[Bug libgcj/44415] New: [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[Bug libgcj/44415] New: [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[ro at gcc dot gnu dot org]

The recent introduction of this patch

2010-05-29  Mike Stump  <mikestump@comcast.net>

        * configure.ac: Add multilib support for gmp.  Proper -I and -L
        flags for gmp are added.

broke bootstrap on i386-pc-solaris2.11 and almost certainly at least all
Solaris 2
platforms with a static libgmp:

/bin/ksh ../../../libtool --tag=CC   --mode=link
/vol/gcc/obj/gcc-4.6.0-20100602/11-gcc/./gcc/xgcc
-B/vol/gcc/obj/gcc-4.6.0-20100602/11-gcc/./gcc/
-B/vol/gcc/i386-pc-solaris2.11/bin/ -B/vol/gcc/i386-pc-solaris2.11/lib/
-isystem /vol/gcc/i386-pc-solaris2.11/include -isystem
/vol/gcc/i386-pc-solaris2.11/sys-include    -W -Wall -Wmissing-declarations
-Wwrite-strings -Wmissing-prototypes -Wno-long-long  -I/vol/gcc/include  -g -O2
-module -version-info 0:0:0 -no-undefined -L/vol/gcc/lib -lgmp -avoid-version 
-o libjavamath.la -rpath /vol/gcc/lib/gcj-4.6.0-11 gnu_java_math_GMP.lo
../../../native/jni/classpath/jcl.lo -lsocket -lnsl 
libtool: link: /vol/gcc/obj/gcc-4.6.0-20100602/11-gcc/./gcc/xgcc
-B/vol/gcc/obj/gcc-4.6.0-20100602/11-gcc/./gcc/
-B/vol/gcc/i386-pc-solaris2.11/bin/ -B/vol/gcc/i386-pc-solaris2.11/lib/
-isystem /vol/gcc/i386-pc-solaris2.11/include -isystem
/vol/gcc/i386-pc-solaris2.11/sys-include    -shared -Wl,-z -Wl,text -Wl,-h
-Wl,libjavamath.so -o .libs/libjavamath.so  .libs/gnu_java_math_GMP.o
../../../native/jni/classpath/.libs/jcl.o   -L/vol/gcc/lib
/vol/gcc/lib/libgmp.a -lsocket -lnsl -lc   
Text relocation remains                         referenced
    against symbol                  offset      in file
.rodata.str1.4 (section)            0x2e       
/vol/gcc/lib/libgmp.a(lt19-get_str.o)
[...]

The static libgmp is usually non-PIC, while we need PIC code to link
libjavamath.so.

It turned out to be remarkably difficult to work around this:

* While gcc on Solaris supports -mimpure-text to allow linking of non-PIC code
  into a shared library, all this does is disable the implicit -z text enabled
  by gcc -shared.  Since libtool passes -z text by itself, this doesn't help.

* Sun ld doesn't allow -z text -z textoff, so I had to resort to edititing
  -z text out of libtool and adding -mimpure-text after the -shared there.

To fix this for real, one needs to build a PIC libgmp.a, which at the very
least
needs to be documented.

Apart from that, the patch has a couple of other problems:

* The ChangeLog entry belongs into classpath/ChangeLog, not libjava/ChangeLog.

* For multilibbed targets, it used to be enough to have libgmp for the default
  multilib.  Now you need libgmp for every multilib, otherwise the facitilies
  between e.g. 32-bit and 64-bit libjavamath.so differ.

* There seems to be no provision to pass in -L and -R flags for the non-default
  multilibs, and neither libtool nor the libjava/classpath make machinery seem
  to support automatically looking in e.g. $libdir/amd64 for 64-bit multilibs.

It seems this needs considerably more work to be complete and useful.


-- 
           Summary: [4.6 regression] gmp multilib support broke bootstrap
                    with static libgmp
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: build
          Severity: normal
          Priority: P3
         Component: libgcj
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: ro at gcc dot gnu dot org
 GCC build triplet: *-*-solaris2.*
  GCC host triplet: *-*-solaris2.*
GCC target triplet: *-*-solaris2.*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2010-06-04 17:48 -------
First off this is not a regression; I ran into this issue back a couple of
years ago with libjava requiring libgmp.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2010-06-04 17:49 -------
See PR 39747.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  BugsThisDependsOn|                            |39747


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[ro at CeBiTec dot Uni-Bielefeld dot DE]

------- Comment #3 from ro at CeBiTec dot Uni-Bielefeld dot DE  2010-06-07 14:03 -------
Subject: Re:  [4.6 regression] gmp multilib support broke bootstrap with static
libgmp

> ------- Comment #1 from pinskia at gcc dot gnu dot org  2010-06-04 17:48 -------
> First off this is not a regression; I ran into this issue back a couple of
> years ago with libjava requiring libgmp.

It sure is: you used to be able to bootstrap with a static, non-PIC
libgmp until this change went in, and now bootstrap fails.  If this is
not a regression, what is?

        Rainer


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[mrs at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

mrs@gcc.gnu.org <mrs at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mrs at gcc dot gnu.org

--- Comment #5 from mrs at gcc dot gnu.org <mrs at gcc dot gnu.org> 2011-01-19 18:10:01 UTC ---
If we are to document bugs, please, let's include a bug number for people to
reference so that they can see when the problem is fixed.

[Bug libgcj/44415] [4.6 regression] gmp multilib support broke bootstrap with static libgmp

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44415

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-01-19 12:01:02 UTC ---
(In reply to comment #3)
> Subject: Re:  [4.6 regression] gmp multilib support broke bootstrap with static
> libgmp
> 
> > ------- Comment #1 from pinskia at gcc dot gnu dot org  2010-06-04 17:48 -------
> > First off this is not a regression; I ran into this issue back a couple of
> > years ago with libjava requiring libgmp.
> 
> It sure is: you used to be able to bootstrap with a static, non-PIC
> libgmp until this change went in, and now bootstrap fails.  If this is
> not a regression, what is?

How did it then succeed to link libjavamath.so?

Can one of you prepare an install manual patch to say that a PIC libgmp
is required for building libjava?