GCC 3.2 suddenly fails every java test, CLASSPATH wrong

GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Kaveh R. Ghazi]

All of a sudden, every test in the java testsuite is failing in GCC-3.2:
http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00486.html

It's not just me:
http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00462.html

In the logfile, all of the failures look like this:

 > Can't find default package `java.lang'. Check the CLASSPATH
 > 	environment variable and the access to the archives
 > 1 error

The code in /libjava/testsuite/lib/libjava.exp is not setting the
CLASSPATH correctly:

 >     # Determine the version so we can find the libgcj jar file.
 >     set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
 >     regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
 >     verbose "jar file is libgcj-$gcc_version.jar"
 > 
 >     [...]
 > 
 >     # Set the CLASSPATH environment variable
 >     verbose "CLASSPATH is
 >     .:$srcdir/$subdir:$objdir:$objdir/../libgcj-$gcc_version.jar"
 >     global env
 >     set env(CLASSPATH)
 >     ".:$srcdir/$subdir:$objdir:$objdir/../libgcj-$gcc_version.jar"

I think the regexp expects the date and "(prerelease)" strings which
normally appear, and when the version simply becomes "3.2" for the
actual release it gets confused.  I don't know this stuff well enough
to suggest a fix that works for both prerelease and actual release.

Looking back through the testresult archives, I see the same thing
happened for 3.1.1 and then magically fixed itself when the version
string returned to normal.  However it didn't fail for everybody.

Perhaps those who installed java already would get the jar file from
the system installation area when it couldn't find the local one
supposedly being tested.  That's just a guess.

		--Kaveh
--
Kaveh R. Ghazi			Director of Systems Architecture
ghazi@caip.rutgers.edu		Qwest Solutions

Re: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[H. J. Lu]

On Thu, Aug 15, 2002 at 11:01:29AM -0400, Kaveh R. Ghazi wrote:
> All of a sudden, every test in the java testsuite is failing in GCC-3.2:
> http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00486.html
> 
> It's not just me:
> http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00462.html
> 
> In the logfile, all of the failures look like this:
> 
>  > Can't find default package `java.lang'. Check the CLASSPATH
>  > 	environment variable and the access to the archives
>  > 1 error
> 
> The code in /libjava/testsuite/lib/libjava.exp is not setting the
> CLASSPATH correctly:
> 
>  >     # Determine the version so we can find the libgcj jar file.
>  >     set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
>  >     regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
>  >     verbose "jar file is libgcj-$gcc_version.jar"
>  > 
>  >     [...]
>  > 
>  >     # Set the CLASSPATH environment variable
>  >     verbose "CLASSPATH is
>  >     .:$srcdir/$subdir:$objdir:$objdir/../libgcj-$gcc_version.jar"
>  >     global env
>  >     set env(CLASSPATH)
>  >     ".:$srcdir/$subdir:$objdir:$objdir/../libgcj-$gcc_version.jar"
> 
> I think the regexp expects the date and "(prerelease)" strings which
> normally appear, and when the version simply becomes "3.2" for the
> actual release it gets confused.  I don't know this stuff well enough
> to suggest a fix that works for both prerelease and actual release.
> 

I was wondering the same thing. This works

http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00457.html

because I have gcc 3.2 installed on that machine. All my other machines
failed. I thought it might be my new binutils :-). It has to be fixed.


H.J.

Re: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Mark Wielaard]

Hi,

On Thu, 2002-08-15 at 17:05, H. J. Lu wrote:
> On Thu, Aug 15, 2002 at 11:01:29AM -0400, Kaveh R. Ghazi wrote:
> > All of a sudden, every test in the java testsuite is failing in GCC-3.2:
> > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00486.html
> > 
> > It's not just me:
> > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00462.html
> > 
> > In the logfile, all of the failures look like this:
> > 
> >  > Can't find default package `java.lang'. Check the CLASSPATH
> >  > 	environment variable and the access to the archives
> >  > 1 error
> > 
> > [...]
> I was wondering the same thing. This works
> 
> http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00457.html
> 
> because I have gcc 3.2 installed on that machine. All my other machines
> failed. I thought it might be my new binutils :-). It has to be fixed.

I just tested the official gcc-3.2.tar.gz release from ftp.gnu.org and
it works for me.

=== libjava Summary ===

# of expected passes		2075
# of expected failures		18
# of untested testcases		14

See: http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00489.html

This was on a i686-pc-linux-gnu machine running:

Debian GNU/Linux testing/unstable
Linux 2.4.18
binutils Version: 2.12.90.0.15-1
glibc Version: 2.2.5-13

With the following configure flags:
--enable-clocale=gnu --enable-shared --enable-threads=posix
--enable-languages=c++,java --prefix=/usr/local/gcc32

I know that at least for the CVS HEAD (3.3) you will need the above
binutils on powerpc-unknown-linux-gnu and maybe other architectures.

One thing that I always do before running 'make check -k' is a 'make
install' because in the past the tests sometimes picked up the wrong
libgcj.jar to test against. Maybe that is still needed. If you haven't
done a 'make install' yet could you do that and then run make check in
the libjava subdir? Otherwise try updating your binutils to 2.12.90.0.15
or higher and rebuild everything with that.

Thanks,

Mark

PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[H. J. Lu]

On Thu, Aug 15, 2002 at 08:52:45PM +0200, Mark Wielaard wrote:
> Hi,
> 
> On Thu, 2002-08-15 at 17:05, H. J. Lu wrote:
> > On Thu, Aug 15, 2002 at 11:01:29AM -0400, Kaveh R. Ghazi wrote:
> > > All of a sudden, every test in the java testsuite is failing in GCC-3.2:
> > > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00486.html
> > > 
> > > It's not just me:
> > > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00462.html
> > > 
> > > In the logfile, all of the failures look like this:
> > > 
> > >  > Can't find default package `java.lang'. Check the CLASSPATH
> > >  > 	environment variable and the access to the archives
> > >  > 1 error
> > > 
> > > [...]
> > I was wondering the same thing. This works
> > 
> > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00457.html
> > 
> > because I have gcc 3.2 installed on that machine. All my other machines
> > failed. I thought it might be my new binutils :-). It has to be fixed.
> 
> I just tested the official gcc-3.2.tar.gz release from ftp.gnu.org and
> it works for me.
> 
> === libjava Summary ===
> 
> # of expected passes		2075
> # of expected failures		18
> # of untested testcases		14
> 
> See: http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00489.html
> 
> This was on a i686-pc-linux-gnu machine running:
> 
> Debian GNU/Linux testing/unstable
> Linux 2.4.18
> binutils Version: 2.12.90.0.15-1
> glibc Version: 2.2.5-13
> 
> With the following configure flags:
> --enable-clocale=gnu --enable-shared --enable-threads=posix
> --enable-languages=c++,java --prefix=/usr/local/gcc32
> 
> I know that at least for the CVS HEAD (3.3) you will need the above
> binutils on powerpc-unknown-linux-gnu and maybe other architectures.
> 
> One thing that I always do before running 'make check -k' is a 'make
> install' because in the past the tests sometimes picked up the wrong
> libgcj.jar to test against. Maybe that is still needed. If you haven't
> done a 'make install' yet could you do that and then run make check in
> the libjava subdir? Otherwise try updating your binutils to 2.12.90.0.15
> or higher and rebuild everything with that.

"make install" shouldn't be required. It is a bug. This one works for
me.


H.J.

Re: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Kaveh R. Ghazi]

 > From: Mark Wielaard <mark@klomp.org>
 > 
 > One thing that I always do before running 'make check -k' is a 'make
 > install' because in the past the tests sometimes picked up the wrong
 > libgcj.jar to test against. Maybe that is still needed. If you haven't
 > done a 'make install' yet could you do that and then run make check in
 > the libjava subdir? Otherwise try updating your binutils to
 > 2.12.90.0.15 or higher and rebuild everything with that.
 > Thanks,
 > Mark

I thought I clearly indicated in my initial report that having a prior
(or current) version of the jar file installed probably made the
warning go away.  That's an explanation why the problem only occurs
for some people and not other, its not the solution.

I agree with HJ, this needs to be fixed.  It's a perception of quality
issue, people will erroneously think gcj is broken when it's in fact a
bug in the testsuite and they'll file spurious bug reports which waste
our time.

		--Kaveh
--
Kaveh R. Ghazi			Director of Systems Architecture
ghazi@caip.rutgers.edu		Qwest Solutions

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Mark Wielaard]

Hi,

On Thu, 2002-08-15 at 21:13, H. J. Lu wrote:
> "make install" shouldn't be required. It is a bug. This one works for
> me.

It would be very nice if this was always true. That would save quite
some time and disk space on my test machines. I have CCed java-patches
to increase the chance that someone that can approve it sees it. I
tested it on the head (3.3) with a small change (see below inside the
patch) and it works nicely. Thanks.

Sorry that I immediately assumed that the real problem was probably an
old version of binutils. I had clearly not read the problem reports
carefully enough (and I should have recognized your name before
suggesting it to you... sorry again).

Cheers,

Mark

> 2002-08-15  H.J. Lu <hjl@gnu.org>
> 
> 	* lib/libjava.exp (gcc_version): Removed.
> 	(libgcj_jar): New.
> 	(libjava_init): Set libgcj_jar with find.
> 	(libjava_arguments): Use libgcj_jar.
> 
> --- libjava/testsuite/lib/libjava.exp.release	Sun May 19 21:51:07 2002
> +++ libjava/testsuite/lib/libjava.exp	Thu Aug 15 12:09:04 2002
> @@ -112,7 +112,7 @@ proc libjava_init { args } {
>      global TOOL_EXECUTABLE
>      global original_ld_library_path
>      global env objdir
> -    global env gcc_version
> +    global env libgcj_jar
>      global tool_root_dir
>      global libjava_libgcc_s_path
>  
> @@ -130,10 +130,9 @@ proc libjava_init { args } {
>  	}
>      }
>  
> -    # Determine the version so we can find the libgcj jar file.
> -    set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
> -    regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
> -    verbose "jar file is libgcj-$gcc_version.jar"
> +    # Find the libgcj jar file.
> +    set libgcj_jar [eval exec "find $objdir/../ -name \"libgcj-*.jar\" 2>@ stdout"]
> +    verbose "jar file is $libgcj_jar"

It might be a little bit more efficient/readable to use glob here:
    set libgcj_jar [glob $objdir/../libgcj-*.jar]

>      # FIXME: This finds libgcj.spec for the default multilib.
>      # If thread models differ between multilibs, this has to be moved
> @@ -271,7 +270,7 @@ proc libjava_arguments {{mode compile}} 
>      global runtests
>      global env
>      global tool_root_dir
> -    global gcc_version
> +    global libgcj_jar
>      global libjava_libgcc_s_path
>  
>      if [info exists LIBJAVA] {
> @@ -337,9 +336,9 @@ proc libjava_arguments {{mode compile}} 
>      verbose "LD_LIBRARY_PATH = $env(LD_LIBRARY_PATH)"
>  
>      # Set the CLASSPATH environment variable
> -    verbose "CLASSPATH is .:$srcdir/$subdir:$objdir:$objdir/../libgcj-$gcc_version.jar"
> +    verbose "CLASSPATH is .:$srcdir/$subdir:$objdir:$libgcj_jar"
>      global env
> -    set env(CLASSPATH) ".:$srcdir/$subdir:$objdir:$objdir/../libgcj-$gcc_version.jar"
> +    set env(CLASSPATH) ".:$srcdir/$subdir:$objdir:$libgcj_jar"
>  
>      if {$mode == "link"} {
>  	global wrapper_file wrap_compile_flags;

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Guillermo Ballester Valor]

Hi, 

Definitely something is buggy, if it is necessary make install before check it 
has to be advertised clearly. I made check after bootstrap (with success) but 
before install.

See my summary for java test (Suse 8.0, glib 2.2.5, linux 2.4.18) :

Compiler version: 3.2
Platform: i686-pc-linux-gnu
configure flags: --host=i686-pc-linux-gnu --prefix=/usr/local/gcc-3.2 
--program-suffix=-3.2 --enable-threads=posix


                === libjava Summary ===

# of expected passes            80
# of unexpected failures        697
# of expected failures          8
# of untested testcases         1270



On Thursday 15 August 2002 21:13, H. J. Lu wrote:
> On Thu, Aug 15, 2002 at 08:52:45PM +0200, Mark Wielaard wrote:
> > Hi,
> >
> > On Thu, 2002-08-15 at 17:05, H. J. Lu wrote:
> > > On Thu, Aug 15, 2002 at 11:01:29AM -0400, Kaveh R. Ghazi wrote:
> > > > All of a sudden, every test in the java testsuite is failing in
> > > > GCC-3.2: http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00486.html
> > > >
> > > > It's not just me:
> > > > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00462.html
> > > >
> > > > In the logfile, all of the failures look like this:
> > > >  > Can't find default package `java.lang'. Check the CLASSPATH
> > > >  > 	environment variable and the access to the archives
> > > >  > 1 error
> > > >
> > > > [...]
> > >
> > > I was wondering the same thing. This works
> > >
> > > http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00457.html
> > >
> > > because I have gcc 3.2 installed on that machine. All my other machines
> > > failed. I thought it might be my new binutils :-). It has to be fixed.
> >
> > I just tested the official gcc-3.2.tar.gz release from ftp.gnu.org and
> > it works for me.
> >
> > === libjava Summary ===
> >
> > # of expected passes		2075
> > # of expected failures		18
> > # of untested testcases		14
> >
> > See: http://gcc.gnu.org/ml/gcc-testresults/2002-08/msg00489.html
> >
> > This was on a i686-pc-linux-gnu machine running:
> >
> > Debian GNU/Linux testing/unstable
> > Linux 2.4.18
> > binutils Version: 2.12.90.0.15-1
> > glibc Version: 2.2.5-13
> >
> > With the following configure flags:
> > --enable-clocale=gnu --enable-shared --enable-threads=posix
> > --enable-languages=c++,java --prefix=/usr/local/gcc32
> >
> > I know that at least for the CVS HEAD (3.3) you will need the above
> > binutils on powerpc-unknown-linux-gnu and maybe other architectures.
> >
> > One thing that I always do before running 'make check -k' is a 'make
> > install' because in the past the tests sometimes picked up the wrong
> > libgcj.jar to test against. Maybe that is still needed. If you haven't
> > done a 'make install' yet could you do that and then run make check in
> > the libjava subdir? Otherwise try updating your binutils to 2.12.90.0.15
> > or higher and rebuild everything with that.
>
> "make install" shouldn't be required. It is a bug. This one works for
> me.
>
>
> H.J.

-- 
Guillermo Ballester Valor
gbv@oxixares.com
Ogijares, Granada  SPAIN
 

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Tom Tromey]

>>>>> "H.J." == H J Lu <hjl@lucon.org> writes:

H.J.> "make install" shouldn't be required. It is a bug.

I agree.

-    # Determine the version so we can find the libgcj jar file.
-    set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
-    regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
-    verbose "jar file is libgcj-$gcc_version.jar"
+    # Find the libgcj jar file.
+    set libgcj_jar [eval exec "find $objdir/../ -name \"libgcj-*.jar\" 2>@ stdout"]
+    verbose "jar file is $libgcj_jar"

I don't really like this hunk.  I'd rather we not use `find' in the
test suite.

What is the problem here?  Why doesn't the existing code work?

If it can't be made to work (I'd like to understand why first), let's
use the Tcl glob command instead of `exec find'.

Tom

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[H. J. Lu]

On Fri, Aug 16, 2002 at 11:11:39AM -0600, Tom Tromey wrote:
> >>>>> "H.J." == H J Lu <hjl@lucon.org> writes:
> 
> H.J.> "make install" shouldn't be required. It is a bug.
> 
> I agree.
> 
> -    # Determine the version so we can find the libgcj jar file.
> -    set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
> -    regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
> -    verbose "jar file is libgcj-$gcc_version.jar"
> +    # Find the libgcj jar file.
> +    set libgcj_jar [eval exec "find $objdir/../ -name \"libgcj-*.jar\" 2>@ stdout"]
> +    verbose "jar file is $libgcj_jar"
> 
> I don't really like this hunk.  I'd rather we not use `find' in the
> test suite.
> 
> What is the problem here?  Why doesn't the existing code work?

http://gcc.gnu.org/ml/gcc/2002-08/msg00901.html

> 
> If it can't be made to work (I'd like to understand why first), let's
> use the Tcl glob command instead of `exec find'.
> 

I don't know much about tcl. But I have no problems with that as long
as it works.


H.J.

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Tom Tromey]

H.J.> http://gcc.gnu.org/ml/gcc/2002-08/msg00901.html

Thanks.  Try the appended patch.  I think it will work (I did some
testing in tclsh).  If it works for you I will check it in.

Tom

Index: ChangeLog
from  Tom Tromey  <tromey@redhat.com>

	* lib/libjava.exp (libjava_init): Handle case where gcc has a
	release version number.

Index: lib/libjava.exp
===================================================================
RCS file: /cvs/gcc/gcc/libjava/testsuite/lib/libjava.exp,v
retrieving revision 1.43
diff -u -r1.43 libjava.exp
--- lib/libjava.exp 24 Jun 2002 04:44:53 -0000 1.43
+++ lib/libjava.exp 16 Aug 2002 17:41:01 -0000
@@ -132,7 +132,9 @@
 
     # Determine the version so we can find the libgcj jar file.
     set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
-    regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
+    # Extract the first line and operate on that.
+    set text [lindex [split $text \n] 0]
+    regexp -- "gcj \[^ \]+ (\[^ \]+)( .*|)$" $text ignore gcc_version
     verbose "jar file is libgcj-$gcc_version.jar"
 
     # FIXME: This finds libgcj.spec for the default multilib.

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[H. J. Lu]

On Fri, Aug 16, 2002 at 11:43:46AM -0600, Tom Tromey wrote:
> H.J.> http://gcc.gnu.org/ml/gcc/2002-08/msg00901.html
> 
> Thanks.  Try the appended patch.  I think it will work (I did some
> testing in tclsh).  If it works for you I will check it in.
> 
> Tom
> 
> Index: ChangeLog
> from  Tom Tromey  <tromey@redhat.com>
> 
> 	* lib/libjava.exp (libjava_init): Handle case where gcc has a
> 	release version number.
> 
> Index: lib/libjava.exp
> ===================================================================
> RCS file: /cvs/gcc/gcc/libjava/testsuite/lib/libjava.exp,v
> retrieving revision 1.43
> diff -u -r1.43 libjava.exp
> --- lib/libjava.exp 24 Jun 2002 04:44:53 -0000 1.43
> +++ lib/libjava.exp 16 Aug 2002 17:41:01 -0000
> @@ -132,7 +132,9 @@
>  
>      # Determine the version so we can find the libgcj jar file.
>      set text [eval exec "$GCJ_UNDER_TEST --version 2>@ stdout"]
> -    regexp -- "gcj \[^ \]+ (\[^ \]+) .*" $text ignore gcc_version
> +    # Extract the first line and operate on that.
> +    set text [lindex [split $text \n] 0]
> +    regexp -- "gcj \[^ \]+ (\[^ \]+)( .*|)$" $text ignore gcc_version
>      verbose "jar file is libgcj-$gcc_version.jar"
>  
>      # FIXME: This finds libgcj.spec for the default multilib.

I don't like it. We got into trouble with it before already. What is
wrong with tcl glob? Here is mine replacing find with tcl glob.


H.J.

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Tom Tromey]

>>>>> "H.J." == H J Lu <hjl@lucon.org> writes:

H.J.> What is wrong with tcl glob? Here is mine replacing find with
H.J.> tcl glob.

Ok.

Tom

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[H. J. Lu]

On Fri, Aug 16, 2002 at 12:15:59PM -0600, Tom Tromey wrote:
> >>>>> "H.J." == H J Lu <hjl@lucon.org> writes:
> 
> H.J.> What is wrong with tcl glob? Here is mine replacing find with
> H.J.> tcl glob.
> 
> Ok.
> 

Done. Can I check it into the 3.2 branch?


H.J.

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Tom Tromey]

>>>>> "H.J." == H J Lu <hjl@lucon.org> writes:

H.J.> Done. Can I check it into the 3.2 branch?

I haven't been paying attention to the rules for the branch.
If I can approve it, consider it approved.

Tom

Re: PATCH: GCC 3.2 suddenly fails every java test, CLASSPATH wrong

[Gerald Pfeifer]

On 16 Aug 2002, Tom Tromey wrote:
>> Done. Can I check it into the 3.2 branch?
> I haven't been paying attention to the rules for the branch.
> If I can approve it, consider it approved.

Just a general remark (triggered by this): This is why I've added the
status table with all relevant release branches to our main page at

  http://gcc.gnu.org/

earlier this year. ;-)

(Mostly) Mark and me try to keep this up to date, so that the current
state of all these branches is reflected there; for example, for GCC 3.2
we currently have:

  Open for all maintainers; fixes for regressions only.

Hope this helps,
Gerald
-- 
Gerald "Jerry" pfeifer@dbai.tuwien.ac.at http://www.dbai.tuwien.ac.at/~pfeifer/