On Sat, Mar 26, 2011 at 21:20, Barry Smith <bsmith at mcs.anl.gov> wrote:
> I thought of that. But it takes seconds to build and it could be a later > run did come after some change to the files so see no harm in rebuilding. If > setting up the locking mechanism is as messy as doing them in parallel we > might as well do them in parallel. Seems to me that we should write dependencies for those sources and rebuild exactly when they are out of date. One option is to generate them out of the source tree and then move them in atomically. Then it won't matter if concurrent processes occasionally duplicate a small amount of the same work. Note that we pay significantly more to recompile those sources than to generate the fortran stubs. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110326/be7125fa/attachment.html>
