On Wed, Dec 10, 2014 at 2:54 PM, Jan Blechta <[email protected]> wrote: > On Wed, 10 Dec 2014 09:57:10 +0000 > Anders Logg <[email protected]> wrote: > >> This is something I haven't figured out. It would be good to have a >> discussion on this. >> >> One option might be to just keep all of FEniCS with dependencies in >> HashDist (using the fenics-install.sh script). >> >> Then for all FEniCS Python components, one sets up a separate >> directory for installation during development, in combination with a >> config file that can be sourced to put that directory first in the >> path. >> >> For DOLFIN development, one would source the hashdist config and work >> with cmake.local. >> >> But there are likely better ways to handle this using suitable >> hashdist commands. Even if the install script does not install >> hashdist, I assume it will be a good thing for developers to actually >> install hashdist so that they may talk to hashdist outside of running >> the install script. > > So couldn't the script use installed hashdist and hashstack if present > instead of installing fresh clones to /tmp?
Yes, I agree. I think it could be useful for many to keep hashdist and hashstack also after the build, such that one can easily add more packages that are not in the default profiles. Another bad thing about having this in /tmp, is that the profile /tmp/fenics-install.XXXXXX/hashstack/default is lost when you restart the machine. This profile is what the garbage collector in hashdist tracks, so when this is removed, "hit gc" will remove everything. Johannes _______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
