Bug 172205

Created attachment 121084 [details]
stack trace

This is what appeared in my terminal after a failed session where I tried to
debug but got an error at each debug step.  The variables view never populated
but the console did show the output of my program as I stepped.

This is with rawhide.  gcc 4.0.2-3

Relevant version #s:

java-1.4.2-gcj-compat-1.4.2.0-40jpp_51rh.i386
libgcj-4.0.2-3.i386
libgcj-devel-4.0.2-3.i386
java-1.4.2-gcj-compat-devel-1.4.2.0-40jpp_51rh.i386

eclipse-*-3.1.1-1jpp_6fc.i386

eclipse-cdt-3.0*

This is how I *was* reproducing but it appears to not happen every time:

File->New Project->C->Managed Make C Project->Next
Give a project name (say, "blahblah")
Finish
"Do you want to switch to ..." -> Yes
Expand project blahblah (little arrow on the left-hand side)
Right-click on blahblah -> New -> (c) Source File
Give a name (say, "main.c"), press enter
Enter this program:

#include <stdio.h>

int main(int argc, char **argv) {
	int i = 0;
	for (i = 0; i < 3; ++i) {
		printf("%d\n", i);
	}
	return 0;
}

File->Save
Notice the new entry under project blahblah:  Binaries
Expand Binaries, right-click on "blahblah - [x86le]"
Debug As -> Debug Local C/C++ Application
Hit OK
"Do you want to switch ..." -> Yes
Use Run->Step Over to step over each line in the program and watch the output in
the console
I was getting errors at each step request

Created attachment 121092 [details]
Program to test the cdtparser

This is a program I've been using to test the cdtparser when natively compiled.
All it does it takes a C++ source file as an argument, then prints out the
functions it finds. When it works properly, everything is as it should be, but
when I compiled the cdtparser jar with find-and-aot-compile (which uses the
options '-findirectdispatch -shared -Wl,-Bsymbolic') it produces very strange
output. I will attach a test file I have been using as input shortly

Created attachment 121093 [details]
test input file for CDTParserTest

This is the sample input file that I have been using for CDTParserTest. When
working properly, it should say that function c starts at offset 4 and goes
until offset 5 on line 1. When working incorrectly, it says that function c
starts at character 4, but ends on an extremely large offset (> 20000) .

To get these results I've compiled everything as follows:
gcj -shared -fPIC -fjni -o libcdtparser.jar.so cdtparser.jar
gcj -fPIC -fjni -classpath ./cdtparser.jar -L. -lcdtparser.jar 
	--main=CDTParserTest CDTParserTest.java
LD_LIBRARY_PATH=. ./a.out main.cpp

I was able to reproduce with gcc-4.0.2-3.i386:

wget https://bugzilla.redhat.com/bugzilla/attachment.cgi?id=121092
mv attachment.cgi\?id\=121092 CDTParserTest.java
wget https://bugzilla.redhat.com/bugzilla/attachment.cgi?id=121093
mv attachment.cgi\?id\=121093 main.cpp

(with eclipse-cdt-3.0.1 (rawhide as of 2005-11-16) installed)
Correct behaviour (no BC ABI):
=============================
gcj -shared -fPIC -fjni -o libcdtparser.jar.so \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.1/cdtparser.jar
gcj -fPIC -fjni -classpath \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.1/cdtparser.jar \
  -L. -lcdtparser.jar --main=CDTParserTest CDTParserTest.java
LD_LIBRARY_PATH=. ./a.out main.cpp
Function found
Name: c
Starting at: 0
Ending at: 0
Name starts at: 4
Name ends at: 5

INcorrect behaviour (BC ABI):
=============================
gcj -shared -findirect-dispatch -Wl,-Bsymbolic -fPIC -fjni -o \
  libcdtparser.jar.so \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.1/cdtparser.jar
gcj -fPIC -fjni -classpath \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.1/cdtparser.jar \
  -L. -lcdtparser.jar --main=CDTParserTest CDTParserTest.java
LD_LIBRARY_PATH=. ./a.out main.cpp
Function found
Name: c
Starting at: 17409568
Ending at: 5
Aborted

(with eclipse-cdt-3.0.0 (FC4 or older than 2005-11-16's rawhide) installed)
Correct behaviour (no BC ABI):
=============================
gcj -shared -fPIC -fjni -o libcdtparser.jar.so \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.0/cdtparser.jar
gcj -fPIC -fjni -classpath \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.0/cdtparser.jar \
  -L. -lcdtparser.jar --main=CDTParserTest CDTParserTest.java
LD_LIBRARY_PATH=. ./a.out main.cpp
Function found
Name: c
Starting at: 0
Ending at: 0
Name starts at: 4
Name ends at: 5

INcorrect behaviour (BC ABI):
=============================
gcj -shared -findirect-dispatch -Wl,-Bsymbolic -fPIC -fjni -o \
  libcdtparser.jar.so \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.0/cdtparser.jar
gcj -fPIC -fjni -classpath \
  /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.0/cdtparser.jar \
  -L. -lcdtparser.jar --main=CDTParserTest CDTParserTest.java
LD_LIBRARY_PATH=. ./a.out main.cpp
Function found
Name: c
Starting at: 17409568
Ending at: 5
Aborted

I tried this with the 3.0 CDT RPM installed.

With my FC4 system gcj (4.0.something), it failed when BC-compiled.

With gcc 4.1 and gcc svn head, it worked.

I can confirm this is OK with current rawhide gcc version 4.1.0 20060121 (Red
Hat 4.1.0-0.16)

Target: x86_64

Works for me with 4.0.0-6.

aph@zorro ~]$ gcj -shared -findirect-dispatch -Wl,-Bsymbolic -fPIC -fjni -o \
>   libcdtparser.jar.so \
>   /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.0/cdtparser.jar
[aph@zorro ~]$ gcj -fPIC -fjni -classpath \
>   /usr/share/eclipse/plugins/org.eclipse.cdt.core_3.0.0/cdtparser.jar \
>   -L. -lcdtparser.jar --main=CDTParserTest CDTParserTest.java
[aph@zorro ~]$ LD_LIBRARY_PATH=. ./a.out main.cpp
Function found
Name: c
Starting at: 0
Ending at: 0
Name starts at: 4
Name ends at: 5
[aph@zorro ~]$ gcj -v
Using built-in specs.
Reading specs from /usr/lib/gcc/i386-redhat-linux/4.0.2/libgcj.spec
rename spec lib to liborig
Target: i386-redhat-linux
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
--infodir=/usr/share/info --enable-shared --enable-threads=posix
--enable-checking=release --with-system-zlib --enable-__cxa_atexit
--disable-libunwind-exceptions --enable-libgcj-multifile
--enable-languages=c,c++,objc,java,f95,ada --enable-java-awt=gtk
--with-java-home=/usr/lib/jvm/java-1.4.2-gcj-1.4.2.0/jre --host=i386-redhat-linux
Thread model: posix
gcc version 4.0.2 20051109 (Red Hat 4.0.2-6)

But fails with 4.0.0-8:

[aph@zorro ~]$ LD_LIBRARY_PATH=. ./a.out main.cpp
Function found
Name: c
Starting at: 18582304
Ending at: 5
Aborted

So this is a regression.  Ths differences between these two gcj versions are
fairly small, so this should be easy to find.

OK, found it.  This is http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24251.

The Red Hat 4.0 package needs to be updated with this patch from gcc-4_0-branch.

It is updated from gcc-4_0-branch, RHEL4 U3 contains 20051130 snapshot
in gcc4-java-4.0.2-9.EL4 and above.  You just need to use gcj4 and gij4
instead of gcj and gij.

I'm testing on FC4.  This patch is not in gcc-4.0.0-8.

It is needed.

FC4 has gcc-4.0.2-8.fc4, which is a 20051126 snapshot.
sudo yum update

Confirmed.  

Andrew O, please close.

I have verified that things do indeed work as expected with an updated FC4
system and also with current rawhide.  Closing.