Fixed issue #428

[zsxwing]

I fixed the issue #428. This issus is because the wrong order of unsubscribe and emitChunk.

unsubscribe will cancel the Future, and the currrent thread's interrupt status will be set. If unsubscribe is called before emitChunk, the currrent thread's interrupt status has already set before emitChunk. It may confuse users when they have some codes depending on the interrupt status.

I put emitChunk before unsubscribe and added a unit test for it.

[cloudbees-pull-request-builder]

RxJava-pull-requests #341 FAILURE
Looks like there's a problem with this pull request

[zsxwing]

It seems rx.operators.OperationConcat$UnitTest.testNestedAsyncConcat does not use TimeAndSizeBasedChunks. Is this some potential issue?

[cloudbees-pull-request-builder]

RxJava-pull-requests #342 SUCCESS
This pull request looks good

[benjchristensen]

Thank you for researching and fixing this!

[benjchristensen]

It seems rx.operators.OperationConcat$UnitTest.testNestedAsyncConcat does not use TimeAndSizeBasedChunks. Is this some potential issue?

What about that unit test suggests the need for TimeAndSizeBasedChunks?

[zsxwing]

What about that unit test suggests the need for TimeAndSizeBasedChunks?

No, I just want to say this is another example of #383.

[benjchristensen]

But what does a flaky unit test in concat have to do with TimeAndSizeBasedChunks? Or are you just saying rx.operators.OperationConcat$UnitTest.testNestedAsyncConcat is a flaky unit test?

[zsxwing]

I just said rx.operators.OperationConcat$UnitTest.testNestedAsyncConcat was a flaky unit test.