Re: PR 10656 and PR 10657

Re: PR 10656 and PR 10657

[Carl Eugen Hoyos]

On Wed, 2003-05-07 at 15:30, Jeff Sturm wrote:
> On 6 May 2003, Mark Mitchell wrote:
> > These PRs represent serious problems with the Java front end --
problems
> > that prevent building GCC if Java is enabled.  Both problems occur on
> > Solaris machines.
>
> I have a patch for PR 10656, but cannot test it myself on Solaris 7.  I
> may have some help.  Is tomorrow too late for 3.3?

I tested the part of the patch concerning v8.S by deleting the line
containing .register: The bootstrap succeeds, but ffitest fails:
When linked against the dynamic library, ffitest fails immediately with
bus error, linked against the static libffi.a, the following output can be
seen:

$ ffitest
709 return value tests run
long double return value tests ok!
2 vs 2
float arg tests ok!
strlen tests passed
double return value tests ok!
many arg tests ok!
1383096 promotion tests run
structure test 1 ok!

3.26278e-309
0
4.55
5.66
structure test 2 ok!
246
246
structure test 3 ok!
structure test 4 ok!
structure test 5 ok!
5.55
6.66
6.55
7.66
structure test 6 ok!
5.55
55.5
6.66
6.55
56.5
7.66
structure test 7 ok!
5.55
55.5
-5.55
-55.5
6.55
56.5
-4.55
-54.5
structure test 8 ok!
5.55
5
6.55
6
structure test 9 ok!

Enter FFI_CLOSURES

1 2 3 4 127 429 7 8 9 10 11 12 13 19 21 1 3: 680
sh: bus error  ffitest

Re: PR 10656 and PR 10657

[Jeff Sturm]

On Thu, 8 May 2003, Carl Eugen Hoyos wrote:
> On Wed, 2003-05-07 at 15:30, Jeff Sturm wrote:
> > On 6 May 2003, Mark Mitchell wrote:
> > > These PRs represent serious problems with the Java front end --
> problems
> > > that prevent building GCC if Java is enabled.  Both problems occur on
> > > Solaris machines.
> >
> > I have a patch for PR 10656, but cannot test it myself on Solaris 7.  I
> > may have some help.  Is tomorrow too late for 3.3?
>
> I tested the part of the patch concerning v8.S by deleting the line
> containing .register: The bootstrap succeeds, but ffitest fails:
> When linked against the dynamic library, ffitest fails immediately with
> bus error, linked against the static libffi.a, the following output can be
> seen:
...
> sh: bus error  ffitest

I get the same.  The failing test here was written since SPARC closures
were added to libffi.

Regardless, bootstrapping without failure is an improvement.  Thanks
for testing.

Jeff

Re: PR 10656 and PR 10657

[Eric Botcazou]

> I couldn't reproduce PR 10657.  Solaris 8 has iconv() in libc, anyway, so
> I don't believe this is a Solaris problem, nor that it should affect the
> 3.3 release.

I couldn't either (Solaris 8, Sun cc/as/ld).

> (Java wants to use iconv() whenever it's available regardless of
> --disable-nls.  See also prs 10013, 9045.)

Warren already reported the same problem for GCC 3.2.1 on Solaris 2.5.1 (see 
PR bootstrap/9047).

-- 
Eric Botcazou

Re: PR 10656 and PR 10657

[Anthony Green]

On Wed, 2003-05-07 at 15:37, Mark Mitchell wrote:
> No, if it gets done tomorrow, we can get it in.

I will test this today.

AG

Re: PR 10656 and PR 10657

[Jeff Sturm]

On 7 May 2003, Mark Mitchell wrote:
> > > --disable-nls does not seem to be working correctly when building jc1.
> >
> > I couldn't reproduce PR 10657.  Solaris 8 has iconv() in libc, anyway, so
> > I don't believe this is a Solaris problem, nor that it should affect the
> > 3.3 release.
>
> Huh.  Does Solaris even have libiconv, then?

Solaris 8 does not.  Earlier releases may have.

> Note that the PR involves
> using GNU ld; does that somehow turn on an attempt to link in -liconv?

libiconv is detected by the bootstrap compiler, which could certainly
search directories that the built compiler will not.

Jeff

Re: PR 10656 and PR 10657

[Mark Mitchell]

On Wed, 2003-05-07 at 15:30, Jeff Sturm wrote:
> On 6 May 2003, Mark Mitchell wrote:
> > These PRs represent serious problems with the Java front end -- problems
> > that prevent building GCC if Java is enabled.  Both problems occur on
> > Solaris machines.
> 
> I have a patch for PR 10656, but cannot test it myself on Solaris 7.  I
> may have some help.  Is tomorrow too late for 3.3?

No, if it gets done tomorrow, we can get it in.

> > PR 10657 looks like it might be a problem on other systems as well;
> > --disable-nls does not seem to be working correctly when building jc1.
> 
> I couldn't reproduce PR 10657.  Solaris 8 has iconv() in libc, anyway, so
> I don't believe this is a Solaris problem, nor that it should affect the
> 3.3 release.

Huh.  Does Solaris even have libiconv, then?  Note that the PR involves
using GNU ld; does that somehow turn on an attempt to link in -liconv?

Thanks,

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com

Re: PR 10656 and PR 10657

[Jeff Sturm]

On 6 May 2003, Mark Mitchell wrote:
> These PRs represent serious problems with the Java front end -- problems
> that prevent building GCC if Java is enabled.  Both problems occur on
> Solaris machines.

I have a patch for PR 10656, but cannot test it myself on Solaris 7.  I
may have some help.  Is tomorrow too late for 3.3?

> PR 10657 looks like it might be a problem on other systems as well;
> --disable-nls does not seem to be working correctly when building jc1.

I couldn't reproduce PR 10657.  Solaris 8 has iconv() in libc, anyway, so
I don't believe this is a Solaris problem, nor that it should affect the
3.3 release.

(Java wants to use iconv() whenever it's available regardless of
--disable-nls.  See also prs 10013, 9045.)

Jeff

Re: PR 10656 and PR 10657

[Mark Mitchell]

>  
> > We need a fix for these problems ASAP; otherwise, I'll be forced to
> > disable Java by default on Solaris boxes in GCC 3.3, which would be a
> > step backwards.  
> 
> I would recommend against doing this, unless we can more specifically
> narrow down which Solaris configurations have problems.

At the very least, if we cannot get successful builds out of the box, we
need to document some procedure which is known to work.

For example, I don't mind if we say "--disable-nls does not work on
Solaris and you must use GNU binutils."

That would be better than disabling Java, I agree.

Thanks,

-- 
Mark Mitchell <mark@codesourcery.com>
CodeSourcery, LLC

Re: PR 10656 and PR 10657

[Joe Buck]

On Tue, May 06, 2003 at 11:10:21PM -0700, Mark Mitchell wrote:
> These PRs represent serious problems with the Java front end -- problems
> that prevent building GCC if Java is enabled.  Both problems occur on
> Solaris machines.

I have a successful Java build of the prerelease tarball on a Solaris 2.8
machine; test results are at 

http://gcc.gnu.org/ml/gcc-testresults/2003-05/msg00426.html

Some details: binutils 2.13.1 are used, and CONFIG_SHELL and SHELL are
set to /bin/ksh (the latter setting might not be needed, but the former
definitely is).
 
> We need a fix for these problems ASAP; otherwise, I'll be forced to
> disable Java by default on Solaris boxes in GCC 3.3, which would be a
> step backwards.  

I would recommend against doing this, unless we can more specifically
narrow down which Solaris configurations have problems.

 

Re: PR 10656 and PR 10657

[Eric Botcazou]

> (re: 10656)
> Is libffi even supported on Solaris/SPARC 7? I mean, judging by the files
> in libffi/src/sparc, it should support only Solaris 8 or 9. I could
> be wrong on this though...

You are confusing the version of the SPARC architecture with the version of 
the Solaris operating system: v8 and v9 are in the former class while 
Solaris 7, Solaris 8 and Solaris 9 (aka 2.7, 2.8 and 2.9) are in the latter.

-- 
Eric Botcazou

Re: PR 10656 and PR 10657

[Ranjit Mathew]

Hi,

(re: 10656) 
Is libffi even supported on Solaris/SPARC 7? I mean, judging by the files
in libffi/src/sparc, it should support only Solaris 8 or 9. I could
be wrong on this though...


> PR 10657 looks like it might be a problem on other systems as well;
> --disable-nls does not seem to be working correctly when building jc1.

AFAICT, PR java/2812 *could* explain this and I suspect AM_ICONV
(in gcc/aclocal.m4) is not as robust as we would have liked it to 
be.

BTW, I and at least one other person (Mohan Embar) have been happily 
building GCJ 3.3 with --disable-nls without facing any problems 
(MinGW/Windows).

Ranjit.

RE: PR 10656 and PR 10657

[Hamdy Mobarak]

Please unsubscribe me!
Please unsubscribe me!
Please unsubscribe me!


-----Original Message-----
From: java-owner@gcc.gnu.org [mailto:java-owner@gcc.gnu.org]On Behalf Of
Mark Mitchell
Sent: Wednesday, May 07, 2003 08:10
To: java@gcc.gnu.org
Cc: gcc@gcc.gnu.org
Subject: PR 10656 and PR 10657


These PRs represent serious problems with the Java front end -- problems
that prevent building GCC if Java is enabled.  Both problems occur on
Solaris machines.

We need a fix for these problems ASAP; otherwise, I'll be forced to
disable Java by default on Solaris boxes in GCC 3.3, which would be a
step backwards.  

PR 10657 looks like it might be a problem on other systems as well;
--disable-nls does not seem to be working correctly when building jc1.

Thanks,

-- 
Mark Mitchell <mark@codesourcery.com>
CodeSourcery, LLC

PR 10656 and PR 10657

[Mark Mitchell]

These PRs represent serious problems with the Java front end -- problems
that prevent building GCC if Java is enabled.  Both problems occur on
Solaris machines.

We need a fix for these problems ASAP; otherwise, I'll be forced to
disable Java by default on Solaris boxes in GCC 3.3, which would be a
step backwards.  

PR 10657 looks like it might be a problem on other systems as well;
--disable-nls does not seem to be working correctly when building jc1.

Thanks,

-- 
Mark Mitchell <mark@codesourcery.com>
CodeSourcery, LLC