Test failed: BadInnerTest about krakatau

[Storyyeller]

Are you using a nonstandard JVM? It appears that your JVM is handling invalid innerclass attribute lengths differently than Hotspot does. At any rate, you can ignore the test if you want.

For what it's worth, here's mine.

java version "1.7.0_21"
Java(TM) SE Runtime Environment (build 1.7.0_21-b11)
Java HotSpot(TM) 64-Bit Server VM (build 23.21-b01, mixed mode)

[Storyyeller]

The relevant code is in http://hg.openjdk.java.net/jdk7/hotspot-rt/hotspot/file/5dce25362b8a/src/share/vm/classfile/classFileParser.cpp. Does your build include this?

[daveloyall]

I'm using the standard JVM. Well, the standard JVM for my family of operating systems. http://openjdk.java.net/projects/icedtea/

I used to install Sun Java on my GNU/Linux systems, but starting with openjdk-7, I was promised that openjdk-7 would BE the standard.

...Hm, yes, that still seems to be the case. According to http://openjdk.java.net/faq/, Oracle Java is based on OpenJDK, not the other way around. :)

That being said, I'll certainly see what JVMs are available for my Debian machine and see which ones pass or fail this test.

Hm, java.lang.ClassFormatError: Truncated class file seems like a reasonable response to the .innerlength 0xFFFFFFFF thing, right?

Cheers,

[Storyyeller]

BadInnerTest and OldVersionTest are effectively testing undocumented corner cases in Hotspot. You can ignore them if you wish.

I'm surprised yours is behaving differently though since the code is part of the OpenJDK code base.

[daveloyall]

I have learned that I can use different VMs by employing command line arguments when invoking java.

Abbreviated java -help output:

-server   to select the "server" VM
-zero     to select the "zero" VM
-jamvm    to select the "jamvm" VM
-avian    to select the "avian" VM
          The default VM is server,
          because you are running on a server-class machine.

(server is apparently IcedTea, though I have also read that my default VM is Hotspot and I'm not sure about the relationship between them in OpenJDK. It's not even clear to me that IcedTea is a VM and not merely a build environment for OpenJDK.)

However, it's not immediately apparent to me whether I need to modify runtests.py or tests/__init__.py in order to add a command line argument to the java invocation.

[Storyyeller]

You need to modify runtests.py line 27. For consistency, you'd probably want to change line 74 too.

init.py only has the name and inputs for each test.

[daveloyall]

So, I modified runtests.py to take a command line argument --vm= and here are my results, just FYI. The results for avian are apparently a local problem.

$ ./runtests.py --vm=server |tail -22:

VM was: server. Test results:
  ArgumentTypes: Pass
  ArrayTest: Pass
  BadInnerTest: Fail
  BoolizeTest: Pass
  ControlFlow: Pass
  DoubleEdge: Pass
  DuplicateInit: Pass
  NullInference: Pass
  OddsAndEnds: Pass
  OldVersionTest: Pass
  SkipJSR: Pass
  StaticInitializer: Pass
  Switch: Pass
  Synchronized: Pass
  TryCatchTest: Pass
  UnicodeTest: Pass
  WhileLoops: Pass
  floattest: Pass
  splitnew: Pass
18/19 tests passed

$ ./runtests.py --vm=zero |tail -22:

VM was: zero. Test results:
  ArgumentTypes: Pass
  ArrayTest: Fail
  BadInnerTest: Fail
  BoolizeTest: Pass
  ControlFlow: Pass
  DoubleEdge: Pass
  DuplicateInit: Pass
  NullInference: Pass
  OddsAndEnds: Pass
  OldVersionTest: Pass
  SkipJSR: Pass
  StaticInitializer: Pass
  Switch: Fail
  Synchronized: Pass
  TryCatchTest: Pass
  UnicodeTest: Pass
  WhileLoops: Pass
  floattest: Pass
  splitnew: Pass
16/19 tests passed

$ ./runtests.py --vm=jamvm |tail -22:

VM was: jamvm. Test results:
  ArgumentTypes: Pass
  ArrayTest: Fail
  BadInnerTest: Fail
  BoolizeTest: Pass
  ControlFlow: Fail
  DoubleEdge: Pass
  DuplicateInit: Pass
  NullInference: Pass
  OddsAndEnds: Pass
  OldVersionTest: Pass
  SkipJSR: Pass
  StaticInitializer: Pass
  Switch: Pass
  Synchronized: Pass
  TryCatchTest: Pass
  UnicodeTest: Pass
  WhileLoops: Pass
  floattest: Pass
  splitnew: Pass
16/19 tests passed

$ ./runtests.py --vm=avian |tail -22:

VM was: avian. Test results:
  ArgumentTypes: Fail
  ArrayTest: Fail
  BadInnerTest: Fail
  BoolizeTest: Fail
  ControlFlow: Fail
  DoubleEdge: Fail
  DuplicateInit: Fail
  NullInference: Fail
  OddsAndEnds: Fail
  OldVersionTest: Fail
  SkipJSR: Fail
  StaticInitializer: Fail
  Switch: Fail
  Synchronized: Fail
  TryCatchTest: Fail
  UnicodeTest: Fail
  WhileLoops: Fail
  floattest: Fail
  splitnew: Fail
0/19 tests passed

[Storyyeller]

That's weird. The pure Java tests (ArrayTest and Switch) shouldn't fail regardless of which JVM is being used. Did you record what the error messages were? The only thing I can think of is that it might be printing out stack traces differently after recompilation despite compiling without debugging info.

P.S. Did you remember to regenerate the expected outputs (tests/*.test) in between tests? If you switch VMs without changing the expected output you'll get false failures.

[Storyyeller]

I went ahead and disabled the implementation specific tests (BadInnerTest and OldVersionTest).

[Storyyeller]

FYI, I tested this again a couple weeks ago and it no longer works even in Java 8. It seems that this undocumented feature is only present up to Java 7.