RFR: 8295099: vmTestbase/nsk/stress/strace/strace013.java failed with "TestFailure: wrong lengths of stack traces: strace013Thread0: NNN strace013Thread83: MMM"

[David Holmes]

On Mon, 7 Nov 2022 07:04:27 GMT, Thomas Stuefe <stuefe at openjdk.org> wrote:

>> The test is incorrectly synchronized and may take the stack trace of a thread before it has entered the wait() and thus be three frames short of the correct stack. So I cleaned that up to ensure things are correctly synchronized.
>> 
>> A number of comments in the test description etc were wrong so I also fixed those.
>> 
>> These tests are stylistically awful but I refrained from embarking on a major cleanup and adopted the prevailing styles.
>> 
>> Testing: local plus 50x on each platform (in progress)
>> 
>> Thanks.
>
> This looks fine. 
> 
> About that check at Line 171ff, where the test first compares stack lengths, I wonder whether the "k>2" was wiggle room for the problem you now solve? Could not explain it otherwise. If yes, you could probable require the length to be just equal now.
> 
> A CountDownLatch may make the coding a bit clearer maybe.

Thanks @tstuefe !

> I wonder whether the "k>2" was wiggle room for the problem you now solve? Could not explain it otherwise. If yes, you could probable require the length to be just equal now.

Yes I think you are right. I'll try making that change - though I'm loathe to spend too much time trying to clean this up as the more you look the more you find.

> A CountDownLatch may make the coding a bit clearer maybe.

Possibly it could avoid the sleep loop, but we still need a sync-block to guarantee the last thread is in the wait(). I'll stick with the current crude mechanism.

-------------

PR: https://git.openjdk.org/jdk/pull/11011