RFR: 8232213: runtime/MemberName/MemberNameLeak.java fails intermittently

[Claes Redestad]

Hi,

this patch resolves an intermittent issue in this test, which became
reproducible on some of our CI hosts after a fix to have String concat
be a bit less wasteful of resources, which showed that we never
triggered the explicit GC in the test but relied on GC activity
happening anyway, which appears to have been likely.

I can reproduce the failure locally by changing the methodCount
to a higher value, e.g., 20000. With the below patch it passes no
matter what I set methodCount to, as long as it's around 1024 or
higher (too low and it'll loop forever).

Bug: https://bugs.openjdk.java.net/browse/JDK-8232213
Patch:

diff -r a08f267427c4 
test/hotspot/jtreg/runtime/MemberName/MemberNameLeak.java
--- a/test/hotspot/jtreg/runtime/MemberName/MemberNameLeak.java	Fri May 
15 18:37:08 2020 +0200
+++ b/test/hotspot/jtreg/runtime/MemberName/MemberNameLeak.java	Fri May 
15 23:45:49 2020 +0200
@@ -103,7 +103,7 @@
              System.gc();  // make mh unused
            }

-          if (after != wb.resolvedMethodItemsCount()) {
+          if (after > wb.resolvedMethodItemsCount() + 50) {
              // Entries have been removed.
              break;
            }

Testing: tier1

Thanks!

/Claes