Java 25's new CPU-Time Profiler (mostlynerdless.de)

[flagged]

The JVMin in the last 6-8 years has been a powerhouse of innovation and cool features. Incredibly impressive!

[dead]

I never thought I would be excited for a new release of Java, but ever since Java 21, I have grown to actually enjoy writing the language. Whomever is running it has really done a good job making the language actually fun to write in the last few years.

If you care about performance at that level, you should not be using Java or any other language with a GC.

The more I search for a new language to learn the more I want to go back to Java. I feel so nostalgic :)

(author here) This is part of a series of blog post. The next blog posts are: https://mostlynerdless.de/blog/2025/07/30/java-25s-new-cpu-t... (on the implementation) https://mostlynerdless.de/blog/2025/08/25/java-25s-new-cpu-t... (on sizing the queue) and a fourth on improving the performance: https://mostlynerdless.de/blog/2025/09/01/java-25s-new-cpu-t...

This is built on a Linux sampling API that offers more accuracy, but it is still sampling.

For CPU tracing, with no sampling errors, use Apple’s M4 with the latest Xcode’s Instruments.

I just hope with new lightweight threads I never have to write async reactive code again. It was a such an unproductive mistake, most applications do not need that level of complexity - now we can confidently say no applications need that level of added complexity.

The biggest problem with Java these days is java developers are not really good at all. I would say n the last 10 years 99 percent of "Java" devs cant competently read a stack trace.

Im not sure why anyone would improve this language when the developers are so frankly pathetic. Let them use 1.8. The ecosystem has improved. Devs have gotten worse.

Java has become a land of opportunists just trying to pass of their non-Java maybe non-programmer skills

CPU-time would over-emphasize regions where many threads are running, right? I find wall-time useful for finding serial regions that aren't yet parallelized.

More detail here: https://github.com/dvyukov/perf-load. We recently implemented the same idea without requiring context-switch events: https://github.com/google/highway/blob/master/hwy/profiler.h...