GCC-3.2.2 pre-release

GCC-3.2.2 pre-release

[Gabriel Dos_Reis]

I uploaded 3.2.2 pre-release tarballs on the FTP server as

    ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*

Please donwload and test them.  Report problems to me and fill GNATS
PRs. 

We've got 6 days testing (and fixing).

Thanks,

-- Gaby

Re: GCC-3.2.2 pre-release

[Joe Buck]

On Tue, Jan 28, 2003 at 10:12:22AM +0100, Gabriel Dos_Reis wrote:
> 
> I uploaded 3.2.2 pre-release tarballs on the FTP server as
> 
>     ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*

That's

	ftp://gcc.gnu.org/pub/gcc/snapshots/gcc-*3.2.2-20030128.*

Re: GCC-3.2.2 pre-release

[Gabriel Dos Reis]

Joe Buck <jbuck@synopsys.com> writes:

| On Tue, Jan 28, 2003 at 10:12:22AM +0100, Gabriel Dos_Reis wrote:
| > 
| > I uploaded 3.2.2 pre-release tarballs on the FTP server as
| > 
| >     ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*
| 
| That's
| 
| 	ftp://gcc.gnu.org/pub/gcc/snapshots/gcc-*3.2.2-20030128.*

Oops. Thanks for the correction.

-- Gaby

Re: GCC-3.2.2 pre-release

[Joe Buck]

On Tue, Jan 28, 2003 at 10:12:22AM +0100, Gabriel Dos_Reis wrote: 
> I uploaded 3.2.2 pre-release tarballs on the FTP server as
>     ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*
> Please donwload and test them.  Report problems to me and fill GNATS
> PRs. 

I've successfully built and run tests for sparc-sun-solaris2.7,
hppa2.0w-hp-hpux11.00, and i686-pc-linux-gnu.  A sparc-sun-solaris2.8
run is in process.  In all cases, binutils 2.13.1 was used for as and ld.

Test results can be found at:

Solaris: http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg01254.html

HPUX: http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg01251.html

GNU/Linux: http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg01245.html

HP has a number of failures.  I see lots of failures for weak symbols that
aren't in the following report (from John David Anglin).  Perhaps a
different binutils version is being used?  Other than those, the failures
seem similar to what we had before.

JDA's report:
http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg00881.html

Re: GCC-3.2.2 pre-release

[Gabriel Dos_Reis]

| On Tue, Jan 28, 2003 at 10:12:22AM +0100, Gabriel Dos_Reis wrote: 
| > I uploaded 3.2.2 pre-release tarballs on the FTP server as
| >     ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*
| > Please donwload and test them.  Report problems to me and fill GNATS
| > PRs. 
| 
| I've successfully built and run tests for sparc-sun-solaris2.7,
| hppa2.0w-hp-hpux11.00, and i686-pc-linux-gnu.  A sparc-sun-solaris2.8
| run is in process.  In all cases, binutils 2.13.1 was used for as and ld.

Thanks for the tests.

| HPUX: http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg01251.html
| 
| GNU/Linux: http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg01245.html
| 
| HP has a number of failures.  I see lots of failures for weak symbols that
| aren't in the following report (from John David Anglin).  Perhaps a
| different binutils version is being used?  Other than those, the failures
| seem similar to what we had before.
| 
| JDA's report:
| http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg00881.html

JDA, please could you tell us which version you used in your testing
(I can't find it in the result you sent).  Or do you have a chance to
confirm/infirm the results obtained by Joe by running the binutils
tool set?

Thanks,

-- Gaby

Re: GCC-3.2.2 pre-release

[Joe Buck]

My report:

> | HPUX: http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg01251.html

JDA's report:
> | http://gcc.gnu.org/ml/gcc-testresults/2003-01/msg00881.html

On Wed, Jan 29, 2003 at 02:49:15AM +0100, Gabriel Dos_Reis wrote:
> JDA, please could you tell us which version you used in your testing
> (I can't find it in the result you sent).  Or do you have a chance to
> confirm/infirm the results obtained by Joe by running the binutils
> tool set?

I think that the word you're looking for is "refute".

And the easiest way to refute it is by noting that the JDA result is for
hppa-unknown-linux-gnu, a different target.  Whoops!  I got confused
because JDA posted some 3.3 tests for hppa-hpux, I didn't see lets of
weak-symbol failures in that test, and I managed to pick the wrong target.
I vaguely recall a report that support for weak symbols on HPUX has
been improved for 3.3, though I don't see any mention of this on
http://gcc.gnu.org/gcc-3.3/changes.html .

Re: GCC-3.2.2 pre-release

[Eric Botcazou]

> HP has a number of failures.  I see lots of failures for weak symbols that
> aren't in the following report (from John David Anglin).  Perhaps a
> different binutils version is being used?  Other than those, the failures
> seem similar to what we had before.

Is the following bug a known problem, or... ? Because I think it comes close 
to being a show stopper: 
http://gcc.gnu.org/ml/gcc-bugs/2003-01/msg01706.html

However, gcc 3.2 and gcc 3.2.1 segfault too according to my preliminary 
tests.

-- 
Eric Botcazou

Re: GCC-3.2.2 pre-release

[Gabriel Dos_Reis]

| > HP has a number of failures.  I see lots of failures for weak symbols that
| > aren't in the following report (from John David Anglin).  Perhaps a
| > different binutils version is being used?  Other than those, the failures
| > seem similar to what we had before.
| 
| Is the following bug a known problem, or... ? Because I think it comes close 
| to being a show stopper: 
| http://gcc.gnu.org/ml/gcc-bugs/2003-01/msg01706.html
| 
| However, gcc 3.2 and gcc 3.2.1 segfault too according to my preliminary 
| tests.

While a serious problem, if it^[OB were already present in 3.2.1 then it
does not qualify as regression or a showstopper.  That does not mean I
would not welcome a fix. 

-- Gaby

Re: GCC-3.2.2 pre-release [and PR 9493]

[Joe Buck]

On Wed, Jan 29, 2003 at 02:38:25PM +0100, Gabriel Dos_Reis wrote:
> | > HP has a number of failures.  I see lots of failures for weak symbols that
> | > aren't in the following report (from John David Anglin).  Perhaps a
> | > different binutils version is being used?  Other than those, the failures
> | > seem similar to what we had before.
> | 
> | Is the following bug a known problem, or... ? Because I think it comes close 
> | to being a show stopper: 
> | http://gcc.gnu.org/ml/gcc-bugs/2003-01/msg01706.html
> | 
> | However, gcc 3.2 and gcc 3.2.1 segfault too according to my preliminary 
> | tests.
> 
> While a serious problem, if it^[OB were already present in 3.2.1 then it
> does not qualify as regression or a showstopper.  That does not mean I
> would not welcome a fix. 

By the criteria we use for GNATS, it is a regression, because
gcc 2.95.3 successfully compiles it.  However, the only strict requirement
for gcc 3.2.2 is that it be an improvement over gcc 3.2.1, so for Gaby's
purposes in getting the release out, it isn't a reason to hold the
release.

However, since 3.3 doesn't have it, it would be nice if someone could
dig out Janis's regression finder and determine when the failure went
away.  If this can be done quickly, perhaps we could isolate a patch that
Gaby could put into 3.2.2.

Sorry, I'm not going to have the cycles to do that in the next couple of
days.
 

Re: GCC-3.2.2 pre-release [and PR 9493]

[Gabriel Dos Reis]

Joe Buck <jbuck@synopsys.com> writes:

| However, since 3.3 doesn't have it, it would be nice if someone could
| dig out Janis's regression finder and determine when the failure went
| away.

I'm going to leave office and I won't have connection this night,..
(I will nevertheless try with local copies but any help is welcome :-)

-- Gaby

Re: GCC-3.2.2 pre-release [and PR 9493]

[Eric Botcazou]

> I'm going to leave office and I won't have connection this night,..
> (I will nevertheless try with local copies but any help is welcome :-)

The following one-liner certainly makes the segfault go away, but I don't 
know if the assembly output is valid (I don't speak PA-RISC).

Index: gcc/cfgcleanup.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/cfgcleanup.c,v
retrieving revision 1.38.2.4.2.1
diff -u -p -r1.38.2.4.2.1 cfgcleanup.c
--- gcc/cfgcleanup.c	16 Nov 2002 01:55:47 -0000	1.38.2.4.2.1
+++ gcc/cfgcleanup.c	29 Jan 2003 19:56:17 -0000
@@ -469,7 +469,7 @@ try_forward_edges (mode, b)
 		    && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG)
 		  break;
 
-	      if (GET_CODE (insn) == NOTE)
+	      if (insn && GET_CODE (insn) == NOTE)
 		break;
 	    }
 
-- 
Eric Botcazou

Re: GCC-3.2.2 pre-release [and PR 9493]

[Joe Buck]

On Wed, Jan 29, 2003 at 09:07:44PM +0100, Eric Botcazou wrote:
> The following one-liner certainly makes the segfault go away, but I don't 
> know if the assembly output is valid (I don't speak PA-RISC).
> 
> Index: gcc/cfgcleanup.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/cfgcleanup.c,v
> retrieving revision 1.38.2.4.2.1
> diff -u -p -r1.38.2.4.2.1 cfgcleanup.c
> --- gcc/cfgcleanup.c	16 Nov 2002 01:55:47 -0000	1.38.2.4.2.1
> +++ gcc/cfgcleanup.c	29 Jan 2003 19:56:17 -0000
> @@ -469,7 +469,7 @@ try_forward_edges (mode, b)
>  		    && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG)
>  		  break;
>  
> -	      if (GET_CODE (insn) == NOTE)
> +	      if (insn && GET_CODE (insn) == NOTE)
>  		break;
>  	    }

It would appear that a null-pointer check can't make things worse, but
it does raise a question as to whether this merely masks a bug that is
somewhere else: is the compiler legitimately expecting a non-null pointer
here?
  
> -- 
> Eric Botcazou

-- 
Q. What's more of a headache than a bug in a compiler.
A. Bugs in six compilers.  -- Mark Johnson

Re: GCC-3.2.2 pre-release [and PR 9493]

[Jan Hubicka]

> On Wed, Jan 29, 2003 at 09:07:44PM +0100, Eric Botcazou wrote:
> > The following one-liner certainly makes the segfault go away, but I don't 
> > know if the assembly output is valid (I don't speak PA-RISC).
> > 
> > Index: gcc/cfgcleanup.c
> > ===================================================================
> > RCS file: /cvs/gcc/gcc/gcc/cfgcleanup.c,v
> > retrieving revision 1.38.2.4.2.1
> > diff -u -p -r1.38.2.4.2.1 cfgcleanup.c
> > --- gcc/cfgcleanup.c	16 Nov 2002 01:55:47 -0000	1.38.2.4.2.1
> > +++ gcc/cfgcleanup.c	29 Jan 2003 19:56:17 -0000
> > @@ -469,7 +469,7 @@ try_forward_edges (mode, b)
> >  		    && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG)
> >  		  break;
> >  
> > -	      if (GET_CODE (insn) == NOTE)
> > +	      if (insn && GET_CODE (insn) == NOTE)
> >  		break;
> >  	    }
> 
> It would appear that a null-pointer check can't make things worse, but
> it does raise a question as to whether this merely masks a bug that is
> somewhere else: is the compiler legitimately expecting a non-null pointer
> here?

No, I guess the fix is correct - it is merely looking for loop preheader
and should crash when there is empty block at the end of insn stream.

Honza
>   
> > -- 
> > Eric Botcazou
> 
> -- 
> Q. What's more of a headache than a bug in a compiler.
> A. Bugs in six compilers.  -- Mark Johnson

Re: GCC-3.2.2 pre-release [and PR 9493]

[Eric Botcazou]

> It would appear that a null-pointer check can't make things worse, but
> it does raise a question as to whether this merely masks a bug that is
> somewhere else: is the compiler legitimately expecting a non-null pointer
> here?

I should have said that the line at stake comes after:

	      for (; insn && GET_CODE (insn) != CODE_LABEL && !INSN_P (insn);
		   insn = NEXT_INSN (insn))
		if (GET_CODE (insn) == NOTE
		    && NOTE_LINE_NUMBER (insn) == NOTE_INSN_LOOP_BEG)
		  break;

So the answer might be no indeed.

-- 
Eric Botcazou

Re: GCC-3.2.2 pre-release

[Gabriel Dos_Reis]

| Gabriel Dos Reis wrote:
| 
| >I uploaded 3.2.2 pre-release tarballs on the FTP server as
| >
| >    ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*
| >
| >Please donwload and test them.  Report problems to me and fill GNATS
| >PRs.
| 
| Bootstrap and regression test on s390-ibm-linux and s390x-ibm-linux
| succeeded;  I've also built and tested glibc 2.2.5 with the compiler
| on both platforms without problems.

Thanks.

-- Gaby

Re: GCC-3.2.2 pre-release

[Ulrich Weigand]

Gabriel Dos Reis wrote:

>I uploaded 3.2.2 pre-release tarballs on the FTP server as
>
>    ftp://gcc.gnu.org/pub/gnu/snapshots/gcc-*3.2.2-20030128.*
>
>Please donwload and test them.  Report problems to me and fill GNATS
>PRs.

Bootstrap and regression test on s390-ibm-linux and s390x-ibm-linux
succeeded;  I've also built and tested glibc 2.2.5 with the compiler
on both platforms without problems.


Mit freundlichen Gruessen / Best Regards

Ulrich Weigand

--
  Dr. Ulrich Weigand
  Linux for S/390 Design & Development
  IBM Deutschland Entwicklung GmbH, Schoenaicher Str. 220, 71032 Boeblingen
  Phone: +49-7031/16-3727   ---   Email: Ulrich.Weigand@de.ibm.com