On 23/05/2020 21:13, Rob Spoor wrote:

Hi,

I was working on upgrading a library of mine from Java 8 to 11, and I noticed my unit tests started failing. Some investigation showed that in Java 9, java.util.Properties was rewritten to no longer rely on the fact that it extends Hashtable. One of the changes was to use a private static class called EntrySet. However, while this class implements most methods from java.util.Set, it's actually missing two: equals and hashCode. As a result it does not adhere to the general contract of java.util.Set. It's also missing a toString implementation, thereby inheriting its String representation from java.lang.Object. I've checked the source code up to Java 14, and even the Mercurial repository, and this issue still exists.

I think this could be solved as simply as by having this private EntrySet class extend AbstractSet, or otherwise delegate its methods to its entrySet field. The latter is probably preferred, as it matches the implementation of the same methods of the java.util.Properties class itself.

Rob

I've also noticed that keySet() and values() may behave differently if you add an entry with the keySet() as key or values() as value. While before that would have printed "(this Collection)", it will now cause a StackOverflowError because the key set or values collection no longer contains itself (which is the keySet() or values() from the backing ConcurrentHashMap), but a synchronized wrapper. While this is a rare corner case (nobody in their right mind would add anything but Strings to a java.util.Properties object), it is probably an unintended side effect of the rewrite of java.util.Properties.