BTW: What exact error message do you get? I've used symlinks for PETSC_DIR before - and don't remember any issues. [although I've never moved the install after things got built]
Satish On Thu, 26 Nov 2009, Satish Balay wrote: > peraps this is one case where --prefix install is useful? > > Satish > > On Thu, 26 Nov 2009, Jed Brown wrote: > > > I was recently building PETSc on a system where /home is a slow NFS > > mount so of course I wanted to build on /scratch. However, since I want > > other nodes to use it, PETSC_DIR should really be in /home. My initial > > thought was to symlink from /home to /scratch, then move the directory > > after compilation of several configurations. This does not work due to > > os.path.isdir returning false for symlinks to directories, and the use > > of os.path.realpath to confirm that PETSC_DIR is the current directory. > > > > Would it be reasonable to relax these checks to merely confirm that > > PETSC_DIR *points* to the current directory, or will that cause more > > confusion than it's worth? I wouldn't expect people to set up such > > symlinks unless they knew the score. > > > > Jed > > > >
