我认为榴莲的错误课程结合了上面各种建议的许多优点。

将抛出函数包装到标准Java 8函数接口。 轻松指定各种处理错误的策略 在包装返回值的方法时，指定默认值和重新抛出RuntimeException之间有重要区别。 抛出Java 8的函数接口版本 类似于fge的答案 用于抛出特定异常的标准接口 这解决了Zoltán的问题

在你的项目中加入榴莲，你可以:

从jcenter或maven中心com.diffplug.durian:durian:3.3.0获取 或者只是复制粘贴两个小类到你的代码中:throws .java和Errors.java

免责声明:我还没有使用过Java 8，只是阅读过它。

Function<String, Integer>不抛出IOException，所以你不能在其中放入任何抛出IOException的代码。如果你正在调用一个期望Function<String, Integer>的方法，那么你传递给该方法的lambda不能抛出IOException，句号。你可以这样写一个lambda(我认为这是lambda语法，不确定):

(String s) -> {
    try {
        return myMethod(s);
    } catch (IOException ex) {
        throw new RuntimeException(ex);
        // (Or do something else with it...)
    }
}

或者，如果你传递lambda的方法是你自己写的，你可以定义一个新的函数接口，并使用它作为参数类型，而不是Function<String, Integer>:

public interface FunctionThatThrowsIOException<I, O> {
    O apply(I input) throws IOException;
}

我在课堂上遇到了这个问题。forName和Class。newInstance在lambda中，所以我做了:

public Object uncheckedNewInstanceForName (String name) {

    try {
        return Class.forName(name).newInstance();
    }
    catch (ClassNotFoundException | InstantiationException | IllegalAccessException e) {
        throw new RuntimeException(e);
    }
}

在lambda中，我没有调用Class.forName("myClass").newInstance()，而是调用uncheckedNewInstanceForName ("myClass")

这里已经贴出了很多很棒的回复。只是试图用不同的角度来解决问题。这只是我的两毛钱，如果我哪里说错了，请指正。

在FunctionalInterface中抛出子句不是一个好主意

我认为强制抛出IOException可能不是一个好主意，原因如下

This looks to me like an anti-pattern to Stream/Lambda. The whole idea is that the caller will decide what code to provide and how to handle the exception. In many scenarios, the IOException might not be applicable for the client. For example, if the client is getting value from cache/memory instead of performing actual I/O. Also, the exceptions handling in streams becomes really hideous. For example, here is my code will look like if I use your API acceptMyMethod(s -> { try { Integer i = doSomeOperation(s); return i; } catch (IOException e) { // try catch block because of throws clause // in functional method, even though doSomeOperation // might not be throwing any exception at all. e.printStackTrace(); } return null; }); Ugly isn't it? Moreover, as I mentioned in my first point, that the doSomeOperation method may or may not be throwing IOException (depending on the implementation of the client/caller), but because of the throws clause in your FunctionalInterface method, I always have to write the try-catch.

如果我知道这个API抛出IOException怎么办

Then probably we are confusing FunctionalInterface with typical Interfaces. If you know this API will throw IOException, then most probably you also know some default/abstract behavior as well. I think you should define an interface and deploy your library (with default/abstract implementation) as follows public interface MyAmazingAPI { Integer myMethod(String s) throws IOException; } But, the try-catch problem still exists for the client. If I use your API in stream, I still need to handle IOException in hideous try-catch block. Provide a default stream-friendly API as follows public interface MyAmazingAPI { Integer myMethod(String s) throws IOException; default Optional<Integer> myMethod(String s, Consumer<? super Exception> exceptionConsumer) { try { return Optional.ofNullable(this.myMethod(s)); } catch (Exception e) { if (exceptionConsumer != null) { exceptionConsumer.accept(e); } else { e.printStackTrace(); } } return Optional.empty(); } } The default method takes the consumer object as argument, which will be responsible to handle the exception. Now, from client's point of view, the code will look like this strStream.map(str -> amazingAPIs.myMethod(str, Exception::printStackTrace)) .filter(Optional::isPresent) .map(Optional::get).collect(toList()); Nice right? Of course, logger or other handling logic could be used instead of Exception::printStackTrace. You can also expose a method similar to https://docs.oracle.com/javase/8/docs/api/java/util/concurrent/CompletableFuture.html#exceptionally-java.util.function.Function- . Meaning that you can expose another method, which will contain the exception from previous method call. The disadvantage is that you are now making your APIs stateful, which means that you need to handle thread-safety and which will be eventually become a performance hit. Just an option to consider though.

我所做的是允许用户在异常情况下给出他实际想要的值。 我有一个像这样的东西

public static <T, R> Function<? super T, ? extends R> defaultIfThrows(FunctionThatThrows<? super T, ? extends R> delegate, R defaultValue) {
    return x -> {
        try {
            return delegate.apply(x);
        } catch (Throwable throwable) {
            return defaultValue;
        }
    };
}

@FunctionalInterface
public interface FunctionThatThrows<T, R> {
    R apply(T t) throws Throwable;
}

这个可以被称为:

defaultIfThrows(child -> child.getID(), null)

如果你不介意使用第三方自由库(Vavr)，你可以写

CheckedFunction1<String, Integer> f = this::myMethod;

它也有所谓的Try单子来处理错误:

Try(() -> f.apply("test")) // results in a Success(Integer) or Failure(Throwable)
        .map(i -> ...) // only executed on Success
        ...

请在这里阅读更多。

声明:我是Vavr的创造者。