Correct, so it depends on how the system properties are used.
In my area, sometimes a system property is always re-read so it's safe
to run related tests in agentvm. Sometimes it's read at initialization
time and a static field is set, if there is no way to reset that static
field, your tests might have to run in othervm mode.
Thanks
Max
On 09/22/2012 05:18 PM, Alan Bateman wrote:
On 22/09/2012 00:12, Weijun Wang wrote:
I think jtreg would automatically reset any changed system property so
othervm is not necessary. Have you tried running it with other tests
in agentvm mode? Thanks Max
There are many areas in the JDK where system properties are read once
during initialization and that fixes the configuration for the lifetime
of the VM. I think java.util.prefs.userRoot is one of them. Tests that
set any of these properties are usually better run in their own VM
because they can cause side effects for tests that run subsequently in
the same VM. Such issues can be difficult to track down, we had quite a
few of these when trying to get the tests to run in samevm or agentvm mode.
So you are right that jtreg could unset these properties before running
the next test but unless it knows whether the property is only read once
or not then I don't think it will be reliable.
-Alan.
