[petsc-dev] did someone mess up parts of TS?

Jed Brown jed at 59A2.org
Thu Sep 30 09:09:48 CDT 2010


This is caused by the MatAXPY updates (TS uses MatAXPY more than
elsewhere), but isn't Shri's fault.  MatHeaderReplace(A,B) did not
preserve the reference count of A.  I fixed this in c41b34d038d0.  I'm
not sure if this should be backported to release, but MatHeaderReplace
is used in a few places and I have to expect that all uses are
erroneous.  i would also lean towards requiring that B have a
reference count of exactly one, because otherwise calling this
function necessarily leaves a hanging reference.

Jed



More information about the petsc-dev mailing list