> I'm working on updating the MELT plugin package that is living in
> mandriva cooker. Turns out that it does not work. First stage melt.so is
> built successfully but is unable to be loaded to continue:

> ***** doing  melt-stage1/warmelt-first.c from melt-build.tpl line 343
> doing warmelt-first+melt-stage1 from melt-build.tpl line 344
> gcc -c @warmelt-first+melt-stage1.args
> cc1: error: cannot load plugin
> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so
> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so:
> undefined symbol: warning_at

>> Hello,

>> I'm working on updating the MELT plugin package that is living in
>> mandriva cooker. Turns out that it does not work. First stage melt.so is
>> built successfully but is unable to be loaded to continue:

>> ***** doing  melt-stage1/warmelt-first.c from melt-build.tpl line 343
>> doing warmelt-first+melt-stage1 from melt-build.tpl line 344
>> gcc -c @warmelt-first+melt-stage1.args
>> cc1: error: cannot load plugin
>> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so
>> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so:
>> undefined symbol: warning_at

> Are you building it for GCC 4.7? Some versions of GCC 4.7 (in particular Debian/Sid) are
> built with the g++ compiler (ie the entire GCC is compiled in C++ mode, not in C99 mode).

> I suspect that by compiling both melt.so and all MELT modules with g++ it should be ok.

> You'll need to edit the MELT-Plugin-Makefile and the melt-module.mk for that.

> BTW, this is not a MELT specific issue. All GCC plugins have it too!
> It was discussed on gcc@ or gcc-patches@

> Cheers.

> >> Hello,

> >> I'm working on updating the MELT plugin package that is living in
> >> mandriva cooker. Turns out that it does not work. First stage melt.so is
> >> built successfully but is unable to be loaded to continue:

> >> ***** doing  melt-stage1/warmelt-first.c from melt-build.tpl line 343
> >> doing warmelt-first+melt-stage1 from melt-build.tpl line 344
> >> gcc -c @warmelt-first+melt-stage1.args
> >> cc1: error: cannot load plugin
> >> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so
> >> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so:
> >> undefined symbol: warning_at

> > Are you building it for GCC 4.7? Some versions of GCC 4.7 (in particular Debian/Sid) are
> > built with the g++ compiler (ie the entire GCC is compiled in C++ mode, not in C99 mode).

> Yep, I mentionned it in the subject :)

> > % nm -D /usr/lib/gcc/x86_64-linux-gnu/4.6/cc1 | grep warning_at
> > 000000000057d9f0 T warning_at
> >  % nm -D /usr/lib/gcc/x86_64-linux-gnu/4.7/cc1 | grep warning_at
> > 0000000000c27ef0 T _Z10warning_atjiPKcz
> >  % nm -C -D /usr/lib/gcc/x86_64-linux-gnu/4.7/cc1 | grep warning_at
> > 0000000000c27ef0 T warning_at(unsigned int, int, char const*, ...)

> > I suspect that by compiling both melt.so and all MELT modules with g++ it should be ok.

> > You'll need to edit the MELT-Plugin-Makefile and the melt-module.mk for that.

> Just adding a dummy warning_at entry point in melt is okay ?

>> Le 25/03/2012 19:03, Basile Starynkevitch a crit :
>>> On Sun, 25 Mar 2012 16:40:31 +0200
>>> ali...@mandriva.com wrote:

>>>> Hello,

>>>> I'm working on updating the MELT plugin package that is living in
>>>> mandriva cooker. Turns out that it does not work. First stage melt.so is
>>>> built successfully but is unable to be loaded to continue:

>>>> ***** doing  melt-stage1/warmelt-first.c from melt-build.tpl line 343
>>>> doing warmelt-first+melt-stage1 from melt-build.tpl line 344
>>>> gcc -c @warmelt-first+melt-stage1.args
>>>> cc1: error: cannot load plugin
>>>> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so
>>>> /home/alex/BuildSystem/gcc-plugin-melt/BUILD/melt-0.9.4.b-plugin-for-gcc-4. 6-or-4.7/melt.so:
>>>> undefined symbol: warning_at

>>> Are you building it for GCC 4.7? Some versions of GCC 4.7 (in particular Debian/Sid) are
>>> built with the g++ compiler (ie the entire GCC is compiled in C++ mode, not in C99 mode).

>> Yep, I mentionned it in the subject :)

>>> % nm -D /usr/lib/gcc/x86_64-linux-gnu/4.6/cc1 | grep warning_at
>>> 000000000057d9f0 T warning_at
>>>  % nm -D /usr/lib/gcc/x86_64-linux-gnu/4.7/cc1 | grep warning_at
>>> 0000000000c27ef0 T _Z10warning_atjiPKcz
>>>  % nm -C -D /usr/lib/gcc/x86_64-linux-gnu/4.7/cc1 | grep warning_at
>>> 0000000000c27ef0 T warning_at(unsigned int, int, char const*, ...)

>>> I suspect that by compiling both melt.so and all MELT modules with g++ it should be ok.

>>> You'll need to edit the MELT-Plugin-Makefile and the melt-module.mk for that.

>> Just adding a dummy warning_at entry point in melt is okay ?

> I don't think that would help or be enough. We need to be sure that

>    melt.so

>    warmelt*.so

> are built with g++, not gcc. Maybe we need to use ie check ENABLE_BUILD_WITH_CXX
> from /usr/lib/gcc/x86_64-linux-gnu/4.7/plugin/include/auto-host.h

> and we surely need to change the CC in MELT-Plugin-Makefile (actually, the occurrences of
> CC) to something else. We also need to change the GCCMELT_CC in melt-module.mk (but that
> is easy).

> However, this issue is not MELT specific. Every GCC plugin (for 4.7, which can be
> compiled both with C++ or with C) has it. Do you know how other packaged plugins solved
> that issue? See e.g. http://gcc.gnu.org/ml/gcc/2012-03/msg00379.html

> My feeling is that we have to do something specific to 4.7, which is the (only) release
> of GCC which can be entirely (I'm talking of the C & Fortran & C++ front-ends, not Go or
> Ada front-ends) compiled with either C or C++ compiler.

> Do you have any suggestions? How do you solve that issue for other GCC plugins in cooker?

> Cheers.

> > However, this issue is not MELT specific. Every GCC plugin (for 4.7, which can be
> > compiled both with C++ or with C) has it. Do you know how other packaged plugins solved
> > that issue? See e.g. http://gcc.gnu.org/ml/gcc/2012-03/msg00379.html

> > My feeling is that we have to do something specific to 4.7, which is the (only) release
> > of GCC which can be entirely (I'm talking of the C & Fortran & C++ front-ends, not Go or
> > Ada front-ends) compiled with either C or C++ compiler.

> > Do you have any suggestions? How do you solve that issue for other GCC plugins in cooker?

> I don't know.

> But at least we can force those variables at make time, since they are
> declared as "?=", but it seems there are things to fix in the code for
> it to be buildable :(

> > > However, this issue is not MELT specific.

> > > > However, this issue is not MELT specific.

> I committed several changes (on the MELT branch, svn rev 185803) which make
> the MELT plugin more friendly with C++ compiled gcc 4.7, but I am not done
> yet.

> I'll probably make a MELT plugin release as soon as I have a plugin
> buildable with a C++ compiled GCC 4.7 & with a C compiled GCC 4.6

>> On Sun, Mar 25, 2012 at 09:51:51PM +0200, Basile Starynkevitch
>> wrote:
>> > On Sun, 25 Mar 2012 21:18:49 +0200
>> > Alexandre Lissy <ali...@mandriva.com> wrote:

>> > > > However, this issue is not MELT specific.

>> I committed several changes (on the MELT branch, svn rev 185803)
>> which make
>> the MELT plugin more friendly with C++ compiled gcc 4.7, but I am
>> not done
>> yet.

>> I'll probably make a MELT plugin release as soon as I have a plugin
>> buildable with a C++ compiled GCC 4.7 & with a C compiled GCC 4.6

> Hello Alexandre & the list

> I just uploaded to
> http://gcc-melt.org/melt-0.9.5rc0-plugin-for-gcc-4.6-or-4.7.tar.gz
> a gzipped tar archive of 4476533 bytes, of md5sum
> a66365f5bae0068a4f23d203180f5a74
> a tentative tarball for the next melt 0.9.5 release which should be
> compilable for a C++
> compiled gcc-4.7 (extracted from MELT branch svn rev 185881)

> at the moment, you need to build it with   make MELTGCC=gcc-4.7
> GCCMELT_CC=g++-4.7

> so it is not release ready yet, since I need to find out how to do
> tricks to detect if
> gcc-4.7 has been compiled in C++ or in C.

> See e.g. http://gcc.gnu.org/ml/gcc/2012-03/msg00375.html
> http://gcc.gnu.org/ml/gcc/2012-03/msg00380.html
> http://gcc.gnu.org/ml/gcc/2012-03/msg00411.html

>  Alexandre, could you check that you can compile it? Please report
> bugs here.

> (I won't be able to read email tomorrow probably)

> Cheers.

> --
> Basile STARYNKEVITCH         http://starynkevitch.net/Basile/
> email: basile<at>starynkevitch<dot>net mobile: +33 6 8501 2359
> 8, rue de la Faiencerie, 92340 Bourg La Reine, France
> *** opinions {are only mine, sont seulement les miennes} ***

> > I just uploaded to
> > http://gcc-melt.org/melt-0.9.5rc0-plugin-for-gcc-4.6-or-4.7.tar.gz
> > a gzipped tar archive of 4476533 bytes, of md5sum
> > a66365f5bae0068a4f23d203180f5a74
> > a tentative tarball for the next melt 0.9.5 release which should be
> > compilable for a C++
> > compiled gcc-4.7 (extracted from MELT branch svn rev 185881)

> > at the moment, you need to build it with   make MELTGCC=gcc-4.7
> > GCCMELT_CC=g++-4.7

> > so it is not release ready yet, since I need to find out how to do
> > tricks to detect if
> > gcc-4.7 has been compiled in C++ or in C.

> > See e.g. http://gcc.gnu.org/ml/gcc/2012-03/msg00375.html
> > http://gcc.gnu.org/ml/gcc/2012-03/msg00380.html
> > http://gcc.gnu.org/ml/gcc/2012-03/msg00411.html

> >  Alexandre, could you check that you can compile it? Please report
> > bugs here.
> It fails, see attached build log.