that it will be handled by a user or unchecked to process it in an easier way) or use a specific exception from XML parser you are using.
Assuming the class structure is indeed out of our control, as seems to be the case, I think catching the NPE as suggested in the question is indeed a reasonable solution, unless performance is a major concern. One small improvement might be to wrap the throw/catch logic to avoid clutter:
static <T> T get(Supplier<T> supplier, T defaultValue) {
try {
return supplier.get();
} catch (NullPointerException e) {
return defaultValue;
}
}
Now you can simply do:
return get(() -> wsObject.getFoo().getBar().getBaz().getInt(), -1);
As already pointed out by Tom in the comment,
Following statement disobeys the Law of Demeter,
wsObject.getFoo().getBar().getBaz().getInt()
What you want is int and you can get it from Foo. Law of Demeter says, never talk to the strangers. For your case you can hide the actual implementation under the hood of Foo and Bar.
Now, you can create method in Foo to fetch int from Baz. Ultimately, Foo will have Bar and in Bar we can access Int without exposing Baz directly to Foo. So, null checks are probably divided to different classes and only required attributes will be shared among the classes.
My answer goes almost in the same line as #janki, but I would like to modify the code snippet slightly as below:
if (wsObject.getFoo() != null && wsObject.getFoo().getBar() != null && wsObject.getFoo().getBar().getBaz() != null)
return wsObject.getFoo().getBar().getBaz().getInt();
else
return something or throw exception;
You can add a null check for wsObject as well, if there's any chance of that object being null.
You say that some methods "may return null" but do not say in what circumstances they return null. You say you catch the NullPointerException but you do not say why you catch it. This lack of information suggests you do not have a clear understanding of what exceptions are for and why they are superior to the alternative.
Consider a class method that is meant to perform an action, but the method can not guarantee it will perform the action, because of circumstances beyond its control (which is in fact the case for all methods in Java). We call that method and it returns. The code that calls that method needs to know whether it was successful. How can it know? How can it be structured to cope with the two possibilities, of success or failure?
Using exceptions, we can write methods that have success as a post condition. If the method returns, it was successful. If it throws an exception, it had failed. This is a big win for clarity. We can write code that clearly processes the normal, success case, and move all the error handling code into catch clauses. It often transpires that the details of how or why a method was unsuccessful are not important to the caller, so the same catch clause can be used for handling several types of failure. And it often happens that a method does not need to catch exceptions at all, but can just allow them to propagate to its caller. Exceptions due to program bugs are in that latter class; few methods can react appropriately when there is a bug.
So, those methods that return null.
Does a null value indicate a bug in your code? If it does, you should not be catching the exception at all. And your code should not be trying to second guess itself. Just write what is clear and concise on the assumption that it will work. Is a chain of method calls clear and concise? Then just use them.
Does a null value indicate invalid input to your program? If it does, a NullPointerException is not an appropriate exception to throw, because conventionally it is reserved for indicating bugs. You probably want to throw a custom exception derived from IllegalArgumentException (if you want an unchecked exception) or IOException (if you want a checked exception). Is your program required to provide detailed syntax error messages when there is invalid input? If so, checking each method for a null return value then throwing an appropriate diagnostic exception is the only thing you can do. If your program need not provide detailed diagnostics, chaining the method calls together, catching any NullPointerException and then throwing your custom exception is clearest and most concise.
One of the answers claims that the chained method calls violate the Law of Demeter and thus are bad. That claim is mistaken.
When it comes to program design, there are not really any absolute rules about what is good and what is bad. There are only heuristics: rules that are right much (even almost all) of the time. Part of the skill of programming is knowing when it is OK to break those kinds of rules. So a terse assertion that "this is against rule X" is not really an answer at all. Is this one of the situations where the rule should be broken?
The Law of Demeter is really a rule about API or class interface design. When designing classes, it is useful to have a hierarchy of abstractions. You have low level classes that uses the language primitives to directly perform operations and represent objects in an abstraction that is higher level than the language primitives. You have medium level classes that delegate to the low level classes, and implement operations and representations at a higher level than the low level classes. You have high level classes that delegate to the medium level classes, and implement still higher level operations and abstractions. (I've talked about just three levels of abstraction here, but more are possible). This allows your code to express itself in terms of appropriate abstractions at each level, thereby hiding complexity. The rationale for the Law of Demeter is that if you have a chain of method calls, that suggests you have a high level class reaching in through a medium level class to deal directly with low level details, and therefore that your medium level class has not provided a medium-level abstract operation that the high level class needs. But it seems that is not the situation you have here: you did not design the classes in the chain of method calls, they are the result of some auto-generated XML serialization code (right?), and the chain of calls is not descending through an abstraction hierarchy because the des-serialized XML is all at the same level of the abstraction hierarchy (right?)?
As others have said, respecting the Law of Demeter is definitely part of the solution. Another part, wherever possible, is to change those chained methods so they cannot return null. You can avoid returning null by instead returning an empty String, an empty Collection, or some other dummy object that means or does whatever the caller would do with null.
To improve readability, you may want to use multiple variables, like
Foo theFoo;
Bar theBar;
Baz theBaz;
theFoo = wsObject.getFoo();
if ( theFoo == null ) {
// Exit.
}
theBar = theFoo.getBar();
if ( theBar == null ) {
// Exit.
}
theBaz = theBar.getBaz();
if ( theBaz == null ) {
// Exit.
}
return theBaz.getInt();
Don't catch NullPointerException. You don't know where it is coming from (I know it is not probable in your case but maybe something else threw it) and it is slow.
You want to access the specified field and for this every other field has to be not null. This is a perfect valid reason to check every field. I would probably check it in one if and then create a method for readability. As others pointed out already returning -1 is very oldschool but I don't know if you have a reason for it or not (e.g. talking to another system).
public int callService() {
...
if(isValid(wsObject)){
return wsObject.getFoo().getBar().getBaz().getInt();
}
return -1;
}
public boolean isValid(WsObject wsObject) {
if(wsObject.getFoo() != null &&
wsObject.getFoo().getBar() != null &&
wsObject.getFoo().getBar().getBaz() != null) {
return true;
}
return false;
}
Edit: It is debatable if it's disobeyes the Law Of Demeter since the WsObject is probably only a data structure (check https://stackoverflow.com/a/26021695/1528880).
If you don't want to refactor the code and you can use Java 8, it is possible to use Method references.
A simple demo first (excuse the static inner classes)
public class JavaApplication14
{
static class Baz
{
private final int _int;
public Baz(int value){ _int = value; }
public int getInt(){ return _int; }
}
static class Bar
{
private final Baz _baz;
public Bar(Baz baz){ _baz = baz; }
public Baz getBar(){ return _baz; }
}
static class Foo
{
private final Bar _bar;
public Foo(Bar bar){ _bar = bar; }
public Bar getBar(){ return _bar; }
}
static class WSObject
{
private final Foo _foo;
public WSObject(Foo foo){ _foo = foo; }
public Foo getFoo(){ return _foo; }
}
interface Getter<T, R>
{
R get(T value);
}
static class GetterResult<R>
{
public R result;
public int lastIndex;
}
/**
* #param args the command line arguments
*/
public static void main(String[] args)
{
WSObject wsObject = new WSObject(new Foo(new Bar(new Baz(241))));
WSObject wsObjectNull = new WSObject(new Foo(null));
GetterResult<Integer> intResult
= getterChain(wsObject, WSObject::getFoo, Foo::getBar, Bar::getBar, Baz::getInt);
GetterResult<Integer> intResult2
= getterChain(wsObjectNull, WSObject::getFoo, Foo::getBar, Bar::getBar, Baz::getInt);
System.out.println(intResult.result);
System.out.println(intResult.lastIndex);
System.out.println();
System.out.println(intResult2.result);
System.out.println(intResult2.lastIndex);
// TODO code application logic here
}
public static <R, V1, V2, V3, V4> GetterResult<R>
getterChain(V1 value, Getter<V1, V2> g1, Getter<V2, V3> g2, Getter<V3, V4> g3, Getter<V4, R> g4)
{
GetterResult result = new GetterResult<>();
Object tmp = value;
if (tmp == null)
return result;
tmp = g1.get((V1)tmp);
result.lastIndex++;
if (tmp == null)
return result;
tmp = g2.get((V2)tmp);
result.lastIndex++;
if (tmp == null)
return result;
tmp = g3.get((V3)tmp);
result.lastIndex++;
if (tmp == null)
return result;
tmp = g4.get((V4)tmp);
result.lastIndex++;
result.result = (R)tmp;
return result;
}
}
Output
241
4
null
2
The interface Getter is just a functional interface, you may use any equivalent.
GetterResult class, accessors stripped out for clarity, hold the result of the getter chain, if any, or the index of the last getter called.
The method getterChain is a simple, boilerplate piece of code, that can be generated automatically (or manually when needed).
I structured the code so that the repeating block is self evident.
This is not a perfect solution as you still need to define one overload of getterChain per number of getters.
I would refactor the code instead, but if can't and you find your self using long getter chains often you may consider building a class with the overloads that take from 2 to, say, 10, getters.
I'd like to add an answer which focus on the meaning of the error. Null exception in itself doesn't provide any meaning full error. So I'd advise to avoid dealing with them directly.
There is a thousands cases where your code can go wrong: cannot connect to database, IO Exception, Network error... If you deal with them one by one (like the null check here), it would be too much of a hassle.
In the code:
wsObject.getFoo().getBar().getBaz().getInt();
Even when you know which field is null, you have no idea about what goes wrong. Maybe Bar is null, but is it expected? Or is it a data error? Think about people who read your code
Like in xenteros's answer, I'd propose using custom unchecked exception. For example, in this situation: Foo can be null (valid data), but Bar and Baz should never be null (invalid data)
The code can be re-written:
void myFunction()
{
try
{
if (wsObject.getFoo() == null)
{
throw new FooNotExistException();
}
return wsObject.getFoo().getBar().getBaz().getInt();
}
catch (Exception ex)
{
log.error(ex.Message, ex); // Write log to track whatever exception happening
throw new OperationFailedException("The requested operation failed")
}
}
void Main()
{
try
{
myFunction();
}
catch(FooNotExistException)
{
// Show error: "Your foo does not exist, please check"
}
catch(OperationFailedException)
{
// Show error: "Operation failed, please contact our support"
}
}
NullPointerException is a run-time exception, so generally speaking is not recommended to catch it, but to avoid it.
You will have to catch the exception wherever you want to call the method (or it will propagate up the stack). Nevertheless, if in your case you can keep working with that result with value -1 and you are sure that it won't propagate because you are not using any of the "pieces" that may be null, then it seems right to me to catch it
Edit:
I agree with the later answer from #xenteros, it wil be better to launch your own exception instead returning -1 you can call it InvalidXMLException for instance.
Have been following this post since yesterday.
I have been commenting/voting the comments which says, catching NPE is bad. Here is why I have been doing that.
package com.todelete;
public class Test {
public static void main(String[] args) {
Address address = new Address();
address.setSomeCrap(null);
Person person = new Person();
person.setAddress(address);
long startTime = System.currentTimeMillis();
for (int i = 0; i < 1000000; i++) {
try {
System.out.println(person.getAddress().getSomeCrap().getCrap());
} catch (NullPointerException npe) {
}
}
long endTime = System.currentTimeMillis();
System.out.println((endTime - startTime) / 1000F);
long startTime1 = System.currentTimeMillis();
for (int i = 0; i < 1000000; i++) {
if (person != null) {
Address address1 = person.getAddress();
if (address1 != null) {
SomeCrap someCrap2 = address1.getSomeCrap();
if (someCrap2 != null) {
System.out.println(someCrap2.getCrap());
}
}
}
}
long endTime1 = System.currentTimeMillis();
System.out.println((endTime1 - startTime1) / 1000F);
}
}
public class Person {
private Address address;
public Address getAddress() {
return address;
}
public void setAddress(Address address) {
this.address = address;
}
}
package com.todelete;
public class Address {
private SomeCrap someCrap;
public SomeCrap getSomeCrap() {
return someCrap;
}
public void setSomeCrap(SomeCrap someCrap) {
this.someCrap = someCrap;
}
}
package com.todelete;
public class SomeCrap {
private String crap;
public String getCrap() {
return crap;
}
public void setCrap(String crap) {
this.crap = crap;
}
}
Output
3.216
0.002
I see a clear winner here. Having if checks is way too less expensive than catch an exception. I have seen that Java-8 way of doing. Considering that 70% of the current applications still run on Java-7 I am adding this answer.
Bottom Line For any mission critical applications, handling NPE is costly.
If efficiency is an issue then the 'catch' option should be considered.
If 'catch' cannot be used because it would propagate (as mentioned by 'SCouto') then use local variables to avoid multiple calls to methods getFoo(), getBar() and getBaz().
It's worth considering to create your own Exception. Let's call it MyOperationFailedException. You can throw it instead returning a value. The result will be the same - you'll quit the function, but you won't return hard-coded value -1 which is Java anti-pattern. In Java we use Exceptions.
try {
return wsObject.getFoo().getBar().getBaz().getInt();
} catch (NullPointerException ignored) {
throw new MyOperationFailedException();
}
EDIT:
According to the discussion in comments let me add something to my previous thoughts. In this code there are two possibilities. One is that you accept null and the other one is, that it is an error.
If it's an error and it occurs, You can debug your code using other structures for debugging purposes when breakpoints aren't enough.
If it's acceptable, you don't care about where this null appeared. If you do, you definitely shouldn't chain those requests.
The method you have is lengthy, but very readable. If I were a new developer coming to your code base I could see what you were doing fairly quickly. Most of the other answers (including catching the exception) don't seem to be making things more readable and some are making it less readable in my opinion.
Given that you likely don't have control over the generated source and assuming you truly just need to access a few deeply nested fields here and there then I would recommend wrapping each deeply nested access with a method.
private int getFooBarBazInt() {
if (wsObject.getFoo() == null) return -1;
if (wsObject.getFoo().getBar() == null) return -1;
if (wsObject.getFoo().getBar().getBaz() == null) return -1;
return wsObject.getFoo().getBar().getBaz().getInt();
}
If you find yourself writing a lot of these methods or if you find yourself tempted to make these public static methods then I would create a separate object model, nested how you would like, with only the fields you care about, and convert from the web services object model to your object model.
When you are communicating with a remote web service it is very typical to have a "remote domain" and an "application domain" and switch between the two. The remote domain is often limited by the web protocol (for example, you can't send helper methods back and forth in a pure RESTful service and deeply nested object models are common to avoid multiple API calls) and so not ideal for direct use in your client.
For example:
public static class MyFoo {
private int barBazInt;
public MyFoo(Foo foo) {
this.barBazInt = parseBarBazInt();
}
public int getBarBazInt() {
return barBazInt;
}
private int parseFooBarBazInt(Foo foo) {
if (foo() == null) return -1;
if (foo().getBar() == null) return -1;
if (foo().getBar().getBaz() == null) return -1;
return foo().getBar().getBaz().getInt();
}
}
return wsObject.getFooBarBazInt();
by applying the the Law of Demeter,
class WsObject
{
FooObject foo;
..
Integer getFooBarBazInt()
{
if(foo != null) return foo.getBarBazInt();
else return null;
}
}
class FooObject
{
BarObject bar;
..
Integer getBarBazInt()
{
if(bar != null) return bar.getBazInt();
else return null;
}
}
class BarObject
{
BazObject baz;
..
Integer getBazInt()
{
if(baz != null) return baz.getInt();
else return null;
}
}
class BazObject
{
Integer myInt;
..
Integer getInt()
{
return myInt;
}
}
Giving answer which seems different from all others.
I recommend you to check for NULL in ifs.
Reason :
We should not leave a single chance for our program to be crashed.
NullPointer is generated by system. The behaviour of System
generated exceptions can not be predicted. You should not leave your
program in the hands of System when you already have a way of handling
it by your own. And put the Exception handling mechanism for the extra safety.!!
For making your code easy to read try this for checking the conditions :
if (wsObject.getFoo() == null || wsObject.getFoo().getBar() == null || wsObject.getFoo().getBar().getBaz() == null)
return -1;
else
return wsObject.getFoo().getBar().getBaz().getInt();
EDIT :
Here you need to store these values wsObject.getFoo(),
wsObject.getFoo().getBar(), wsObject.getFoo().getBar().getBaz() in
some variables. I am not doing it because i don't know the return
types of that functions.
Any suggestions will be appreciated..!!
I wrote a class called Snag which lets you define a path to navigate through a tree of objects. Here is an example of its use:
Snag<Car, String> ENGINE_NAME = Snag.createForAndReturn(Car.class, String.class).toGet("engine.name").andReturnNullIfMissing();
Meaning that the instance ENGINE_NAME would effectively call Car?.getEngine()?.getName() on the instance passed to it, and return null if any reference returned null:
final String name = ENGINE_NAME.get(firstCar);
It's not published on Maven but if anyone finds this useful it's here (with no warranty of course!)
It's a bit basic but it seems to do the job. Obviously it's more obsolete with more recent versions of Java and other JVM languages that support safe navigation or Optional.
I have to analyze a huge data stream which often includes incomplete data. Currently the code is littered with null checks at multiple levels, as there could be incomplete data at any level.
So for example I might have to retrieve:
Model.getDestination().getDevice().getName()
I tried to create a method to try and reduce the null checks to a single method whereby I enter:
IsValid(Model.getDestination(), Model.getDestination().getDevice(), Model.getDestination().getDevice().getName())
this method fails because it evaluates all parameters before it sends them, rather than checking each at a time like
Model.getDestination() != null && Model.getDestination().getDevice() != null && etc
but is there a way I could pass in Model.getDestination().getDevice().getName() and do the check at each level without having to evaluate it or split it up before I pass it?
What I really want it to do is if there is a null/nullexception it should quietly return "", and continue processing incoming data
I know there are ways to do this elegantly in Java 8, but I am stuck with Java 7
I struggled with a similar problem with deeply nested structures, and if I'd have had the opportunity to introduce additional structures just to navigate the underlying data, I think, I had done that.
This was C# which in the meantime has a save navigation/Elvis operator, for which we'll wait in vain with Java (proposed for Java 7 but discarded. Groovy has it btw.). Also looks like there are arguments against using Elvis, even if you have it). Also lambdas (and extension methods) didn't improve things really. Also every other approach has been discredited as ugly in other posts here.
Therefore I propose a secondary structure purely for navigation, each element with a getValue() method to access the original structure (also the shortcuts proposed by #Michael are straight forward to add this way). Allowing you null save navigation like this:
Model model = new Model(new Destination(null));
Destination destination = model.getDestination().getValue(); // destination is not null
Device device = model.getDestination().getDevice().getValue(); // device will be null, no NPE
String name = destination.getDevice().getName().getValue(); // name will be null, no NPE
NavDevice navDevice = model.getDestination().getDevice(); // returns an ever non-null NavDevice, not a Device
String name = navDevice.getValue().getName(); // cause an NPE by circumventing the navigation structure
With straight forward original structures
class Destination {
private final Device device;
public Destination(Device device) {
this.device = device;
}
public Device getDevice() {
return device;
}
}
class Device {
private final String name;
private Device(String name) {
this.name = name;
}
public String getName() {
return name;
}
}
And secondary structures for the purpose of save navigation.
Obviously this is debatable, since you always can access the original structure directly and run into a NPE. But in terms of readability perhaps I'd still take this, especially for large structures where a shrub of ifs or optionals really is an eyesore (which matters, if you have to tell, which business rules actually were implemented here).
A memory/speed argument could be countered by using only one navigation object per type and re-set their internals to approriate underlying objects as you navigate.
class Model {
private final Destination destination;
private Model(Destination destination) {
this.destination = destination;
}
public NavDestination getDestination() {
return new NavDestination(destination);
}
}
class NavDestination {
private final Destination value;
private NavDestination(Destination value) {
this.value = value;
}
public Destination getValue() {
return value;
}
public NavDevice getDevice() {
return new NavDevice(value == null ? null : value.getDevice());
}
}
class NavDevice {
private final Device value;
private NavDevice(Device value) {
this.value = value;
}
public Device getValue() {
return value;
}
public NavName getName() {
return new NavName(value == null ? null : value.getName());
}
}
class NavName {
private final String value;
private NavName(String value) {
this.value = value;
}
public String getValue() {
return value;
}
}
Option 1 - if statement
You already provided it in your question. I think using am if statementlike the following is perfectly acceptable:
Model.getDestination() != null && Model.getDestination().getDevice() != null && etc
Option 2 - javax Validation and checking the result - before sending
You could make use of javax validation.
See: https://www.baeldung.com/javax-validation
You would annotate the fields that you want with #NotNull.
Then you could use programmatic validation.
You could check the validation result to see if there is a problem.
Example:
So in your class you would do:
#NotNull
Public String Destination;
And you could feed your object to the validater:
ValidatorFactory factory = Validation.buildDefaultValidatorFactory();
Validator validator = factory.getValidator();
Set<ConstraintViolation<Model>> violations = validator.validate(Model);
for (ConstraintViolation<User> violation : violations) {
log.error(violation.getMessage());
}
Option 3 - fromNullable and Maps ( if you have Java 8)
I'm taking this one from https://softwareengineering.stackexchange.com/questions/255503/null-checking-whilst-navigating-object-hierarchies . This is very simular to your question.
import java.util.Optional;
Optional.fromNullable(model)
.map(Model::getDestination)
.map(Lounge::getDevice)
.ifPresent(letter -> .... do what you want ...);
Option 4 - Just using a try/catch
Everyone hates this one due to the slowness of exception.
So you want to simplify Model.getDestination().getDevice().getName(). First, I want to list a few things that should not be done: Don't use exceptions. Don't write an IsValid method, because it just doesn't work, because all functions (or methods) are strict in Java: that means that every time you call a function, all arguments are evaluated before they are passed to the function.
In Swift I would just write let name = Model.getDestination()?.getDevice()?.getName() ?? "". In Haskell it would be like name <- (destination >>= getDevice >>= getName) <|> Just "" (assuming the Maybe monad). And this has different semantics from this Java code:
if(Model.getDestination() && Model.getDestination().getDevice() && Model.getDestination().getDevice().getName() {
String name = Model.getDestination().getDevice().getName();
System.out.println("We got a name: "+name);
}
because this snippet calls getDestination() 4 times, getDevice() 3 times, getName() 2 times. This has more than just performance implications: 1) It introduces race conditions. 2) If any of the methods have side-effects, you don't want them to be called multiple times. 3) It makes everything harder to debug.
The only correct way of doing it is something like this:
Destination dest = Model.getDestination();
Device device = null;
String name = null;
if(dest != null) {
device = dest.getDevice();
if(device != null) {
name = device.getName();
}
}
if(name == null) {
name = "";
}
This code sets name to Model.getDestination().getDevice().getName(), or if any of these method calls return null, it sets name to "". I think correctness is more important than readability, especially for production applications (and even for example code IMHO). The above Swift or Haskell code is equivalent to that Java code.
If you have a production app, I guess that something like that is what you are already doing, because everything that is fundamentally different than that is error-prone.
Every better solution has to provide the same semantics and it MUST not call any of the methods (getDestination, getDevice, getName) more than once.
That said, I don't think you can simplify the code much with Java 7.
What you can do of course, is shorten the call chains: E.g. you could create a method getDeviceName() on Destination, if you need this functionality often. If this makes the code more readable depends on the concrete situation.
Forcing you to code on this low level also has advantages: you can do common subexpression elimination, and you'll see the advantages of it, because it will make the code shorter. E.g. if you have:
String name1 = Model.getDevice().getConnection().getContext().getName();
String name2 = Model.getDevice().getConnection().getContext().getLabel();
you can simplify them to
Context ctx = Model.getDevice().getConnection().getContext();
String name1 = ctx.getName();
String name2 = ctx.getLabel();
The second snippet has 3 lines, while the first snippet has only two lines. But if you unroll the two snippets to include null-checks, you will see that the second version is in fact much shorter. (I'm not doing it now because I'm lazy.)
Therefore (regarding Optional-chaining), Java 7 will make the code of the performance-aware coder look better, while many more high-level languages create incentives to make slow code. (Of course you can also do common subexpression elimination in higher level languages (and you probably should), but in my experience most developers are more reluctant to do it in high level languages. Whereas in Assembler, everything is optimized, because better performance often means you have to write less code and the code that you write is easier to understand.)
In a perfect word, we would all use languages that have built-in optional chaining, and we would all use it responsibly, without creating performance problems and race conditions.
You can use try-catch. Because there is no processing required in your case, like
try{
if(IsValid(Model.getDestination(), Model.getDestination().getDevice(), Model.getDestination().getDevice().getName())){
}catch(Exception e){
//do nothing
}
Alternatively you can improve your isValid method by passing only Model object
boolean isValid(Model model){
return (model != null && model.getDestination() != null && model.getDestination().getDevice() != null && model.getDestination().getDevice().getName() != null)
}
A web service returns a huge XML and I need to access deeply nested fields of it. For example:
return wsObject.getFoo().getBar().getBaz().getInt()
The problem is that getFoo(), getBar(), getBaz() may all return null.
However, if I check for null in all cases, the code becomes very verbose and hard to read. Moreover, I may miss the checks for some of the fields.
if (wsObject.getFoo() == null) return -1;
if (wsObject.getFoo().getBar() == null) return -1;
// maybe also do something with wsObject.getFoo().getBar()
if (wsObject.getFoo().getBar().getBaz() == null) return -1;
return wsObject.getFoo().getBar().getBaz().getInt();
Is it acceptable to write
try {
return wsObject.getFoo().getBar().getBaz().getInt();
} catch (NullPointerException ignored) {
return -1;
}
or would that be considered an antipattern?
Catching NullPointerException is a really problematic thing to do since they can happen almost anywhere. It's very easy to get one from a bug, catch it by accident and continue as if everything is normal, thus hiding a real problem. It's so tricky to deal with so it's best to avoid altogether. (For example, think about auto-unboxing of a null Integer.)
I suggest that you use the Optional class instead. This is often the best approach when you want to work with values that are either present or absent.
Using that you could write your code like this:
public Optional<Integer> m(Ws wsObject) {
return Optional.ofNullable(wsObject.getFoo()) // Here you get Optional.empty() if the Foo is null
.map(f -> f.getBar()) // Here you transform the optional or get empty if the Bar is null
.map(b -> b.getBaz())
.map(b -> b.getInt());
// Add this if you want to return null instead of an empty optional if any is null
// .orElse(null);
// Or this if you want to throw an exception instead
// .orElseThrow(SomeApplicationException::new);
}
Why optional?
Using Optionals instead of null for values that might be absent makes that fact very visible and clear to readers, and the type system will make sure you don't accidentally forget about it.
You also get access to methods for working with such values more conveniently, like map and orElse.
Is absence valid or error?
But also think about if it is a valid result for the intermediate methods to return null or if that is a sign of an error. If it is always an error then it's probably better throw an exception than to return a special value, or for the intermediate methods themselves to throw an exception.
Maybe more optionals?
If on the other hand absent values from the intermediate methods are valid, maybe you can switch to Optionals for them also?
Then you could use them like this:
public Optional<Integer> mo(Ws wsObject) {
return wsObject.getFoo()
.flatMap(f -> f.getBar())
.flatMap(b -> b.getBaz())
.flatMap(b -> b.getInt());
}
Why not optional?
The only reason I can think of for not using Optional is if this is in a really performance critical part of the code, and if garbage collection overhead turns out to be a problem. This is because a few Optional objects are allocated each time the code is executed, and the VM might not be able to optimize those away. In that case your original if-tests might be better.
I suggest considering Objects.requireNonNull(T obj, String message). You might build chains with a detailed message for each exception, like
requireNonNull(requireNonNull(requireNonNull(
wsObject, "wsObject is null")
.getFoo(), "getFoo() is null")
.getBar(), "getBar() is null");
I would suggest you not to use special return-values, like -1. That's not a Java style. Java has designed the mechanism of exceptions to avoid this old-fashioned way which came from the C language.
Throwing NullPointerException is not the best option too. You could provide your own exception (making it checked to guarantee that it will be handled by a user or unchecked to process it in an easier way) or use a specific exception from XML parser you are using.
Assuming the class structure is indeed out of our control, as seems to be the case, I think catching the NPE as suggested in the question is indeed a reasonable solution, unless performance is a major concern. One small improvement might be to wrap the throw/catch logic to avoid clutter:
static <T> T get(Supplier<T> supplier, T defaultValue) {
try {
return supplier.get();
} catch (NullPointerException e) {
return defaultValue;
}
}
Now you can simply do:
return get(() -> wsObject.getFoo().getBar().getBaz().getInt(), -1);
As already pointed out by Tom in the comment,
Following statement disobeys the Law of Demeter,
wsObject.getFoo().getBar().getBaz().getInt()
What you want is int and you can get it from Foo. Law of Demeter says, never talk to the strangers. For your case you can hide the actual implementation under the hood of Foo and Bar.
Now, you can create method in Foo to fetch int from Baz. Ultimately, Foo will have Bar and in Bar we can access Int without exposing Baz directly to Foo. So, null checks are probably divided to different classes and only required attributes will be shared among the classes.
My answer goes almost in the same line as #janki, but I would like to modify the code snippet slightly as below:
if (wsObject.getFoo() != null && wsObject.getFoo().getBar() != null && wsObject.getFoo().getBar().getBaz() != null)
return wsObject.getFoo().getBar().getBaz().getInt();
else
return something or throw exception;
You can add a null check for wsObject as well, if there's any chance of that object being null.
You say that some methods "may return null" but do not say in what circumstances they return null. You say you catch the NullPointerException but you do not say why you catch it. This lack of information suggests you do not have a clear understanding of what exceptions are for and why they are superior to the alternative.
Consider a class method that is meant to perform an action, but the method can not guarantee it will perform the action, because of circumstances beyond its control (which is in fact the case for all methods in Java). We call that method and it returns. The code that calls that method needs to know whether it was successful. How can it know? How can it be structured to cope with the two possibilities, of success or failure?
Using exceptions, we can write methods that have success as a post condition. If the method returns, it was successful. If it throws an exception, it had failed. This is a big win for clarity. We can write code that clearly processes the normal, success case, and move all the error handling code into catch clauses. It often transpires that the details of how or why a method was unsuccessful are not important to the caller, so the same catch clause can be used for handling several types of failure. And it often happens that a method does not need to catch exceptions at all, but can just allow them to propagate to its caller. Exceptions due to program bugs are in that latter class; few methods can react appropriately when there is a bug.
So, those methods that return null.
Does a null value indicate a bug in your code? If it does, you should not be catching the exception at all. And your code should not be trying to second guess itself. Just write what is clear and concise on the assumption that it will work. Is a chain of method calls clear and concise? Then just use them.
Does a null value indicate invalid input to your program? If it does, a NullPointerException is not an appropriate exception to throw, because conventionally it is reserved for indicating bugs. You probably want to throw a custom exception derived from IllegalArgumentException (if you want an unchecked exception) or IOException (if you want a checked exception). Is your program required to provide detailed syntax error messages when there is invalid input? If so, checking each method for a null return value then throwing an appropriate diagnostic exception is the only thing you can do. If your program need not provide detailed diagnostics, chaining the method calls together, catching any NullPointerException and then throwing your custom exception is clearest and most concise.
One of the answers claims that the chained method calls violate the Law of Demeter and thus are bad. That claim is mistaken.
When it comes to program design, there are not really any absolute rules about what is good and what is bad. There are only heuristics: rules that are right much (even almost all) of the time. Part of the skill of programming is knowing when it is OK to break those kinds of rules. So a terse assertion that "this is against rule X" is not really an answer at all. Is this one of the situations where the rule should be broken?
The Law of Demeter is really a rule about API or class interface design. When designing classes, it is useful to have a hierarchy of abstractions. You have low level classes that uses the language primitives to directly perform operations and represent objects in an abstraction that is higher level than the language primitives. You have medium level classes that delegate to the low level classes, and implement operations and representations at a higher level than the low level classes. You have high level classes that delegate to the medium level classes, and implement still higher level operations and abstractions. (I've talked about just three levels of abstraction here, but more are possible). This allows your code to express itself in terms of appropriate abstractions at each level, thereby hiding complexity. The rationale for the Law of Demeter is that if you have a chain of method calls, that suggests you have a high level class reaching in through a medium level class to deal directly with low level details, and therefore that your medium level class has not provided a medium-level abstract operation that the high level class needs. But it seems that is not the situation you have here: you did not design the classes in the chain of method calls, they are the result of some auto-generated XML serialization code (right?), and the chain of calls is not descending through an abstraction hierarchy because the des-serialized XML is all at the same level of the abstraction hierarchy (right?)?
As others have said, respecting the Law of Demeter is definitely part of the solution. Another part, wherever possible, is to change those chained methods so they cannot return null. You can avoid returning null by instead returning an empty String, an empty Collection, or some other dummy object that means or does whatever the caller would do with null.
To improve readability, you may want to use multiple variables, like
Foo theFoo;
Bar theBar;
Baz theBaz;
theFoo = wsObject.getFoo();
if ( theFoo == null ) {
// Exit.
}
theBar = theFoo.getBar();
if ( theBar == null ) {
// Exit.
}
theBaz = theBar.getBaz();
if ( theBaz == null ) {
// Exit.
}
return theBaz.getInt();
Don't catch NullPointerException. You don't know where it is coming from (I know it is not probable in your case but maybe something else threw it) and it is slow.
You want to access the specified field and for this every other field has to be not null. This is a perfect valid reason to check every field. I would probably check it in one if and then create a method for readability. As others pointed out already returning -1 is very oldschool but I don't know if you have a reason for it or not (e.g. talking to another system).
public int callService() {
...
if(isValid(wsObject)){
return wsObject.getFoo().getBar().getBaz().getInt();
}
return -1;
}
public boolean isValid(WsObject wsObject) {
if(wsObject.getFoo() != null &&
wsObject.getFoo().getBar() != null &&
wsObject.getFoo().getBar().getBaz() != null) {
return true;
}
return false;
}
Edit: It is debatable if it's disobeyes the Law Of Demeter since the WsObject is probably only a data structure (check https://stackoverflow.com/a/26021695/1528880).
If you don't want to refactor the code and you can use Java 8, it is possible to use Method references.
A simple demo first (excuse the static inner classes)
public class JavaApplication14
{
static class Baz
{
private final int _int;
public Baz(int value){ _int = value; }
public int getInt(){ return _int; }
}
static class Bar
{
private final Baz _baz;
public Bar(Baz baz){ _baz = baz; }
public Baz getBar(){ return _baz; }
}
static class Foo
{
private final Bar _bar;
public Foo(Bar bar){ _bar = bar; }
public Bar getBar(){ return _bar; }
}
static class WSObject
{
private final Foo _foo;
public WSObject(Foo foo){ _foo = foo; }
public Foo getFoo(){ return _foo; }
}
interface Getter<T, R>
{
R get(T value);
}
static class GetterResult<R>
{
public R result;
public int lastIndex;
}
/**
* #param args the command line arguments
*/
public static void main(String[] args)
{
WSObject wsObject = new WSObject(new Foo(new Bar(new Baz(241))));
WSObject wsObjectNull = new WSObject(new Foo(null));
GetterResult<Integer> intResult
= getterChain(wsObject, WSObject::getFoo, Foo::getBar, Bar::getBar, Baz::getInt);
GetterResult<Integer> intResult2
= getterChain(wsObjectNull, WSObject::getFoo, Foo::getBar, Bar::getBar, Baz::getInt);
System.out.println(intResult.result);
System.out.println(intResult.lastIndex);
System.out.println();
System.out.println(intResult2.result);
System.out.println(intResult2.lastIndex);
// TODO code application logic here
}
public static <R, V1, V2, V3, V4> GetterResult<R>
getterChain(V1 value, Getter<V1, V2> g1, Getter<V2, V3> g2, Getter<V3, V4> g3, Getter<V4, R> g4)
{
GetterResult result = new GetterResult<>();
Object tmp = value;
if (tmp == null)
return result;
tmp = g1.get((V1)tmp);
result.lastIndex++;
if (tmp == null)
return result;
tmp = g2.get((V2)tmp);
result.lastIndex++;
if (tmp == null)
return result;
tmp = g3.get((V3)tmp);
result.lastIndex++;
if (tmp == null)
return result;
tmp = g4.get((V4)tmp);
result.lastIndex++;
result.result = (R)tmp;
return result;
}
}
Output
241
4
null
2
The interface Getter is just a functional interface, you may use any equivalent.
GetterResult class, accessors stripped out for clarity, hold the result of the getter chain, if any, or the index of the last getter called.
The method getterChain is a simple, boilerplate piece of code, that can be generated automatically (or manually when needed).
I structured the code so that the repeating block is self evident.
This is not a perfect solution as you still need to define one overload of getterChain per number of getters.
I would refactor the code instead, but if can't and you find your self using long getter chains often you may consider building a class with the overloads that take from 2 to, say, 10, getters.
I'd like to add an answer which focus on the meaning of the error. Null exception in itself doesn't provide any meaning full error. So I'd advise to avoid dealing with them directly.
There is a thousands cases where your code can go wrong: cannot connect to database, IO Exception, Network error... If you deal with them one by one (like the null check here), it would be too much of a hassle.
In the code:
wsObject.getFoo().getBar().getBaz().getInt();
Even when you know which field is null, you have no idea about what goes wrong. Maybe Bar is null, but is it expected? Or is it a data error? Think about people who read your code
Like in xenteros's answer, I'd propose using custom unchecked exception. For example, in this situation: Foo can be null (valid data), but Bar and Baz should never be null (invalid data)
The code can be re-written:
void myFunction()
{
try
{
if (wsObject.getFoo() == null)
{
throw new FooNotExistException();
}
return wsObject.getFoo().getBar().getBaz().getInt();
}
catch (Exception ex)
{
log.error(ex.Message, ex); // Write log to track whatever exception happening
throw new OperationFailedException("The requested operation failed")
}
}
void Main()
{
try
{
myFunction();
}
catch(FooNotExistException)
{
// Show error: "Your foo does not exist, please check"
}
catch(OperationFailedException)
{
// Show error: "Operation failed, please contact our support"
}
}
NullPointerException is a run-time exception, so generally speaking is not recommended to catch it, but to avoid it.
You will have to catch the exception wherever you want to call the method (or it will propagate up the stack). Nevertheless, if in your case you can keep working with that result with value -1 and you are sure that it won't propagate because you are not using any of the "pieces" that may be null, then it seems right to me to catch it
Edit:
I agree with the later answer from #xenteros, it wil be better to launch your own exception instead returning -1 you can call it InvalidXMLException for instance.
Have been following this post since yesterday.
I have been commenting/voting the comments which says, catching NPE is bad. Here is why I have been doing that.
package com.todelete;
public class Test {
public static void main(String[] args) {
Address address = new Address();
address.setSomeCrap(null);
Person person = new Person();
person.setAddress(address);
long startTime = System.currentTimeMillis();
for (int i = 0; i < 1000000; i++) {
try {
System.out.println(person.getAddress().getSomeCrap().getCrap());
} catch (NullPointerException npe) {
}
}
long endTime = System.currentTimeMillis();
System.out.println((endTime - startTime) / 1000F);
long startTime1 = System.currentTimeMillis();
for (int i = 0; i < 1000000; i++) {
if (person != null) {
Address address1 = person.getAddress();
if (address1 != null) {
SomeCrap someCrap2 = address1.getSomeCrap();
if (someCrap2 != null) {
System.out.println(someCrap2.getCrap());
}
}
}
}
long endTime1 = System.currentTimeMillis();
System.out.println((endTime1 - startTime1) / 1000F);
}
}
public class Person {
private Address address;
public Address getAddress() {
return address;
}
public void setAddress(Address address) {
this.address = address;
}
}
package com.todelete;
public class Address {
private SomeCrap someCrap;
public SomeCrap getSomeCrap() {
return someCrap;
}
public void setSomeCrap(SomeCrap someCrap) {
this.someCrap = someCrap;
}
}
package com.todelete;
public class SomeCrap {
private String crap;
public String getCrap() {
return crap;
}
public void setCrap(String crap) {
this.crap = crap;
}
}
Output
3.216
0.002
I see a clear winner here. Having if checks is way too less expensive than catch an exception. I have seen that Java-8 way of doing. Considering that 70% of the current applications still run on Java-7 I am adding this answer.
Bottom Line For any mission critical applications, handling NPE is costly.
If efficiency is an issue then the 'catch' option should be considered.
If 'catch' cannot be used because it would propagate (as mentioned by 'SCouto') then use local variables to avoid multiple calls to methods getFoo(), getBar() and getBaz().
It's worth considering to create your own Exception. Let's call it MyOperationFailedException. You can throw it instead returning a value. The result will be the same - you'll quit the function, but you won't return hard-coded value -1 which is Java anti-pattern. In Java we use Exceptions.
try {
return wsObject.getFoo().getBar().getBaz().getInt();
} catch (NullPointerException ignored) {
throw new MyOperationFailedException();
}
EDIT:
According to the discussion in comments let me add something to my previous thoughts. In this code there are two possibilities. One is that you accept null and the other one is, that it is an error.
If it's an error and it occurs, You can debug your code using other structures for debugging purposes when breakpoints aren't enough.
If it's acceptable, you don't care about where this null appeared. If you do, you definitely shouldn't chain those requests.
The method you have is lengthy, but very readable. If I were a new developer coming to your code base I could see what you were doing fairly quickly. Most of the other answers (including catching the exception) don't seem to be making things more readable and some are making it less readable in my opinion.
Given that you likely don't have control over the generated source and assuming you truly just need to access a few deeply nested fields here and there then I would recommend wrapping each deeply nested access with a method.
private int getFooBarBazInt() {
if (wsObject.getFoo() == null) return -1;
if (wsObject.getFoo().getBar() == null) return -1;
if (wsObject.getFoo().getBar().getBaz() == null) return -1;
return wsObject.getFoo().getBar().getBaz().getInt();
}
If you find yourself writing a lot of these methods or if you find yourself tempted to make these public static methods then I would create a separate object model, nested how you would like, with only the fields you care about, and convert from the web services object model to your object model.
When you are communicating with a remote web service it is very typical to have a "remote domain" and an "application domain" and switch between the two. The remote domain is often limited by the web protocol (for example, you can't send helper methods back and forth in a pure RESTful service and deeply nested object models are common to avoid multiple API calls) and so not ideal for direct use in your client.
For example:
public static class MyFoo {
private int barBazInt;
public MyFoo(Foo foo) {
this.barBazInt = parseBarBazInt();
}
public int getBarBazInt() {
return barBazInt;
}
private int parseFooBarBazInt(Foo foo) {
if (foo() == null) return -1;
if (foo().getBar() == null) return -1;
if (foo().getBar().getBaz() == null) return -1;
return foo().getBar().getBaz().getInt();
}
}
return wsObject.getFooBarBazInt();
by applying the the Law of Demeter,
class WsObject
{
FooObject foo;
..
Integer getFooBarBazInt()
{
if(foo != null) return foo.getBarBazInt();
else return null;
}
}
class FooObject
{
BarObject bar;
..
Integer getBarBazInt()
{
if(bar != null) return bar.getBazInt();
else return null;
}
}
class BarObject
{
BazObject baz;
..
Integer getBazInt()
{
if(baz != null) return baz.getInt();
else return null;
}
}
class BazObject
{
Integer myInt;
..
Integer getInt()
{
return myInt;
}
}
Giving answer which seems different from all others.
I recommend you to check for NULL in ifs.
Reason :
We should not leave a single chance for our program to be crashed.
NullPointer is generated by system. The behaviour of System
generated exceptions can not be predicted. You should not leave your
program in the hands of System when you already have a way of handling
it by your own. And put the Exception handling mechanism for the extra safety.!!
For making your code easy to read try this for checking the conditions :
if (wsObject.getFoo() == null || wsObject.getFoo().getBar() == null || wsObject.getFoo().getBar().getBaz() == null)
return -1;
else
return wsObject.getFoo().getBar().getBaz().getInt();
EDIT :
Here you need to store these values wsObject.getFoo(),
wsObject.getFoo().getBar(), wsObject.getFoo().getBar().getBaz() in
some variables. I am not doing it because i don't know the return
types of that functions.
Any suggestions will be appreciated..!!
I wrote a class called Snag which lets you define a path to navigate through a tree of objects. Here is an example of its use:
Snag<Car, String> ENGINE_NAME = Snag.createForAndReturn(Car.class, String.class).toGet("engine.name").andReturnNullIfMissing();
Meaning that the instance ENGINE_NAME would effectively call Car?.getEngine()?.getName() on the instance passed to it, and return null if any reference returned null:
final String name = ENGINE_NAME.get(firstCar);
It's not published on Maven but if anyone finds this useful it's here (with no warranty of course!)
It's a bit basic but it seems to do the job. Obviously it's more obsolete with more recent versions of Java and other JVM languages that support safe navigation or Optional.
Let's say I had a lot of code between an if statement. Is it more proper to do a quick if-else check before it, and if it fails, return.
OR create the if statement with a lot of code in-between but not use return?
OR is it just a matter of preference?
so my 2 options are:
if(!something){
return
}
else
//lots of code here
if(something){
//lots of code here
}
From a performance perspective, you should always return from a function as quickly as you can, avoid doing unnecessary computations, "short-circuit" if you will. So checking for error cases and returning quickly would be the better policy.
Edit to add: In the same vein, you should always check the cases that are most likely to be violated first, this is sound advice when structuring your conditionals as well (|| and && checks)
I think this looks much nicer:
func() {
if(someCondition) {
return;
}
if(otherCondition) {
return;
}
//lots of code
}
than this:
func() {
if(someCondition) {
return;
} else if(otherCondition) {
return;
} else {
//lots of code
}
}
or this:
func() {
if(!someCondition) {
if(!otherCondition) {
//lots of code
}
}
}
It looks even uglier with more conditions, so I generally use the first method.
I prefer "shortcut". It has nothing to do with performance, as modern computer can handle if-else very fast, so we should focus on code readability.
However, if there's so many if-else in code, you may re-think your design. Refactory can be a better choice.
Readability and performance are not necessary conflicting constraints but when they are I tend to give readability the front seat.
To enhance readability I tend to follow the following rules.
Rule 1. Keep return as the last line of code, whatever comes in the middle. In other words don't sprinkle return statements whenever you want just because you're not too sure your if-else structure will cascade down just before the final return.
Except may be for the simplest methods I privilege a structure like
MyType func() {
MyType result ;
if ( condition ) {
result = result_1 ;
} else {
result = result_2 ;
}
return result ;
}
over an allegedly simpler
MyType func() {
if ( condition ) {
return result_1 ;
} else {
return result_2 ;
}
}
In my opinion the performance cost, if any, is negligible. However, when scaled up, I find the first coding pattern much more readable.
Rule 2. Refrain from starting a logic by "evacuating" error conditions, just in order to get them out of the way and free your mind. If your logic is well thought these checks will find their place in the logic (also have a look at guava for many well though techniques of encapsulating routine checks in helpers).
Many freshmen in my team start coding things like this
MyType func (ArgType arg1,...) {
if ( arg1 == null ) {
throw new Exception ( "hey dummy, we don't take null arg1) ;
// or return null ;
}
if ( arg2 == null ) {
// you got the picture...
}
// wow at last !!! all checks done
// Combine args and return result...
}
Which I have to say, is already a progress on just taking all conditions for granted
I tend to prefer
MyType func (ArgType arg1,...) {
MyType result ;
if ( try_to_compact_all_checks_here ) {
// Combine args and return result...
} else {
// throw, log, nullify result etc
}
return result ;
}
If the condition "try_to_compact_all_checks_here" does not fit in one line, I even sometimes prefer to get out of my way and I encapsulate all the checks in a private function. Even if it's called only once.
Rule 3. Keep the number of lines in an if/else statement to a reasonable amount (basically should fit on one screen in your IDE). To do so it is sometimes possible to extract some logic and stick it into a private function. No problem at all. All modern IDE do that for you in 2 clicks.
So basically the previous template becomes.
MyType func (ArgType arg1,...) {
MyType result ;
if ( checks_here ) {
// 1 to 20 lines max,
encapsulate lengthy logic in full fledged private methods.
} else {
// throw, log, nullify result etc
}
return result ;
}
Rule 4. Inner IFs should always have an ELSE, and that ELSE should be different from the outer ELSE.
Explanation: If I end up with
MyType func (ArgType arg1,...) {
MyType result ;
if ( check_1 ) {
if (check_2) {
Do the real work
} else {
treat error condition
}
} else {
same error condition as above
}
return result ;
}
Then it's probably because my check analysis is not complete. It happens quite often.
I try to reach
MyType func (ArgType arg1,...) {
MyType result ;
if ( check_1 && check_2) {
Do the real work
} else {
same error condition as above
}
return result ;
}
That's all.
I found that, by observing this kind of conventions, I can process large Java projects with ofter complex business logics (like in ESBs, Web Services etc), at very little performance cost if any.
Is there any real reason of providing the return statement in Scala? (aside from being more "Java-friendly")
Ignoring nested functions, it is always possible to replace Scala calculations with returns with equivalent calculations without returns. This result goes back to the early days of "structured programming", and is called the structured program theorem, cleverly enough.
With nested functions, the situation changes. Scala allows you to place a "return" buried deep inside series of nested functions. When the return is executed, control jumps out of all of the nested functions, into the the innermost containing method, from which it returns (assuming the method is actually still executing, otherwise an exception is thrown). This sort of stack-unwinding could be done with exceptions, but can't be done via a mechanical restructuring of the computation (as is possible without nested functions).
The most common reason you actually would want to return from inside a nested function is to break out of an imperative for-comprehension or resource control block. (The body of an imperative for-comprehension gets translated to a nested function, even though it looks just like a statement.)
for(i<- 1 to bezillion; j <- i to bezillion+6){
if(expensiveCalculation(i, j)){
return otherExpensiveCalculation(i, j)
}
withExpensiveResource(urlForExpensiveResource){ resource =>
// do a bunch of stuff
if(done) return
//do a bunch of other stuff
if(reallyDoneThisTime) return
//final batch of stuff
}
It is provided in order to accommodate those circumstances in which it is difficult or cumbersome to arrange all control flow paths to converge at the lexical end of the method.
While it is certainly true, as Dave Griffith says, that you can eliminate any use of return, it can often be more obfuscatory to do so than to simply cut execution short with an overt return.
Be aware, too, that return returns from methods, not function (literals) that may be defined within a method.
Here is an example
This method has lots of if-else statements to control flow, because there is no return (that is what I came with, you can use your imagination to extend it). I took this from a real life example and modified it to be a dummy code (in fact it is longer than this):
Without Return:
def process(request: Request[RawBuffer]): Result = {
if (condition1) {
error()
} else {
val condition2 = doSomethingElse()
if (!condition2) {
error()
} else {
val reply = doAnotherThing()
if (reply == null) {
Logger.warn("Receipt is null. Send bad request")
BadRequest("Coudln't receive receipt")
} else {
reply.hede = initializeHede()
if (reply.hede.isGood) {
success()
} else {
error()
}
}
}
}
}
With Return:
def process(request: Request[RawBuffer]): Result = {
if (condition1) {
return error()
}
val condition2 = doSomethingElse()
if (!condition2) {
return error()
}
val reply = doAnotherThing()
if (reply == null) {
Logger.warn("Receipt is null. Send bad request")
return BadRequest("Coudln't receive receipt")
}
reply.hede = initializeHede()
if (reply.hede.isGood)
return success()
return error()
}
To my eyes, the second one is more readable and even manageable than the first one. The depth of indentation (with well formatted code) goes deep and deep if you don't use a return statement. And I don't like it :)
I view return as a useful when writing imperative style code, which generally means I/O code. If you're doing pure functional code, you don't need (and should not use) return. But with functional code you may need laziness to get performance equivalent to imperative code that can "escape early" using return.