Is there a type inference regression in JDK 8 update 20?

Question

I have a problem with the following code:

public static <T> T firstNonNull(@Nullable T first, @Nullable T second) {     return first != null ? first : second; }  public static Set<String> getStrings() {    return new HashSet<>(); }  public static Set<String> doesNotCompile = firstNonNull(getStrings(), new HashSet<>()); 

With JDK 8 until update 11, this code compile. With JDK 8 update 20, it does not compile anymore. In the last statement, I have to explicitly specify the String type argument for the last HashSet instantiation.

I was wondering if I am wrong with this code or if it is a regression in the last JDK update.

Answer 1

This is a new bug that exists in the JDK 8u20 update release and in the current JDK 9 development branch. This code worked before, so yes, this would be considered a regression. The JDK langtools team has filed the following bug report:

JDK-8055963 Inference failure with nested invocation

Judging from the comments, it appears that the current behavior actually conforms to the specification (the JLS), but the behavior is clearly wrong, so it might be the case that a clarification to the specification is necessary.

Note that this is a different type inference bug from the one reported in this other StackOverflow question Java 1.8.20 Compiler Error, bug JDK-8051402. That bug has been fixed already, although but the fix isn't in JDK 8u20.