What's the point of Guava checkNotNull

好久不见. 提交于 2019-11-28 18:06:12

The idea is to fail fast. For instance, consider this silly class:

public class Foo {
    private final String s;

    public Foo(String s) {
        this.s = s;
    }

    public int getStringLength() {
        return s.length();
    }
}

Let's say you don't want to allow null values for s. (or else getStringLength will throw a NPE). With the class as-is, by the time you catch that null, it's too late -- it's very hard to find out who put it there. The culprit could well be in a totally different class, and that Foo instance could have been constructed a long time ago. Now you have to comb over your code base to find out who could possibly have put a null value there.

Instead, imagine this constructor:

public Foo(String s) {
    this.s = checkNotNull(s);
}

Now, if someone puts a null in there, you'll find out right away -- and you'll have the stack trace pointing you exactly to the call that went wrong.


Another time this can be useful is if you want to check the arguments before you take actions that can modify state. For instance, consider this class that computes the average of all string lengths it gets:

public class StringLengthAverager {
    private int stringsSeen;
    private int totalLengthSeen;

    public void accept(String s) {
        stringsSeen++;
        totalLengthSeen += s.length();
    }

    public double getAverageLength() {
        return ((double)totalLengthSeen) / stringsSeen;
    }
}

Calling accept(null) will cause an NPE to get thrown -- but not before stringsSeen has been incremented. This may not be what you want; as a user of the class, I may expect that if it doesn't accept nulls, then its state should be unchanged if you pass a null (in other words: the call should fail, but it shouldn't invalidate the object). Obviously, in this example you could also fix it by getting s.length() before incrementing stringsSeen, but you can see how for a longer and more involved method, it might be useful to first check that all of your arguments are valid, and only then modify state:

    public void accept(String s) {
        checkNotNull(s); // that is, s != null is a precondition of the method

        stringsSeen++;
        totalLengthSeen += s.length();
    }

myObject.getAnything(); (which might cause a NullPointerException if myObject is null)

No... it will throw NPE whenever myObject == null. In Java, there's no chance of calling a method with null receiver (a theoretical exception are static methods, but they can and should be always called without any object).


I should use checkNotNull(myObject).getAnything();

No you should not. This would be rather redundant (Update).

You should use checkNotNull in order to fail fast. Without it, you may pass an illegal null to another method, which passes it further, and so on and so on, where it finally fails. Then you can need some good luck to find out that actually the very first method should have refused null.


The answer by yshavit mentions an important point: Passing an illegal value is bad, but storing it and passing it later is even worse.

Update

Actually,

 checkNotNull(myObject).getAnything()

makes sense, too, as you clearly express your intent to not accept any nulls. Without it, someone could think that you forgot the check and convert it into something like

 myObject != null ? myObject.getAnything() : somethingElse

OTOH, I don't think the check is worth the verbosity. In a better language, the type system would consider nullability and give us some semantic sugar like

 myObject!!.getAnything()                    // checkNotNull
 myObject?.getAnything()                     // safe call else null
 myObject?.getAnything() ?: somethingElse    // safe call else somethingElse

for nullable myObject, while the standard dot syntax would be allowed only when myObject is known to be non-null.

I have read this whole thread few minutes ago. Nevertheless I was confused why should we use checkNotNull. Then look over Precondition class doc of Guava and I got what I expected. Excessive use of checkNotNull will degrade performance definitely.

My thought is checkNotNull method is worth required for data validation which comes from user directly or very end API to user interaction. It shouldn't be used in each and every methods of internal API because using it you can't stop exception rather correct your internal API to avoid Exception.

According to DOC: Link

Using checkNotNull:

public static double sqrt(double value) {
     Preconditions.checkArgument(value >= 0.0, "negative value: %s", value);
     // calculate the square root
}

Warning about performance

The goal of this class is to improve readability of code, but in some circumstances this may come at a significant performance cost. Remember that parameter values for message construction must all be computed eagerly, and autoboxing and varargs array creation may happen as well, even when the precondition check then succeeds (as it should almost always do in production). In some circumstances these wasted CPU cycles and allocations can add up to a real problem. Performance-sensitive precondition checks can always be converted to the customary form:

if (value < 0.0) {
     throw new IllegalArgumentException("negative value: " + value);
}
易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!