[#RVM-294] JikesRVM fails to build using a Classpath VM

RVM

JikesRVM fails to build using a Classpath VM

Created: 10/Oct/07 05:04 PM Updated: 06/Oct/08 01:47 PM

Component/s:

Runtime: Class Library: GNU Classpath

Affects Version/s:

None

Fix Version/s:

3.0.2

Time Tracking:

Not Specified

File Attachments:	1. Test.java (0.2 kb)
Environment:	ia32-linux

Testcase included:

yes

Description

« Hide

Building JikesRVM with either JamVM or CACAO and GNU Classpath currently fails due to a difference in the VMRuntime classes. JikesRVM's version is public, while the Classpath copy is package-private.

gen-interface:
[java] java.lang.ExceptionInInitializerError
[java] at org.jikesrvm.ia32.VM_OutOfLineMachineCode.generateThreadSwitchInstructions(VM_OutOfLineMachineCode.java:342)
[java] at org.jikesrvm.ia32.VM_OutOfLineMachineCode.init(VM_OutOfLineMachineCode.java:55)
[java] at org.jikesrvm.VM.init(VM.java:2216)
[java] at org.jikesrvm.VM.initForTool(VM.java:107)
[java] at org.jikesrvm.VM.initForTool(VM.java:95)
[java] at org.jikesrvm.tools.header_gen.GenerateInterfaceDeclarations.main(GenerateInterfaceDeclarations.java:162)
[java] Caused by: java.lang.IllegalAccessException: class is not accessible
[java] at org.jikesrvm.runtime.VM_Entrypoints.<clinit>(VM_Entrypoints.java:78)
[java] at org.jikesrvm.ia32.VM_OutOfLineMachineCode.generateThreadSwitchInstructions(VM_OutOfLineMachineCode.java:342)
[java] ...5 more

The line in question is:

public static final VM_Field gcLockField = getField(java.lang.VMRuntime.class, "gcLock", int.class);

I don't know if this is new or not, but we could build earlier. Trying to put this line in a separate file to test causes the compile to fail because VMRuntime is private:

javac -cp target/prototype_x86_64-linux/jksvm.jar:target/prototype_x86_64-linux/rvmrt.jar Test.java
Test.java:7: java.lang.VMRuntime is not public in java.lang; cannot be accessed from outside package
System.out.println(getField(java.lang.VMRuntime.class, "gcLock", int.class));
^
1 error

Using the jars as the bootclasspath works but the test still fails.

java org/jikesrvm/runtime/Test
java.lang.IllegalAccessException: class is not accessible
at org.jikesrvm.runtime.Test.main(Test.java:9)

Description

Building JikesRVM with either JamVM or CACAO and GNU Classpath currently fails due to a difference in the VMRuntime classes. JikesRVM's version is public, while the Classpath copy is package-private. gen-interface: [java] java.lang.ExceptionInInitializerError [java] at org.jikesrvm.ia32.VM_OutOfLineMachineCode.generateThreadSwitchInstructions(VM_OutOfLineMachineCode.java:342) [java] at org.jikesrvm.ia32.VM_OutOfLineMachineCode.init(VM_OutOfLineMachineCode.java:55) [java] at org.jikesrvm.VM.init(VM.java:2216) [java] at org.jikesrvm.VM.initForTool(VM.java:107) [java] at org.jikesrvm.VM.initForTool(VM.java:95) [java] at org.jikesrvm.tools.header_gen.GenerateInterfaceDeclarations.main(GenerateInterfaceDeclarations.java:162) [java] Caused by: java.lang.IllegalAccessException: class is not accessible [java] at org.jikesrvm.runtime.VM_Entrypoints.<clinit>(VM_Entrypoints.java:78) [java] at org.jikesrvm.ia32.VM_OutOfLineMachineCode.generateThreadSwitchInstructions(VM_OutOfLineMachineCode.java:342) [java] ...5 more The line in question is: public static final VM_Field gcLockField = getField(java.lang.VMRuntime.class, "gcLock", int.class); I don't know if this is new or not, but we could build earlier. Trying to put this line in a separate file to test causes the compile to fail because VMRuntime is private: javac -cp target/prototype_x86_64-linux/jksvm.jar:target/prototype_x86_64-linux/rvmrt.jar Test.java Test.java:7: java.lang.VMRuntime is not public in java.lang; cannot be accessed from outside package System.out.println(getField(java.lang.VMRuntime.class, "gcLock", int.class)); ^ 1 error Using the jars as the bootclasspath works but the test still fails. java org/jikesrvm/runtime/Test java.lang.IllegalAccessException: class is not accessible at org.jikesrvm.runtime.Test.main(Test.java:9)

Show »

All

Comments

Work Log

Change History

Sort Order:

Ascending order - Click to sort in descending order

[ Permlink | « Hide ]

Andrew John Hughes - 10/Oct/07 05:14 PM

Filed in Classpath as well: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33731

[ Permlink | « Hide ]

Andrew John Hughes - 09/Nov/07 03:46 AM

Using the string version of getField means this now fails with a non-existant "java.lang.VMRuntime" Error.

[ Permlink | « Hide ]

Ian Rogers - 09/Nov/07 05:44 AM

That's odd.. does the string need to be formatted as "Ljava/lang/VMRuntime;" ?

[ Permlink | « Hide ]

Andrew John Hughes - 03/Dec/07 02:35 PM

Thanks Ian. You were right, the string does need to use the type name. However, this works only up until it hangs building the bootimage:

build-bootimage-writer:
[mkdir] Created dir: /home/gandalf/projects/java/classpath/jikesrvm.jamvm/target/prototype_x86_64-linux/bootimage-writer
[javac] Compiling 7 source files to /home/gandalf/projects/java/classpath/jikesrvm.jamvm/target/prototype_x86_64-linux/bootimage-writer

gen-primordial-list:

build-bootimage:
[echo] Building bootimage. Output redirected to : /home/gandalf/projects/java/classpath/jikesrvm.jamvm/target/prototype_x86_64-linux/BootImageWriterOutput.txt
[echo] MMTk properties = /home/gandalf/projects/java/classpath/jikesrvm.jamvm/build/mmtk/default.properties

[ Permlink | « Hide ]

Ian Rogers - 03/Dec/07 03:25 PM

That's odd, wonder if boot image writing is stressing parts of the VMs not stressed before. In particular the recent change to make the Intel assembler typed means we're hitting a lot of enumeration code now. There are some verbosity flags for the boot image writer. Some can be configured using ant (iirc - you'll need to look at the build.xml to work out what they are) and some others can be configured by toggling them in the boot image writer code itself. The output is off course all redirected into the target/prototype_x86_64-linux/BootImageWriterOutput.txt file so you'll need to look there to see where boot image writing got up to before it crashed. Is there any output in that file currently? Redirecting output the way we do tends to mask seeing errors

Thanks for looking into this.

[ Permlink | « Hide ]

Andrew John Hughes - 10/Dec/07 04:46 PM

I think this is the stalling thread, seems to never wake up:

[java] "pool-1-thread-1" 0xccba90 priority: 5 tid: 0x42804960 id: 6 state: RUNNABLE (6)
[java] at java.lang.VMThread.sleep(Native method)
[java] at java.lang.Thread.sleep(Thread.java:902)
[java] at java.lang.Thread.sleep(Thread.java:866)
[java] at org.jikesrvm.classloader.VM_Array.instantiate(VM_Array.java:481)
[java] at org.jikesrvm.tools.bootImageWriter.BootImageWorker.run(BootImageWorker.java:44)
[java] at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:679)
[java] at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:704)
[java] at java.lang.Thread.run(Thread.java:743)

[ Permlink | « Hide ]

Andrew John Hughes - 11/Dec/07 05:30 PM

Log ends with:

BootImageWriter: instantiating
BootImageWriter: compiling with 1 threads

[ Permlink | « Hide ]

Ian Rogers - 11/Dec/07 06:22 PM

Hi Andrew, the sleep is there to stop a race when building multithreaded iirc. Previously we used to die during booting if a class was being instantiated in parallel to an array trying to be instantiated of that class. It looks like this may be a case where the class is never getting instantiated, in which case it'd be interesting to know what class we were struggling upon. We probably need some extra debug to understand this. Sorry for not having looked into the problem deeply.

[ Permlink | « Hide ]

Ian Rogers - 11/Jul/08 08:25 AM

The gcLock, original cause of this bug, is now held in a different class (java.lang.VMCommonClassLibrarySupport.GCLock) and may have resolved this issue. Andrew could you test?

[ Permlink | « Hide ]

Andrew John Hughes - 23/Jul/08 04:38 AM

Moving to 3.0.1 as I don't have time to try and replicate this right now.

Building with a Classpath VM needs a simulated JDK layout:

bin/java
bin/javah
bin/apt
jre/lib/rt.jar
lib/tools.jar

Note the need for apt, which means tools.jar must be from Sun's langtools and not Classpath. You can then use this layout for JAVA_HOME.

[ Permlink | « Hide ]

David Grove - 06/Oct/08 01:47 PM

Initial triage for 3.0.1 release; suggesting postponing to 3.0.2; move back to 3.0.1 if you disagree.