[vmkit-commits] Breakage on trunk?

[Will Dietz]

Hi,

I'm not sure what change introduced this, but as of r144456 I cannot
do a from-clean build on any of {32,64}x{classpath,openjdk}.  All
failures happen while building mmtk.

I'm building against LLVM r144492 for what it's worth.

The 32bit variants fail in the very first call to j3ResolveStaticStub
(which is the 1st or 2nd java instruction executed in both runtime
ports), with the following assertion failure:

vmjc: /home/will/vmkit-svn/lib/J3/VMCore/JavaRuntimeJIT.cpp:436: void
*j3ResolveStaticStub(): Assertion `FI->Metadata != __null && "Wrong
stack trace"' failed.

The 64bit variants fail later in the mmtk build process, with the following:

Thread 0x110000000 received a SIGSEGV: either the VM code or an external
native method is bogus. Aborting...

I believe this failure point is the same in both--its when the first
NullPointerException is thrown.

Interestingly, the 64bit builds *do* work when changing
"SupportsHardwareNullCheck" to 'false' in System.h.  Once MMTk is
built, re-enabling the hardware null checks seems to work great
(dacapo, etc).

(As an aside, the hardware null checks result in significantly faster
execution times! ~22s instead of 33s on dacapo's antlr, for example!).

Anyway, does anyone else see these failures or have any ideas towards
their cause?

Thanks!

~Will