RFR (XXS) 8212932: [TESTBUG] Clean up TestVirtualSpaceNode test

[Thomas Stüfe]

Looks good.

.. Thomas

On Wed, Feb 27, 2019, 19:22 gerard ziemski <gerard.ziemski at oracle.com wrote:

> Hi all,
>
> Please review this small change where we remove
> test/hotspot/gtest/memory/test_virtualSpaceNode.cpp
>
> For the reason as to why we decided to remove it, as opposed to cleaning
> it up (which was under consideration - I fixed some of its issues in
> JDK-8216493), I’m providing here a quote from Thomas Stuefe:
>
> “I feel that since 8198423 its usefulness is greatly diminished, and
> there is a bit of a danger of getting it wrong, messing up global
> structures in metaspace and affecting follow up tests. Metaspace is not
> (anymore) designed to be used on this isolated level.
>
> I really thought long and hard about this. Unfortunately, this test
> cannot be done in an isolated fashion anymore without affecting global
> metaspace - sure you encountered that yourself, hence the terrible
> (sorry :-) ChunkManagerRestorer workaround. And I do not see a better
> way to do this.
>
> All these tests do now is to ensure that a particular notion of chunk
> geometry and allocation behavior is adhered to - stuff which may change
> anyway since it is implementation detail. I do not see any really
> worthwhile tests. And the fact that since 8198423 any allocation from a
> VirtualSpaceNode can cause allocation of padding chunks, which get added
> to the global freelist and hence change global state, makes matters
> quite complicated."
>
> bug: https://bugs.openjdk.java.net/browse/JDK-8212932
> webrev: http://cr.openjdk.java.net/~gziemski/8212932_rev1
> testing: passes "make test-hotspot-gtest” locally and Mach5 tier1,2,3
>
>
> cheers
>