Sophie

Sophie

distrib > Mandriva > 2007.0 > i586 > media > contrib-release > by-pkgid > 4c9f17ec5da473f7fb52041bb9197c5a > files > 77

kaffe-devel-1.1.8-0.20060723.1mdv2007.0.i586.rpm

Debugging Kaffe
---------------

This document provides some pointers for debugging the Kaffe VM. (Not
necessarily for debugging Java apps running on Kaffe, though you can
abuse these techniques to that end.)

building Kaffe for debugging
============================

There are a few things to consider when building Kaffe to make your 
debugging experience enjoyable.  There are debugging macros within 
Kaffe that will print out predefined messages, you control which 
messages you are interested in by using -vmdebug (see below).  If
you are planning on using GDB (either alone or with some GUI) you
will want to disable GCC optimizations.  The optimizations that 
GCC perform may change the execution order of some instructions 
and make your debugging session very frustrating!  You will not 
know which like will execute next and your executing line will 
jump around.  The easiest way disable GCC optimizations when 
building Kaffe is to edit the config.frag for your system 
(config/<CPU>/<OS>/config.frag).  Add the -O0 flag to CFLAGS, 
this tells GCC to use zero (0) optimizations. 
        CFLAGS="$CFLAGS -O0"

The other thing that you will want to do is to use
static linking when building Kaffe.  There are issue with some 
dynamic loaders and GDB may not be able to find your source files
when you step into a function that is defined in another directory
or library.  When configuring Kaffe you can pass the static flags
     ./configure --with-staticlib --with-staticbin --with-staticvm

       
-vmdebug
========

Configure Kaffe with the '--enable-debug' option. This turns on a
bunch of debugging infrastructure in Kaffe. Most of the
infrastructure is for tracing, but there are some sanity checks that
can be turned on (e.g., 'GCDIAG'), and other flags and settings (e.g.,
'NOGC').

Run kaffe with '-vmdebug list' to get a list of supported VM debugging
options. Pass the options you're interested in as a comma-separated
list of names. For example, you can use it like so:
        kaffe -vmdebug INIT,VMTHREAD,GCSYSALLOC foo

This will print some stats about initialization, VM-level threading
events, and system-level memory allocations. Note that some of the
options are very, very verbose (e.g., 'GCALLOC' or 'ALL'.)

If you want to add more run-time tracing code, look at
kaffe/kaffevm/debug.h. Grep for the DBG() macros to see how they're
used.

gdb
===

Kaffe is started by a script that determines the BOOTCLASSPATH, 
CLASSPATH and LD_LIBRARY_PATH that is needed for Kaffe to run.
Since the Kaffe executable is not started directly by you, trying
'dbg kaffe' will not work.  You will need to tell the startup script
which debugger you want to use.  This is as easy as setting the 
environment variable KAFFE_DEBUG to your debugger of choice.
Currently Kaffe can be debugged use: gdb, ddd, emacs and cgdb.
        KAFFE_DEBUG=gdb
Once you are done debugging or nolonger want Kaffe to start up 
within a debugger you can simply unset KAFFE_DEBUG.        

There are some gdb macros in developers/gdbinit that are useful for
looking at Kaffe's GC, thread, and object structures (in particular,
look for 'pobject', 'livethreadsbt'). Some of the macros are out of
sync wrt to the actual structures, so you may have to update
them. Comments in the file should explain how to use them.

Also, look at FAQ.xdebugging for a mechanism for including Java
symbols in backtraces in GDB. If you're going to debug Kaffe on a
regular basis, you should always include the xdebugging support.


Internal Tests
==============

Internal tests are kept in the test/internal directory and are used to
do very basic checking of the VM.  If you are doing a port, you will
want to start here since the tests are less demanding than the full
regression tests.  In general, the tests are written in C, link
directly to the VM libraries, check results internally, and exit when
they first encounter a problem.

Currently, there is only one test, jitBasic, which does basic testing
of the jitter.  For example, the tests range from simple checking of
functions returning constants to creating objects and accessing their
fields.  These tests are ideal for those doing a port of the jitter,
and were in fact used to assist the development of the PowerPC port.
Note that the test relies on Java classes, Jikes is required to do the
compilation, we do not currently distribute precompiled versions.

jitBasic environment variables:

	TEST_CLASSES - The classes to test against.

To run it manually, you need to do something like the following 
in sh/bash:

  . ../../BUILD_ENVIRONMENT
  export BOOTCLASSPATH=${BOOTCLASSPATH}:.
  export TEST_CLASSES="ConstMethods.class"
  ./jitBasic

It "should" complain if it has any problems and it should print out 
a bunch of debugging info as it jits the methods.  If its completely 
silent, something is probably wrong.

Regression Tests
================

Regression tests are kept in test/regression/.  Regression tests are
just simple stand-alone .java programs. They have special comments at
the end that influence how the tester treats their output.  Generally,
the "correct" output is included in the comment, and the test script
simply compares the output generated.  

To add a new test, just add it to the list of tests in Makefile.am,
regenerate the Makefile.in, re-configure, and run 'make check'.  (Ugh,
the need to re-generate Makefile.in is awkward and lame.)

To run a single test (or a subset of tests) set the TEST variable,
like so:
	cd test/regression
	make check TESTS="ThreadInterrupt.java IndexTest.java"

Note, you can test new tests without doing the whole Makefile.am thing
by just including the test in the test/regression directory and then
setting the TESTS environment variable to that test.  For example:

	$ vi test/regression/OneOffHackTest.java
	$ make check TESTS="OneOffHackTest.java"


Special Comments in Regression Tests
====================================

Regression tests have special comments that tell the test driver
script what to do with them. The special comments and their
effects are:

// Sources: SOURCES
compile also these sources with current test case.

// javac flags: FLAGS
add some javac flags.  For exemple -nowarn remove warning lines
and therefore don't let's think that javac fail.

// Skip Run
Compile only

// java args: CLASSNAME [OPTIONS]
Replace current classname by this classname and options.

All lines enclose inside line
/* Expected Output:
and
*/
are compared to test-case's output.

// Sort Output
Sort output before compare it to Expected Output.

Other Stuff
===========

Debugging Performance: Look at FAQ.timing and FAQ.xprofiling for
details on getting performance information out of the VM.