code offsets and add a debug flag to StackTrace that enablings printing of
an internal stack trace with bytecode indexes and machine code offsets.

Exposing the information about machine code offsets and bytecode indexes
is very useful for the debugging of compiler problems.

verbose signal handling to mostly mimic the behaviour of -X:verbose
before the bootloader changes.

Before the bootloader changes (RVM-1105), the -X:verbose command line
argument only affected signal handling. After the changes, setting
-X:verbose also prints out information about Jikes RVM syscalls. The
addition of the -X:verboseSignalHandling option allows users to get an
approximation of the old behaviour of -X:verbose . There are some
differences in the print outs but the information should be mostly the
same.

The current -X:verbose implies -X:verboseSignalHandling.

RVM-1129 : Exclude regression test for RVM-966 from pre-commit until it passes reliably on all platforms.

Modify build.xml to automatically update Eclipse project directories when generated files are updated.

The target was used to produce documentation for http://www.jdocs.com
which is now defunct.

PPC assembler registers (original changes for both issues by Ian Rogers
for MRP).

This commit contains changes from the following MRP commits:
- https://github.com/codehaus/mrp/commit/eff1aa489facbf3756bad431c4af4f71497e9c92
(removal of ArchitectureSpecific)
- https://github.com/codehaus/mrp/commit/4846ffb0fca7c2647586ceb5c8bff374909aa303
(BURS bug fixes and debugging improvements)
- https://github.com/codehaus/mrp/commit/ad42eb0bfd76bb0b276e294e1bc8d922344e4760
(update of Eclipse project templates for removal of
ArchitectureSpecific)
- https://github.com/codehaus/mrp/commit/d0561a9726ed285097b77226809254578592676
(remove generated.arch.java from the build)
- https://github.com/codehaus/mrp/commit/155f1c20571b56e8598ba82d4fdb9ee4cc83867c
(bug fixes for PowerPC builds)
- https://github.com/codehaus/mrp/commit/31d11685042d1bb38ed52623f2155f7819ad9484
(restructure logic in Instruction.isTwoWayBranch())

The following changes have been made with respect to MRP:
- Whitespace cleanup in the PPC assembler (removed some empty lines in
emit* methods)
- added the enums from org.jikesrvm.ppc.RegisterConstants to the PPC
primordials to fix NPE when running class initializers during VM booting
- in contrast to MRP, the int constants in
org.jikesrvm.compilers.opt.regalloc.ppc.PhysicalRegisterConstants have
not been converted to GPRs / FPRs
- the JNI code differs from MRP because the Jikes RVM is still using the
old PPC JNI code
- JavaDoc errors have been fixed

Split org.jikesrvm.SizeConstants into org.jikesrvm.runtime.JavaSizeConstants and org.jikesrvm.runtime.UnboxedSizeConstants.

package org.jikesrvm.runtime (original changes by Ian Rogers for MRP).

In MRP, these classes were moved during the removal of
ArchitectureSpecific in
https://github.com/codehaus/mrp/commit/eff1aa489facbf3756bad431c4af4f71497e9c92
.

package org.jikesrvm to the package org.jikesrvm.util (original changes
by Ian Rogers for MRP).

In MRP, these classes were moved during the removal of
ArchitectureSpecific in
https://github.com/codehaus/mrp/commit/eff1aa489facbf3756bad431c4af4f71497e9c92
.

This operator hasn't been used since commit 3836db4e from 2007.

RVM-1063 : Remove BURS rule that I forgot to remove when merging MRP commit be94f3408ac0f07b396f9f040f51f4576085ccd2 .

According to the documentation in org.jikesrvm.runtime.ExitStatus, an
exit code of -1 will get mapped to 0 (i.e. success) by Cygwin. That's
obviously not desirable so all uses of -1 in Jikes RVM have been
replaced with EXIT_STATUS_MISC_TROUBLE. In MMTk +2 has been used instead
of -1 (EXIT_STATUS_PRINTED_HELP_MESSAGE in Jikes RVM already uses +1).

This commit also updates the runCompareTest Ant macro with support for setting an expected exit code. In addition, when a test fails with an unexpected exit code the explanation will now show which code was expected and which was returned.

Initialize architecture-dependent fields in Entrypoints and ArchEntrypoints with a static initializer.

This will make the code more readable when new architectures are added.

it terminates because of an uncaught Throwable in an application thread.

implemented in superinterfaces of a class.

This implements 3. from section 5.4.3.3 in the JVM specification.

The test case was provided by Suriya Subramanian.

MemoryOperands when doing register allocation on x64.

In the current code, operations that should be 32-bit according to the
types of the registers (i.e. TypeReference.Int) can be changed to 64-bit
due to the introduction of spill locations (which are always 64-bit
wide).

In Da Feng's patch set for x64, GenericStackManager saves the type of
the currently processed operand and StackManager uses 32-bit spill
locations instead of 64-bit spill locations when the typing requires it.
This makes the whole operation 32-bit. I think that this approach is
incorrect in the general case because it is not guaranteed that the
spill location's address fits in 32 bits.

This commit changes the StackManager to make sure that the values of
int-typed registers are in registers when the register's type is int.
This prevents potential problems with addressing.

x64 Linux when running with -X:verbose.

Since the removal of libvm in cf01173d ,
any attempt to print the registers in the hardwareTrapHandler has
segfaulted on x64.

It seems that the Jikes RVM signal handler takes more space than the
standard size (SIGSTKSZ) provides. On my Ubuntu x64 system, the Jikes
RVM signal handler took more than 10000 bytes. Unfortunately, SIGSTKSZ
for glibc on Linux x64 is only 8092 bytes.

This commit uses 2 * SIGSTKSZ for all platforms. It's likely that a few
platforms require less space but our current regression machine setup
doesn't cover all platforms that we want to support.

(original changes by Ian Rogers from MRP; thanks to Xi Yang for giving
me the idea of pulling the changes).

The original code for this was commented out in MRP. However, I think
that it can be useful for debugging.

Code that's not executed will bitrot. Performance is not that important
in ExtremeAssertions builds, so it's fine to enable the code there.

See
https://github.com/codehaus/mrp/commit/8780f9f0896aef463c179181587e0a9ef95943f9
for the original MRP commit.

IA32_OLDMASK to be consistent with the MRP implementation.

The original MRP commit was
https://github.com/codehaus/mrp/commit/9db39bf2395ede289957dc3ebee1e1b8e1f951d4
. That MRP commit was merged in Jikes RVM commit
2200f38a . 

This fixes compilation issues under OSX and likely other targets.

streams in order to improve clarity.

reduce duplication.

This problem was found with FindBugs.

RVM-1063 : Apply changes to StackManager from MRP that I forgot when merging the original MRP commit in e34d12d6 .

See https://github.com/codehaus/mrp/commit/08b031163d63875a14ca8e7f29883284451d0cb8 for the original MRP commit.

that represent plain longs (original changes by Da Feng).

It is necessary to track whether longs come from references in order to
make sure that the space on the stack is managed properly. Longs take 2
stack slots but references only take one.

This change is part of Da Feng's x64 patch set.

These changes are part of Da Feng's x64 patch set.