Compilation warnings and a perf question

[Aleksey Shipilev]

Hi,

On 6/21/21 4:02 PM, Dmitry Vyazelenko wrote:
> - Warnings when generating and compiling code.
>    * The annotation processor:
>    ```
>    warning: No processor claimed any of these annotations: org.openjdk.jcstress.annotations.Actor,org.openjdk.jcstress.annotations.Arbiter,org.openjdk.jcstress.annotations.Outcome,org.openjdk.jcstress.annotations.Outcome.Outcomes,org.openjdk.jcstress.annotations.State
>    ```
>    * Generated code:
>    ```
>    warning: [rawtypes] found raw type: CounterThread
>          for (CounterThread t : threads) {
>               ^
>    missing type arguments for generic class CounterThread<R>
>    where R is a type-variable:
>      R extends Object declared in class CounterThread
>    ```
>    Both could be silenced with `-Xlint:-processing,-rawtypes`.

How exactly did you reproduce this?

I understand the issues, but would like to get builds to fail automatically.


> - Generation of the HTML report takes forever when a test has lots of unique outcomes and a wildcard matcher is used.
> The following test takes 6 minutes to complete when the test is executed without special command line options.
> ```
> import java.util.concurrent.atomic.AtomicLong;
> 
> @JCStressTest
> @Outcome(id = ".*, true", expect = Expect.ACCEPTABLE, desc = "Boring")
> @State
> public class WildcardResultsTest

...

> Attached is the CPU profile taken with an async-profiler. As can be seen from the profile 96% of time is taken by the `org.openjdk.jcstress.infra.grading.HTMLReportPrinter#emitTestReport` method.

This pain is instructional: you have way too many results on record, because the test initializes 
AtomicLong to System.currentTimeMillis() *and* records current value of AtomicLong into the result!

You probably want to write it as:

@JCStressTest
@Outcome(id = "true", expect = Expect.ACCEPTABLE, desc = "Boring")
@State
public class WildcardResultsTest {
     private final AtomicLong counter = new AtomicLong(System.currentTimeMillis());

     long c1, c2;

     @Actor
     public void actor1() {
         c1 = counter.getAndIncrement();
     }

     @Actor
     public void actor2() {
         c2 = counter.getAndIncrement();
     }

     @Arbiter
     public void arbiter(Z_Result result) {
         result.r1 = (c1 != c2);
     }
}

I am actually surprised only the HTML report printer is lagging. It lags for several reasons: the 
test grading is not cached, so it matches the regexp multiple times over all results; and HTML 
report builds a huge table that merges the results from many invocations. That HTML table is 42M long!

It can be helped a bit:
   https://github.com/openjdk/jcstress/pull/86

...but really the test needs fixing.

-- 
Thanks,
-Aleksey