On Thu, Dec 2, 2010 at 23:19, Matthew Knepley <knepley at gmail.com> wrote:
> How exactly does that happen? I have just checked that I ran multiple > configures and did not clobber those logs. > When a new configure process starts, it truncates the file. You only relocate it later. It's a race condition, so may not happen every time, or you might not have correct contents in each of your files. > I do not think this works on NTFS. > NTFS is lame and does not have a way to atomic rename when the destination file exists. But it does have both hardlinks and symlinks. In the proposed model $PETSC_DIR/configure.log is only a convenience for the user, it is not being read or written concurrently, so the fact that redirecting the symlink cannot be done atomically should not be a problem (racing a human is safer than racing another process). Jed -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20101202/56509e4a/attachment.html>
