Refactor concurrency

[OliverO2]

This PR covers a number of issues regarding concurrency and multi-threading. It is based on the recent BlockHound PR and offers an alternative to #3604. Although this is a lot, I thought it would be better to put it in the open for everyone to consider and discuss.

I am currently examining the further evolution of concurrent testing around existing, but less documented parameters like concurrentSpecs and concurrentTests. This PR intends to provide a sound basis for that.

Details addressed in this PR:

• Avoid the use of runBlocking inside coroutines. Schedule all coroutines below the test engine layer via regular dispatchers, bringing the test environment more in line with actual production environments.

• Remove direct use of Java's Executors, delegating to kotlinx.coroutines-provided dispatchers instead.

• Preserve coroutine context (which includes lots of Kotest configuration) in multithreaded test executions.

• Remove hard-coded 1-day time limit for multithreaded test executions.

• Change tests relying on single-threading from a coroutine model to a thread model, removing coroutine invocations (delay), replacing those with thread invocations (Thread.sleep).

• Avoid leaking threads by closing dispatchers after use.

• Keep the assertionCounter synchronized with thread-switching coroutines.

• Remove fun <K, V> concurrentHashMap(): MutableMap<K, V>, which was not thread-safe.

• Make FixedThreadCoroutineDispatcherFactory.dispatcherAffinity thread-safe.

• Revert a change in ConcurrentTestSuiteScheduler.schedule by commit c316bbd, which replaced launch with async plus joinAll. This was functionally equivalent, but async was not needed, as there were no values returned, and joinAll was already performed by the enclosing coroutineScope.

Fixes #3022

[OliverO2]

Test failure seems unrelated to this PR:

> Task :kotest-property:jvmTest
com.sksamuel.kotest.property.arbitrary.BindShrinkTest[jvm]

  Test Shrinks all components to minimum value[jvm] FAILED

  java.lang.AssertionError: "Property test failed for inputs
  
  0) MaximumComponents(a=1000, b=1000, c=1000, d=1000, e=1000, f=1000, g=1000, h=1000, i=1000, j=1000, k=1000, l=1000, m=1000, n=1000)
  
  Caused by java.lang.AssertionError: 1000 should be < 100 at
  	com.sksamuel.kotest.property.arbitrary.BindShrinkTest$1$2$stdout$1$1$1.invokeSuspend(BindShrinkTest.kt:52)
  	com.sksamuel.kotest.property.arbitrary.BindShrinkTest$1$2$stdout$1$1$1.invoke(BindShrinkTest.kt)
  	com.sksamuel.kotest.property.arbitrary.BindShrinkTest$1$2$stdout$1$1$1.invoke(BindShrinkTest.kt)
  	io.kotest.property.internal.ProptestKt$proptest$3$2.invokeSuspend(proptest.kt:49)
  
  " should include substring "Shrink result (after 45 shrinks) => MaximumComponents(a=0, b=0, c=0, d=0, e=0, f=0, g=0, h=0, i=0, j=0, k=0, l=0, m=100, n=0)"

[sksamuel]

Prob some very intermittent prop test failure

…

On Sun, Aug 27, 2023, 11:18 AM OliverO2 ***@***.***> wrote: Test failure seems unrelated to this PR: > Task :kotest-property:jvmTest com.sksamuel.kotest.property.arbitrary.BindShrinkTest[jvm] Test Shrinks all components to minimum value[jvm] FAILED java.lang.AssertionError: "Property test failed for inputs 0) MaximumComponents(a=1000, b=1000, c=1000, d=1000, e=1000, f=1000, g=1000, h=1000, i=1000, j=1000, k=1000, l=1000, m=1000, n=1000) Caused by java.lang.AssertionError: 1000 should be < 100 at com.sksamuel.kotest.property.arbitrary.BindShrinkTest$1$2$stdout$1$1$1.invokeSuspend(BindShrinkTest.kt:52) com.sksamuel.kotest.property.arbitrary.BindShrinkTest$1$2$stdout$1$1$1.invoke(BindShrinkTest.kt) com.sksamuel.kotest.property.arbitrary.BindShrinkTest$1$2$stdout$1$1$1.invoke(BindShrinkTest.kt) io.kotest.property.internal.ProptestKt$proptest$3$2.invokeSuspend(proptest.kt:49) " should include substring "Shrink result (after 45 shrinks) => MaximumComponents(a=0, b=0, c=0, d=0, e=0, f=0, g=0, h=0, i=0, j=0, k=0, l=0, m=100, n=0)" — Reply to this email directly, view it on GitHub <#3623 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFVSGXNZQ5OJ7TKCJHLDJ3XXNXMZANCNFSM6AAAAAA3UH74XE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[OliverO2]

Seems like this one is causing the test failure:

inline fun captureStandardOut(fn: () -> Unit): String {
   val previous = System.out
   val buffer = ByteArrayOutputStream()
   System.setOut(PrintStream(buffer))
   try {
      fn()
      return String(buffer.toByteArray())
   } finally {
      System.setOut(previous)
   }
}

PrintStream(buffer) creates a stream without auto-flushing, so before returning the captured output, flush() should be invoked. I could correct this and other invocations in this PR.

[sksamuel]

This is green and ready to merge but I'd like to take a few days to read through and understand the changes if that's ok.

[OliverO2]

That sounds fantastic!

[sksamuel]

Couple more follow up questions added then we can merge in time for 5.7 release.

[sksamuel]

nice work :)

[OliverO2]

Thanks for the excellent cooperation. Way quicker than I had envisioned, given the size.

[sksamuel]

I'm eager to get 5.7 out this weekend :)