Lisandro, The memory leak occurs for all in-place MatConvert(). I suspect your changes might affect MatHeaderReplace() used in the in-place MatConvert(). I have trouble figuring out what causes it. You might have some clue about it.
Hong On Sat, 6 Oct 2007, Satish Balay wrote: > On Fri, 5 Oct 2007, Lisandro Dalcin wrote: > > > And please, send me the failures... > > Lisandro, > > I haven't yet checked the nightly builds properly. Some of the issues > I've been checking were with code already in petsc-dev. [So the > current checks are from the tests I ran on my laptop] > > Most of the 'strict aliasing' I've seen so far are compile issues with > sieve and other external pacakges[spooles,superlu_dist,pmetis.c etc..] > wrt PetscObject usage. > > [so the fix for PetscObject stuff was simple enough for met to do and > test]. I'm attaching the patch of the changes that worked. Let me know > if this is ok to commit and push. > > There is also the following memory leak - which I haven't checked > yet.. Perhaps there is some incorrect reference counting somewhere. > > >>>>>>>>>>>>>>>>>> > testexamples_C in: /home/balay/tmp/petsc-dalcinl/src/mat/examples/tests > 0a1,10 > > [0]Total space allocated 1380 bytes > > [ 0]460 bytes MatCreate() line 80 in src/mat/utils/gcreate.c > > [0] MatConvert_SeqBAIJ_SeqSBAIJ() line 276 in > > src/mat/impls/sbaij/seq/aijsbaij.c > > [0] MatConvert() line 3142 in src/mat/interface/matrix.c > > [ 0]460 bytes MatCreate() line 80 in src/mat/utils/gcreate.c > > [0] MatDuplicate_SeqBAIJ() line 2448 in src/mat/impls/baij/seq/baij.c > > [0] MatConvert() line 3142 in src/mat/interface/matrix.c > > [ 0]460 bytes MatCreate() line 80 in src/mat/utils/gcreate.c > > [0] MatConvert_SeqBAIJ_SeqAIJ() line 17 in > > src/mat/impls/baij/seq/aijbaij.c > > [0] MatConvert() line 3142 in src/mat/interface/matrix.c > Possible problem with ex55_1, diffs above > 52a53,55 > > [0]Total space allocated 460 bytes > > [ 0]460 bytes MatCreate() line 80 in src/mat/utils/gcreate.c > > [0] MatCreateSeqAIJ() line 2777 in src/mat/impls/aij/seq/aij.c > Possible problem with ex114_1, diffs above > <<<<<<<<<<<<<<< > > Prometheus also nees some PetscObject changes - I can send Mark the > changes to update the prometheus tarball later.. > > thanks, > Satish
