The following following bug:
6942989 2/2 Memory leak of java.lang.ref.WeakReference objects
was used to add the following tests:
These tests use 'jmap -histo:live' to sample instance counts to
determine if the memory leak is present or not. The tests were
originally written with a fall back to 'jmap -histo' if the bits
being tested didn't support 'jmap -histo:live'. It turns out that
'jmap -histo' isn't reliable for returning proper instance counts
because it does not require a GC to be done. So while 'jmap -histo'
can show decreased counts in bits that have the memory leak fixed,
it can also make it look like the leak is still present.
The two tests need to be modified to abort if the 'jmap -histo:live'
option isn't available. Just to make life difficult, when using
OpenJDK6 bits on Windows, jmap will report a usage message that
says that the '-histo:live' option isn't available when it is.
That issue will be tracked with a different bug.