On 12/11/2013 17:39, Jonathan Gibbons wrote:
On 11/12/2013 12:16 AM, Uwe Schindler wrote:
Hi Jon,
On 11/11/2013 04:24 PM, Uwe Schindler wrote:
We are testing with latest Java 8 builds and random JVM settings like
bitness, garbage collector, compressed oops,... and found, as Rory
mentioned, many bugs, especially in Hotspot's optimizer leading to
data loss
in Lucene.
Uwe,
Have you reported your findings to the appropriate dev tems,
especially the
Hotspot team, at hotspot-...@openjdk.java.net?
Of course, we worked together with the quality team to get those bugs
reported. The one affecting Java 7u40 and leading to SIGSEGV / data
loss because of incorrect method return values was fixed and
committed yesterday by Vladimir Kozlov
(https://issues.apache.org/jira/browse/LUCENE-5212,
https://bugs.openjdk.java.net/browse/JDK-8024830). We are still
investigating 2 more Java bugs only happening on MacOSX: One of them
is already known, but happens all the time (SocketInputStream
sigsegvs on unexpected errno/whatever). The other one is that OSX's
libc complains about double free() or free() with invalid pointer.
This one is hard to reproduce, but happens from time to time while
running tests.
Hi Jon, Uwe - we are still investigating these issues, as Uwe mentioned
the second issue is really hard to reproduce.
About Java 8 we found (not complete, I just copypasted this list from
an earlier talk - we should mark those bugs in our own bugtracker):
- The recent Object#clone() IllegalAccessException with nested
interfaces, fixed in 8b114
- G1 garbage collector deadlock due to marking stack overflow, fixed
already
- Compile failures with -source 1.7 related to default interface
methods (“isAnnotationPresent”), fixed already
- various javadoc bugs (all fixed): new doclint feature did not work,
doc-files folders were not copied
Thank you Uwe for these examples of how working together we can improve
the quality !
In general, if we find new bugs, we inform Rory and try to open bug
report on the legacy bugtracker bugs.sun.com (Access to JIRA would be
much better)! If we know the reason for the bug, we always send mails
to the approp mailing list. In most cases to the hotspot one to get
Vladimir on the boat.
Uwe
Uwe,
Thanks for the update. It sounds like this is a good effort that is
paying off well.
-- Jon
I agree, this is working well .
Rgds,Rory
--
Rgds,Rory O'Donnell
Quality Engineering Manager
Oracle EMEA , Dublin, Ireland
