Hi Oliver,

I'm sorry for letting this bug report get past me.  I just never saw it.
In any event, please see below for my comments.

On Thu, Aug 12, 2010 at 05:45:54PM +0200, Oliver Sander wrote:
> Package: coinor-libipopt-dev
> Version: 3.8.3-2
> Severity: normal
>
> I have been using a hand-compiled libipopt with a C++ program
> for a while.  The linear solver for ipopt was ma27 (fortran).  The
> C++ program was purely sequential, hence there was no need
> to mention or have MPI.
>
> Now I want to build the same program with libipopt from the new
> package (thanks for packaging, by the way!).  As the linker flags
> I use
>     -lipopt -llapack -lblas -lgfortran
> which is what worked before.  (As ma27 is not used anymore I guess
> the -lgfortran is obsolete.)  The program compiles and links,
> however when running it I get
> symbol lookup error: /usr/lib/libipopt.so.0: undefined symbol: MPI_Init
> on the first call to libipopt, and the program aborts.
>
I think that this might be the same problem reported in #592324.

> Trying to be smart, I add -lmpi to me list of linker flags.  Then I
> get the run time error
> [haile:4637] *** An error occurred in MPI_Comm_rank
> [haile:4637] *** on communicator MPI_COMM_WORLD
> [haile:4637] *** MPI_ERR_COMM: invalid communicator
> [haile:4637] *** MPI_ERRORS_ARE_FATAL (your MPI job will now abort)
>
> Is this a bug or am I using the wrong flags?  How can I find out what
> the correct linker flags are?  The ipopt manual mentions a file
> ipopt_addlibs_cpp.txt with this information, but it is not contained
> in the Debian package.
>
Please try installing libmumps-4.9.2 and libmumps-seq-4.9.2, then try
your application again.  Please try both with and without -lmpi.  It
should work without, but I am curious as to what happens in either case.

> I guess the most elegant solution would be to use pkg-config.
> Support for pkg-config seems to have entered the ipopt code
> shortly after the 3.8.3 release.  Maybe it can be backported?
>
I am working on the new upstream version.  I will have a look at this as
part of getting that packaged.

Regards,

-Roberto

-- 
Roberto C. Sánchez
http://people.connexer.com/~roberto
http://www.connexer.com

Attachment: signature.asc
Description: Digital signature

Reply via email to