fix #23416 instantiate exceptions at right time

[francisdb]

fix #23416 instantiate exceptions at right time

[akka-ci]

Can one of the repo owners verify this patch?

[ktoso]

OK TO TEST

[akka-ci]

Test PASSed.

[ktoso]

I think this was intentional, since the stacktrace won't help you as much as you'd think it would...
NoStacktrace would address the perf issue - but is there really one? How often are these things thrown really

[francisdb]

I'm having this being thrown and am trying to debug what is going on, since the cause might be in the implementation of the reading/writing done by a different lib the stack trace would be more than welcome. Currently the trace points to the location where .run() is called which is very confusing.

java.io.IOException: Reactive stream is terminated, no writes are possible
	at akka.stream.impl.io.OutputStreamAdapter.<init>(OutputStreamSourceStage.scala:164)
	at akka.stream.impl.io.OutputStreamSourceStage.createLogicAndMaterializedValue(OutputStreamSourceStage.scala:151)
	at akka.stream.impl.GraphStageIsland.materializeAtomic(PhasedFusingActorMaterializer.scala:627)
	at akka.stream.impl.PhasedFusingActorMaterializer.materialize(PhasedFusingActorMaterializer.scala:458)
	at akka.stream.impl.PhasedFusingActorMaterializer.materialize(PhasedFusingActorMaterializer.scala:420)
	at akka.stream.impl.PhasedFusingActorMaterializer.materialize(PhasedFusingActorMaterializer.scala:415)
	at akka.stream.scaladsl.RunnableGraph.run(Flow.scala:439)
	at my.project.ImportExport$.readZip(ImportExport.scala:66)
...

where the line in my code is this

val is = source.toMat(StreamConverters.asInputStream(5.seconds))(Keep.right).run()
// the reading of the InputStream is handled further down in the code using a Future

The error is probably caused by the library I am using that writes to the stream and closes/flushes multiple times.

Also this is the only place in Akka streams where I see this pattern of pre-creating the exception. Sure this is not an accidental mistake?

[johanandren]

I don't think the stack trace will be useful, but I also think caching the exception was a premature optimisation, so LGTM

[francisdb]

added some more proof of the usefulness of this patch in the ticket:
#23416 (comment)

[patriknw]

LGTM