[Bug bootstrap/54283] New: [4.8 regression] build tools don't run after cxx-conversion merge

[Bug bootstrap/54283] New: [4.8 regression] build tools don't run after cxx-conversion merge

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

             Bug #: 54283
           Summary: [4.8 regression] build tools don't run after
                    cxx-conversion merge
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: crowl@gcc.gnu.org, dnovillo@gcc.gnu.org
              Host: *-*-solaris2.*
            Target: *-*-solaris2.*
             Build: *-*-solaris2.*


Since the merge of the cxx-conversion branch, Solaris bootstrap fails for me.
In stage 1, the build tools fail to run, e.g.

build/genmddeps /vol/gcc/src/hg/trunk/local/gcc/config/i386/i386.md >
tmp-mddeps
ld.so.1: genmddeps: fatal: libstdc++.so.6: open failed: No such file or
directory
make[3]: *** [s-mddeps] Killed
make[3]: *** Waiting for unfinished jobs....

I'm using a self-built gcc 4.4.2 as bootstrap compiler, installed into
/vol/gcc-4.4.  In order to assure that the libstdc++.so.6 included there is
found,
I've so far been configuring with

--with-host-libstdcxx=\'-L/vol/gcc-4.4/lib -Wl,-Bstatic -lstdc++ -Wl,-Bdynamic
-lm\' --with-boot-ldflags=\'-static-libstdc++' -static-libgcc
-L/vol/gcc-4.4/lib\'

This is not enough after the conversion since the build tools don't use either
$BOOT_LDFLAGS or $HOST_LIBS.

The attached patch allowed the bootstrap to finish, but is obviously wrong
as-is
since it doesn't correctly distinguish between build and host tools.  Having
never run a canadian cross myself, I'm not sure what the right approach is.

The first part (using $CC as linker instead of $CXX_FOR_BUILD) is necessary to
avoid linking with g++ which links with -lstdc++ on its own.

  Rainer

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #1 from Rainer Orth <ro at gcc dot gnu.org> 2012-08-16 13:53:37 UTC ---
Created attachment 28030
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28030
hacky patch

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.0

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-08-16 15:29:24 UTC ---
Maybe related that we don't build a "host" libstdc++-v3 but only a target one:

target_modules = { module= libstdc++-v3;
                   bootstrap=true;
                   lib_path=src/.libs;
                   raw_cxx=true; };

and funnily we bootstrap it.  We should bootstrap a host libstdc++-v3 and
build a target libstdc++-v3 (if C++ is enabled).  The host libstdc++-v3
would not need libstdc++ PCHs or multilibs of course.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-11-16 01:06:20 UTC ---
Does this still happen?

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #4 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-11-16 09:11:36 UTC ---
> --- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-11-16
> 01:06:20 UTC ---
> Does this still happen?

It does.  I've a local patch in my tree to allow bootstrapping at all.
Unfortunately, the patch certainly isn't right, but I still can't fully
wrap my head around the full consequences build/host/target distinction.

Btw., I've got the same problem on i686-linux with a self-built gcc 4.4
as bootstrap compiler, so there's nothing Solaris-specific here.

    Rainer

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-11-17
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org
     Ever Confirmed|0                           |1

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-11-17 10:20:30 UTC ---
Does it happen with a recent bootstrap compiler, i.e. for which
-static-libstdc++ is implemented, if you don't set --with-host-libstdcxx?  I
think it's 4.6 and above for Solaris.

Note that --with-boot-ldflags should be superfluous since -static-libstdc++'
-static-libgcc are automatically passed during stage 2 & 3.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #6 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-11-17 23:31:22 UTC ---
> Does it happen with a recent bootstrap compiler, i.e. for which
> -static-libstdc++ is implemented, if you don't set --with-host-libstdcxx?  I
> think it's 4.6 and above for Solaris.

Replying to myself: no, this works (tested with 4.7.x).  So it's possible to
bootstrap without any special configure options if you have a recent compiler.
Which could mean that using --with-stage1-ldflags would work for the others.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

Alexandre Oliva <aoliva at gcc dot gnu.org> changed:

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

--- Comment #7 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2012-11-28 19:21:10 UTC ---
The general assumption on non-cross scenarios is that we're able to run a
program created by CC and CXX.  If this is not the case, not even
configure-time execution tests involving the compilers would fail.  This
suggests to me that we don't have a bug or even a regression here; it just so
happens that we use CXX now, and CXX has never worked “properly” on Rainer's
setting.  I suggest setting LD_LIBRARY_PATH so that CXX works “properly”, or
using a CXX with -static-libstdc++ and whatever other options are needed to
make newly-created executables run.  I'm not sure this requires a documentation
change.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-11-28 21:40:38 UTC ---
> The general assumption on non-cross scenarios is that we're able to run a
> program created by CC and CXX.  If this is not the case, not even
> configure-time execution tests involving the compilers would fail.  This
> suggests to me that we don't have a bug or even a regression here; it just so
> happens that we use CXX now, and CXX has never worked “properly” on Rainer's
> setting.  I suggest setting LD_LIBRARY_PATH so that CXX works “properly”, or
> using a CXX with -static-libstdc++ and whatever other options are needed to
> make newly-created executables run.

-static-libstdc++ is already automatically used if supported.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-18 10:30:12 UTC ---
So not a bug then?

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #10 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-12-18 12:06:21 UTC ---
> --- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-18 10:30:12 UTC ---
> So not a bug then?

I don't think so: even on x86_64-unknown-linux-gnu, when building an
i686-unknown-linux-gnu compiler with an unbundled gcc 4.4 (which does
support -static-libstdc++) where the 32-bit libstdc++.so.6 doesn't live
in /usr/lib, but only in /vol/gcc-4.4/lib, the build tools are the only
ones that don't run.  If it were host and build tools alike, I'd agree,
but the host tools are fine, as they were before the cxx-conversion.

    Rainer

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-18 12:09:52 UTC ---
If you don't add that /vol/gcc-4.4/lib/ to ld.so.conf (or ld.so.conf.d), it is
your responsibility to build with LD_LIBRARY_PATH, otherwise you are
bootstrapping with a non-working C++ compiler.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #12 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-12-18 12:12:32 UTC ---
> --- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-18
> 12:09:52 UTC ---
> If you don't add that /vol/gcc-4.4/lib/ to ld.so.conf (or ld.so.conf.d), it is
> your responsibility to build with LD_LIBRARY_PATH, otherwise you are
> bootstrapping with a non-working C++ compiler.

Not if you build with -static-libstdc++, which does work for e.g. go1
before the cxx-conversion and cc1* afterwards, just not for the build
tools.  That's what this is all about.

    Rainer

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #13 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2012-12-19 11:58:33 UTC ---
> I don't think so: even on x86_64-unknown-linux-gnu, when building an
> i686-unknown-linux-gnu compiler with an unbundled gcc 4.4 (which does
> support -static-libstdc++) where the 32-bit libstdc++.so.6 doesn't live
> in /usr/lib, but only in /vol/gcc-4.4/lib, the build tools are the only
> ones that don't run.

Are you sure that GCC 4.4 supports -static-libstdc++?  In any case, the
configure script thinks that it doesn't so it disables -static-libstdc++ in
stage 1 for it.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19 12:04:19 UTC ---
Given that -static-libstdc++ has been added to trunk in 2009-06-25, and gcc 4.4
has been branched 2009-03-27, you'd need to have local patches on top of 4.4 if
it would support that option.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #15 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2012-12-19 14:33:55 UTC ---
> --- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-19
> 12:04:19 UTC ---
> Given that -static-libstdc++ has been added to trunk in 2009-06-25, and gcc 4.4
> has been branched 2009-03-27, you'd need to have local patches on top of 4.4 if
> it would support that option.

Seems you're right: the vanilla g++ 4.4.2 doesn't support the option.
I'll give it a try with 4.7 instead, but this may take until after the
christmas holidays.

    Rainer

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #16 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2013-01-07 12:08:27 UTC ---
I can now confirm that using g++ 4.7 as bootstrap compiler works out of
the box.  Even with the problems I've observed with g++ 4.4, this seems
a suitable way out.  Maybe the requirement/suggestion to use a bootstrap
compiler supporting -static-libstdc++ should be documented?

Thanks.
    Rainer

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-07 12:55:39 UTC ---
But that is not a requirement.  The requirement is using a C++ compiler that
works out of the box (compiler configured in a path where its shared libraries
aren't found by the dynamic linker isn't), or making sure through
LD_LIBRARY_PATH or tweaking spec (to add -rpath) that the compiler works out of
the box, or compiler which supports -static-libstdc++.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |INVALID

--- Comment #18 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-07 13:01:38 UTC ---
.

[Bug bootstrap/54283] [4.8 regression] build tools don't run after cxx-conversion merge

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54283

--- Comment #19 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2013-01-07 13:06:16 UTC ---
> --- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-07
> 12:55:39 UTC ---
> But that is not a requirement.  The requirement is using a C++ compiler that
> works out of the box (compiler configured in a path where its shared libraries
> aren't found by the dynamic linker isn't), or making sure through
> LD_LIBRARY_PATH or tweaking spec (to add -rpath) that the compiler works out of
> the box, or compiler which supports -static-libstdc++.

But that first part (every single user of executables built with a g++
installed into non-system directories having to tweak LD_LIBRARY_PATH,
gcc builders/installers having to tweak specs or whatever) has been a
major shortcoming of gcc for years.  There needs to be a way that gcc
DTRT out of the box; a compiler producing executables that don't work
without prior additional setup is a joke.  Maybe I can work on this in
the 4.9 timeframe.

    Rainer