[petsc-dev] ISRestoreIndicesF90 crashes with empty IS

[Blaise Bourdin]

Hi,

There seems to be a bug in ISRestoreIndicesF90 when the locally owned portion of the index set is empty. I am attaching a small example illustrating the issue.

mpiexec -n 2 ./ISBugf90 -start_in_debugger
[0]PETSC ERROR: PETSC: Attaching gdb to ./ISBugf90 of pid 34744 on display /tmp/launch-kaUROB/org.x:0 on machine galerkin.math.lsu.edu
[1]PETSC ERROR: PETSC: Attaching gdb to ./ISBugf90 of pid 34745 on display /tmp/launch-kaUROB/org.x:0 on machine galerkin.math.lsu.edu
[0] Index set is permutation
[0] Number of indices in (stride) set 0
[1] Index set is permutation
[1] Number of indices in (stride) set 2
[1] 0 0
[1] 1 1
           1           0           1
           0
[0]PETSC ERROR: PetscTrFreeDefault() called from ISRestoreIndices_Stride() line 174 in /opt/HPC/petsc-dev/src/vec/is/impls/stride/stride.c
[0]PETSC ERROR: Block at address 0x105647200 is corrupted; cannot free;
may be block not allocated with PetscMalloc()
[0]PETSC ERROR: PetscTrFreeDefault() line 287 in /opt/HPC/petsc-dev/src/sys/memory/mtr.c Bad location or corrupted memory

The stack trace is 
(gdb) where
#0  0x00007fff8e8c282a in __kill ()
#1  0x00007fff8ecf0a9c in abort ()
#2  0x00000001039bd19c in PetscAbortErrorHandler (comm=1140850689, line=287, fun=0x105491428 "PetscTrFreeDefault", file=0x105490a20 "/opt/HPC/petsc-dev/src/sys/memory/mtr.c", dir=0x104ff6808 "", n=78, p=PETSC_ERROR_INITIAL, mess=0x7fff6345e500 "Bad location or corrupted memory", ctx=0x0) at /opt/HPC/petsc-dev/src/sys/error/errabort.c:62
#3  0x00000001039b4c72 in PetscError (comm=1140850689, line=287, func=0x105491428 "PetscTrFreeDefault", file=0x105490a20 "/opt/HPC/petsc-dev/src/sys/memory/mtr.c", dir=0x104ff6808 "", n=78, p=PETSC_ERROR_INITIAL, mess=0x105490d2c "Bad location or corrupted memory") at /opt/HPC/petsc-dev/src/sys/error/err.c:356
#4  0x000000010394c29b in PetscTrFreeDefault (aa=0x105647950, line=174, function=0x1054ad654 "ISRestoreIndices_Stride", file=0x1054ad0a0 "/opt/HPC/petsc-dev/src/vec/is/impls/stride/stride.c", dir=0x104ff6808 "") at /opt/HPC/petsc-dev/src/sys/memory/mtr.c:287
#5  0x0000000103ab8c24 in ISRestoreIndices_Stride (in=0x7ff65901df70, idx=0x7fff6345f2a0) at /opt/HPC/petsc-dev/src/vec/is/impls/stride/stride.c:174
#6  0x0000000103ad3d89 in ISRestoreIndices (is=0x7ff65901df70, ptr=0x7fff6345f2a0) at /opt/HPC/petsc-dev/src/vec/is/interface/index.c:417
#7  0x0000000103c425c6 in isrestoreindicesf90_ (x=0x103863110, ptr=0x103863118, __ierr=0x7fff6345f360) at /opt/HPC/petsc-dev/src/vec/is/interface/f90-custom/zindexf90.c:29
#8  0x0000000103862d83 in isbug () at ISBugf90.F90:18
#9  0x0000000103862a3c in main ()



This can be avoided by always checking the size of the array to be restored before restoring it (uncomment l 18 and 20), but this is a bit annoying. Is this easily fixable in petsc? The C version does not have this issue.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ISBugf90.F90
Type: application/octet-stream
Size: 781 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20120314/f5c6f540/attachment.obj>
-------------- next part --------------



Blaise

-- 
Department of Mathematics and Center for Computation & Technology
Louisiana State University, Baton Rouge, LA 70803, USA
Tel. +1 (225) 578 1612, Fax  +1 (225) 578 4276 http://www.math.lsu.edu/~bourdin