RFR 8245308 : Replace ThreadLocalCoders decoder/encoder cache in java.net.URI

[Claes Redestad]

On 2020-08-27 16:11, Alan Bateman wrote:
> 
> For the micro, I'm curious if iterations is needed.

I'd say the canonical JMH way of doing these might be something as
simple as this:

     @Benchmark
     public URI encodeURI() throws URISyntaxException {
         return new URI("http", "\u00A0", "\u00A0");
     }

     @Benchmark
     public String decodeURI() throws URISyntaxException {
         return decoderUri.getPath();
     }

Sometimes for very simple micros like this this means the
workload gets inlined so aggressively that you no longer
measure what is likely to happen in the real world, and it's then
not uncommon to force the benchmark method to not be inlined up
into the JMH framework by annotating them with this:

     @CompilerControl(CompilerControl.Mode.DONT_INLINE)

See for example org/openjdk/bench/java/util/ImmutableColls.java

If you run with these suggestions you'll only one have encode/decode
per iteration, so you might want to change output to nanoseconds, i.e.:
@OutputTimeUnit(TimeUnit.NANOSECONDS)

Thanks!

/Claes