Nightly test failure on 5/24:

gc/64bit/quicklook/largeheap/GC32BitRange

http://vmsqe.sfbay/nightly/mantis/DTWS/results/05-23-05/ServerVM/Solsparc/mixed/Gc_Baseline-Xinc/vm.gc-NIGHTLY-Gc_Baseline-Xinc-ServerVM-mixed-Solsparc-2005-05-24-07-35-33/analysis.html

When part of the heap has been swept and the remainder of the
heap that has not yet been swept has references to these
swept up objects from now-dead-but-not-yet swept objects,
we can get false assertions. The cross-section of
vulnerability increases when perm gen collection is
turned off or if we use incremental CMS.

We could explore having the sweeper mark the dead objects in
perm gen specially (when perm gen collection is not enabled
and we are running a debug jvm) so that they are skipped during
verification. We should make use of the CMS bit map
to determine object liveness during verification
that happens during a sweep.

###@###.### 2005-05-25 18:10:51 GMT