The library tests have been running in agentvm mode and with concurrency
for several years and the intermittent CNFE/NCDFE issues have always
been worked around by avoiding implicit compilation. There seems to be
several more intermittent failures in jdk10/jdk10 compared to jdk9/dev
with tests that have been changed to revert to implicit compilation. So
I agree with Martin that Jon Gibbons (with his jtreg hat on) is the
right person to comment on this. If the issue is really
CODETOOLS-7901986 (and Ioi's analysis looks good) then I think we need
to get that into a jtreg build before changing lots of tests to not
explicitly compile the test infrastructure that they use.
-Alan
On 01/06/2017 04:04, Igor Ignatyev wrote:
http://cr.openjdk.java.net/~iignatyev//8181391/webrev.00/index.html
55 lines changed: 1 ins; 50 del; 4 mod;
Hi all,
due to CODETOOLS-7901986[1], tests sporadically fail w/ CNFE or NCDFE, having
explicit build actions has been proven to contribute into this[1-2]. this fix
removes explicit @build actions for classes from jdk.test.lib.* packages (the
top level testlibrary).
webrev: http://cr.openjdk.java.net/~iignatyev//8181391/webrev.00/index.html
jbs: https://bugs.openjdk.java.net/browse/JDK-8181391
testing: changed tests
[1] https://bugs.openjdk.java.net/browse/CODETOOLS-7901986
[2] https://bugs.openjdk.java.net/browse/JDK-8157957
[3] http://mail.openjdk.java.net/pipermail/hotspot-dev/2016-August/024198.html
Thanks,
-- Igor
