This is the mail archive of the gcc-prs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

libstdc++/5859: gdb fails to break in C++ MPI code



>Number:         5859
>Category:       libstdc++
>Synopsis:       gdb fails to break in C++ MPI code
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Mar 06 19:06:01 PST 2002
>Closed-Date:
>Last-Modified:
>Originator:     Julian Cummings
>Release:        3.0
>Organization:
>Environment:
i686 running Linux 2.4.9-ac9
>Description:
The gdb debugger fails to break on the symbol MPI_Init
in a trivial MPI code written in C++.  The code uses the
mpich package from Argonne National Laboratory and is 
compiled using mpiCC, a script that invokes the g++ compiler
and links with libmpich.a.  The code is executed using 
"mpirun" and the -gdb flag, which causes gdb to set a 
breakpoint at MPI_Init and run the program.  

The problem is triggered by any C++ MPI code that utilizes 
C++ streams, for example, to write something to std::cout.
In this case, the code will run through all of main() and
stop inside the global object destructor phase.  With no
usage of C++ streams, the code halts at MPI_Init as it 
should.  Also, a comparable MPI code written in C functions
properly with gdb.

I am reporting this as a gcc bug because it does not happen
with gcc 2.96.  It seems that something has changed with
respect to the C++ standard library and streams that makes
debugging of MPI codes with gdb no longer possible.  
>How-To-Repeat:
You will need to have a recent version of mpich
installed (http://www.mcs.anl.gov/mpi/mpich).  I am using 
mpich-1.2.1.  With this installed, build my example code
with the command "mpiCC -g -o cctest main.cc".  Then run it
on a single processor, for example, with the command
"mpirun -np 1 ./cctest".  Now try to run it under the gdb
debugger by adding the flag -gdb to the mpirun command.
You should see that the code does not break at MPI_Init
as expected.
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: application/octet-stream; name="main.cc"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename="main.cc"
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=


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]