Heads-up: Lucene/Solr bug in loop predicates

[Aleksey Shipilev]

On 08/11/2017 10:16 AM, Roland Westrelin wrote:
> 
>> I have just reproduced it off the clean workspace. But I did set JAVA_HOME and PATH:
>>
>> $ export JAVA_HOME=~/trunks/shenandoah-jdk10/build/linux-x86_64-normal-server-fastdebug/images/jdk
>> $ export PATH=$JAVA_HOME/bin:$PATH
>>
>> If that fails, then we have to suspect native compiler bug?
> 
> It's not the same stack trace, right? It can't be because some stuff in
> the stack trace you sent doesn't exist upstream.

Yes, I have the same assert (n != _null) in phaseX.hpp as I sent before.


> I couldn't reproduce the failure you saw in loop predicates but I got
> another one:
> 
> #  assert(_refresh_tick < 2*100000) failed: DU iteration must converge quickly
> 
> in split if which is an upstream bug that I'm working on fixing.

Okay. I wonder if that points to timing, and my machine is just faster to arrive to assertion, while
yours is slower to finish the run until this hits. To test that, ran it on my much slower ultrabook,
and had the same assertion in phaseX!

The build I was using:
 https://shipilev.net/stuff/shenandoah-jdk10-fastdebug-lucenefail.tar.gz

-Aleksey