Javac benchmarks

[Aleksey Shipilev]

On 01/23/2014 01:30 AM, Staffan Friberg wrote:
> When running Javac as JMH microbenchmark it automatically picks up the
> annotation processor files that JMH use when generating the
> microbenchmarks.

Right. That is usually not the concern since presumably nobody uses the
microbenchmark artifact as dependency.

> This happens because the runtime microbenchmark.jar still contains the
> META-INF/services/javax.annotation.processing.Processor file which
> enumerates the annotation processing classes.
> 
> Would it be possible to automatically remove/not include this file when
> building the runtime microbenchmark.jar file?
> Guess this would not really be a change in JMH itself, but rather in a
> HOWTO or example pom.xml file on how to best use JMH.

Yup, unfortunately, this can only be done in users' POMs. However, we
can fix the archetype (which is the preferred way to bootstrap the JMH
benchmark), as well as disseminate the filters over JMH codebase in case
users pull off, say, jmh-sample/pom.xml as the example.

Archetype fix:
 http://hg.openjdk.java.net/code-tools/jmh/rev/9d70c6308d91

JMH subprojects fix:
 http://hg.openjdk.java.net/code-tools/jmh/rev/5990033f24ae

-Aleksey.