Bug 5091921: Sign flip issues in loop optimizer

[Vladimir Kozlov]

Hi Ismael,

It did not "fall through the cracks". We gave it relatively low priority since, 
as easy to reproduce it, it is also easy to avoid it by modifying java program. 
There were several attempts to fix it but they were not finished. The main 
problem is a performance regression associated with the fix since it is in the 
core of our loop optimizations. It is also complicated since, as  5091921 shows, 
there are several related problems which should be fixed. Recently we decided to 
accept the performance regression and fix it: currently the bug is targeted for 
jdk 7 release but I can't comment when exactly it will be fixed.

Vladimir

Ismael Juma wrote:
> Hi,
> 
> I came across a Lucene issue where it would read an index incorrectly due to a
> HotSpot bug[1]. In the issue tracker, they refer to two unfixed bugs in the
> Sun bug tracker: 5091921[2] and 7024475[3].
> 
> I am interested in the former because it was filed in 2004 and it has a number
> of duplicates (the most recent of which was filed last month - 7020614[4]). It
> seems pretty nasty and easy to reproduce with many isolated test cases, so I
> was wondering why it has remained unfixed for so long. Did it just fall
> through the cracks, is it complicated to fix or something else?
> 
> Best,
> Ismael
> 
> [1] https://issues.apache.org/jira/browse/LUCENE-2975 (hotspot bug in readvint 
> gives wrong results)
> [2] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=5091921
> [3] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7024475
> [4] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=7020614
> 
>