RFR (S): 8023158 - hotspot/test/gc/7168848/HumongousAlloc.java fails 14 full gcs, expect 0 full gcs

Thomas Schatzl thomas.schatzl at oracle.com
Tue Sep 24 05:49:21 PDT 2013


Hi,

On Mon, 2013-09-23 at 18:13 +0200, Per Liden wrote:
> Hi!
> 
> http://cr.openjdk.java.net/~pliden/8023158/webrev.01/
> 
> https://bugs.openjdk.java.net/browse/JDK-8023158
> 
> Summary: A side effect of fixing JDK-8020155 was that the HumongousAlloc 
> test for G1 broke. This test is designed to make sure than an allocation 
> of an humongous object initiates a mark phase if that allocation pushes 
> the heap size above InitiatingHeapOccupancyPercent. Due to the changes 
> made in JDK-8020155 this test fails since it instead provokes Full GCs. 
> This patch fixes the test. Also renamed the test according to the new 
> naming schema.

- the name of the class does not correspond to the java file name, so it
does not compile (probably a last-minute change).
- one suggestion for the test: please assign the dummy LOB to a global
static instead of a local variable. The compiler can remove that
allocation easily, say when run with -Xcomp and the server compiler. I
tried adding "-Xcomp" and "-server" to the invocation of the
ProcessBuilder and it did not fail though. It may not hurt to do that
anyway.

One personal style comment: not sure if you need to prefix all variables
in the test with "g1", but that's just my opinion. I am fine either way.

Thanks,
Thomas



More information about the hotspot-gc-dev mailing list