Craig P. Motlin

While the performance argument may not be convincing, the point about toArray(emptyArray) being an atomic way of dealing with concurrent collections is compelling to me.

I have this impression that on concurrent collections, the only method that can conveniently and atomically give you O(n) elements that were at least in the collection at some point...

I am seeing different behaviors for different sizes. ```java public static void main(String[] args) { testWithPreSize(0); testWithPreSize(2); } private static void testWithPreSize(int preSize) { ConcurrentHashMap map = new java.util.concurrent.ConcurrentHashMap(); map.put("One",...

I agree that this isn't a real issue in practice. ConcurrentHashMap doesn't allow null keys nor null values, so if `keySet().toArray(preSizedArray)` or `values().toArray(preSizedArray)` results contain nulls, it unambiguously means that...

> The `@aQute.bnd.annotation.spi.ServiceProvider` is an annotation with class time retention. I'm familiar with com.google.auto.service.AutoService which seems similar and also has class time retention. What's the difference between then? Does ServiceProvider...

> I have no idea why the Javadoc build fails now. I tested it locally and there it works. Maybe you have an idea on this? The error message is:...

I think we should land the three items separately. If you don't mind, I'll split this up into three separate PRs.

Another matcher I just found that's not converted: ```java assertThat(object, instanceOf(MyType.class)) ```

That's right! I noticed that MDC iteration order is pseudorandom when working with logstash/logstash-logback-encoder. This surprised me a bit so I searched to see if anyone had written about it...

I was thinking plain JCF when I wrote up the issue. But I'm looking at what it would take to make boxed() return MutableList instead of List, and it looks...