This oughta be fun:

[test] [error] 1 of 12 test tasks failed:
[test] [error] - macroAnnot/test
[test] [error]   - macroAnnot/compile:compileIncremental failed: java.lang.NoSuchMethodError: scala.tools.nsc.Global.<init>(Lscala/tools/nsc/Settings;Lscala/tools/nsc/reporters/Reporter;)V
[test] java.lang.RuntimeException
[test] 	at $33f7ec16b05f4d72fd8b$$anonfun$root$26.apply(build.sbt:919)
[test] 	at $33f7ec16b05f4d72fd8b$$anonfun$root$26.apply(build.sbt:858)
[test] 	at scala.Function1$$anonfun$compose$1.apply(Function1.scala:47)
[test] 	at sbt.$tilde$greater$$anonfun$$u2219$1.apply(TypeFunctions.scala:40)
[test] 	at sbt.std.Transform$$anon$4.work(System.scala:63)
[test] 	at sbt.Execute$$anonfun$submit$1$$anonfun$apply$1.apply(Execute.scala:228)
[test] 	at sbt.Execute$$anonfun$submit$1$$anonfun$apply$1.apply(Execute.scala:228)
[test] 	at sbt.ErrorHandling$.wideConvert(ErrorHandling.scala:17)
[test] 	at sbt.Execute.work(Execute.scala:237)
[test] 	at sbt.Execute$$anonfun$submit$1.apply(Execute.scala:228)
[test] 	at sbt.Execute$$anonfun$submit$1.apply(Execute.scala:228)
[test] 	at sbt.ConcurrentRestrictions$$anon$4$$anonfun$1.apply(ConcurrentRestrictions.scala:159)
[test] 	at sbt.CompletionService$$anon$2.call(CompletionService.scala:28)
[test] 	at java.util.concurrent.FutureTask.run(FutureTask.java:266)
[test] 	at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
[test] 	at java.util.concurrent.FutureTask.run(FutureTask.java:266)
[test] 	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
[test] 	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
[test] 	at java.lang.Thread.run(Thread.java:748)
[test] [error] (root/*:testAll) java.lang.RuntimeException
[test] [error] Total time: 1430 s, completed May 12, 2018 9:08:52 PM

Unrelatedly, I wanted to benchmark for refactor AbstractReporter to a filter, but managed to get jmh to fail with a VerifyError. I don't know if this is a usual interaction? When I'm more comfortable with jmh I'll update.

java.lang.VerifyError: Inconsistent stackmap frames at branch target 81
Exception Details:
  Location:
    scala/tools/nsc/reporters/ReportingBenchmark.init()V @81: aload_1
  Reason:
    Type 'scala/tools/nsc/reporters/ReportingBenchmark$$anon$1' (current frame, locals[1]) is not assignable to 'scala/reflect/internal/Reporter' (stack map, locals[1])

I guess it happens sometimes.

I did a quickie to restore the previous Global ctor, but now:

[test] [error] - macroAnnot/compile:compileIncremental failed: java.lang.NoSuchMethodError: xsbt.CachedCompiler0$Compiler.reporter_$eq(Lscala/tools/nsc/reporters/Reporter;)V

I guess I'll cross that bridge, too, now that I've come to it.

I guess that was it. Not sure exactly what is different for sbt about macro-annot, or what a migration path would be; but maybe just leave it as-is.

@adriaanm re: scala/bug#10869 (comment)

This takes internal.Reporter so that you get error, warn, and echo. The built-in controls are -Xmaxerrs, -Xmaxwarns, and -nowarn, and -Xfatal-warnings.

A custom reporter could filter echo on some criterion if it wants to. Or maybe I already added reporters.VerboseReporter.

The compiler (after previous refactor) tests -verbose before echoing, rather than the reporter checking verbosity.

I think the simplicity will benefit any solution for "warning suppression", in avoiding the "where is my output?" moments you get with logging systems.

Regarding macroAnnot, Lukas diagnosed it as a limitation with the sbt compiler interface. The macro annot project uses the quick compiler for its scala instance, but the rest of the project uses starr. The macroAnnot tests should be ported to partest, and the quick scalaInstance dropped.

@adriaanm thx for the xplanation. Also, my previous pun, "cross that bridge [method]", was a good one. Maybe an old one.

I'm not sure what to think of this PR. Maybe I misunderstand it / its goals.

The PR adds casts and type tests, and IMO doesn't make the code easier to understand, while nsc.reporters.Reporter is still there. It also duplicates some functionality (eg from ConsoleReporter into DisplayReporter). I fear that the resulting code we end up having is even more confusing than the current state, and it doesn't really bring us closer to removing nsc.reporters.Reporter, since sbt continues to use it (https://github.com/sbt/sbt/blob/1.x/main-actions/src/main/scala/sbt/compiler/Eval.scala#L15). I think it would be better work on the other end first (remove usages of reporters.Reporter, think what to do for def comment, var cancelled), and clean up the hierarchy when we actually can.

@lrytz I see there are two commits, one to address the TODO about fixing internal.Reporter when partest is fixed (which is possible now); the second refactors the AbstractReporter/ConsoleReporter.

The comments about sbt/ide/partest now just say sbt/ide. If you can't do anything because sbt, then I'd give up.

I could let Global still take reporters.Reporter natively and adapt internal.Reporter; that would avoid the downcasting for cancel and comment.

That sounds better. Let's do the actual cleanups, but not anticipate future ones by adding more code.

My broom broke.

Ah, merge conflict.. Also maybe @adriaanm wants to take a quick scan of the change (maybe not).

Rebased, accommodated recent change to avoid System.exit.

Interesting failure in https://scala-ci.typesafe.com/job/scala-2.13.x-validate-main/2311/testReport/junit/scala.collection/FloatFormatTest/toOption/

Could this be due to recent changes to LazyList?

/cc @szeiger could you take a look?

I think it's just a bug in #6726

/cc @martijnhoekstra

#6746

/rebuild

Noting that this PR did not change the default for -Xreporter, since it would require adapting for Global.

I noticed that in the 2.13.x branch, the benchmark harness which sets settings.nowarn = true no longer manages to suppress warnings. I guessing that it is related to this PR, but haven't figured out the details.

@retronym I'll take a look. I also neglected to update ReplReporter, which I'm doing now.

@retronym Don't be mad, but apparently it was intentional to provide a minimal reporter for displaying errors before the compiler is built. Normally the real reporter is created with Global, and then Driver.process adopts global.reporter, replacing the temporary. In this case, Global is created with the current, minimal reporter; the benchmark global wouldn't respect -Xreporter if it were set, for instance.

I'll do the obvious thing and revert to creating a full-featured reporter for that start-up sequence. Not sure if there's a reason not to use Global.reporter (on the companion) right away.