NonDaemonThreadLogger and SBT 1.6.x

[mattroberts297]

Using SBT 1.6.x the NonDaemonThreadLogger appears to print out SBT-related threads. Reproducible:

Main.scala

import cats.effect.{ExitCode, IO, IOApp}

object Main extends IOApp {
  def run(args: List[String]): IO[ExitCode] = {
    IO(ExitCode.Success)
  }
}

build.sbt

ThisBuild / scalaVersion     := "2.13.7"
ThisBuild / version          := "0.1.0-SNAPSHOT"
ThisBuild / organization     := "com.example"
ThisBuild / organizationName := "example"

lazy val root = (project in file("."))
  .settings(
    name := "sbt-cats-repro",
    libraryDependencies ++= Seq(
      "org.typelevel" %% "cats-effect" % "3.3.4"
    )
  )

project/build.properties

sbt.version=1.6.1

Output:

sbt:sbt-cats-repro> run
[info] running Main
[success] Total time: 1 s, completed 11-Jan-2022 16:44:22
sbt:sbt-cats-repro> Non-daemon threads currently preventing JVM termination: - 175: Thread[com.swoval.files.apple.FileEventsMonitor.runloop,5,main]
 -  - 23: Thread[pool-1-thread-4,5,main]
 -  - 180: Thread[sbt-socket-server,5,main]
 -  - 173: Thread[com.swoval.files.SymlinkWather.callback-executor-1,5,main]
 -  - 181: Thread[com.swoval.files.apple.FileEventsMonitor.callback-1,5,main]
 -  - 22: Thread[pool-1-thread-3,5,main]
 -  - 177: Thread[com.swoval.files.SymlinkWather.callback-executor-1,5,main]
 -  - 176: Thread[com.swoval.files.apple.FileEventsMonitor.runloop,5,main]
 -  - 179: Thread[com.swoval.files.apple.FileEventsMonitor.runloop,5,main]
 -  - 178: Thread[FileTreeRepository-callback-executor-1,5,main]
 -  - 29: Thread[sbt-console-input-reader,5,main]
 -  - 1: Thread[main,5,main]
 -  - 21: Thread[pool-1-thread-2,5,main]
 -  - 19: Thread[pool-1-thread-1,5,main]
 -  - 172: Thread[com.swoval.files.apple.FileEventsMonitor.runloop,5,main]
 -  - 174: Thread[FileTreeRepository-callback-executor-1,5,main]
 -  - 24: Thread[pool-1-thread-5,5,main]
 -  - 193: Thread[com.swoval.files.apple.FileEventsMonitor.callback-1,5,main]

I suspect this issue is now observable due to this SBT change: sbt/sbt#6665. Though when I attached a debugger to SBT 1.5.x I observed that the NonDaemonThreadLogger is terminated before it gets to detectThreads(), so perhaps this has always been an issue, but just not observable.

[djspiewak]

Yeah I think this has always been an issue, just not observable. The NonDaemonThreadLogger shouldn't be run in an unforked sbt run. It just doesn't have any meaning in that context, and it's not useful anyway because sbt's thread cleaner will take a wrecking ball to all of that anyway. Seeing as we now have the ability to detect this scenario, we just shouldn't run the logger in that case.

Quick workaround while we get to a fix: run / fork := true. This will also fix other issues since unforked runs really don't work very well in sbt anymore (e.g. ctrl-c works perfectly in Cats Effect in every case except unforked sbt, simply because unforked sbt has an unresolved bug which prevents it from forwarding the interrupt).

[mattroberts297]

Thank you @djspiewak and @vasilmkd 🙇‍♂️