On Nov 3, 2010, at 11:39 AM, Jed Brown wrote: > [moving to petsc-dev] > > On Wed, Nov 3, 2010 at 10:54, Barry Smith <bsmith at mcs.anl.gov> wrote: > Right, this is very bad. We set the r2 and r3 up front so the compiler won't > complain about possible not set values? > > I don't know, you wrote that code. I stripped it out of a local build and I > don't get any warnings in optimized mode (with gcc or clang). I think those > warnings were a compiler bug, unless you only saw them with-errorchecking=0.
Because of the || *r2 may never get set, but it gets used later in the routine. Maybe compilers shouldn't give the warning but some did. Barry > > I guess we need to not use PetscMallocN() for structs and values in the > structs. > > Maybe, but only if we really need to set r2 up front. > > Jed
