Hi Ken, On 19/08/2012 14:31, Kenneth Hoste wrote: > In the latter case, I don't see how "module install" picks up the .eb files > found with "module search"... > > I must be overlooking something obvious.
ah, yes, now I understand what you missed: "my copy-pasting?" LOL ;-) but, seriously some algorithm could pick up a default choice and try that; my thinking about it is, that this doesn't have a clear-cut optimal answer since on one hand we can prioritize items based on directories searchpath (like in PATH, MODULEPATH) but on the other hand, we will stumble across the very same problem that these guys have met: http://www.mancoosi.org/cudf/primer/ """ versions are positive integers, usual version strings like "1.2.3-4" are not accepted (as they have no clear cross-distribution semantics). Ideally, each set of versions in a given distribution has a total order and can then be easily mapped to positive integers. """ ie. if all of (easybuild, pkgsrc, portage, macports, ports, rpm, srpm, tgz) were able to provide the latest zlib v1.2.7, it would not be at all clear which one is the optimal one and should be picked when asked. This is why the CUDF guys used integers for versions: they force the suppliers of the CUDF bits, to do some ordering & turn versions into an enumerable set. If somebody has some clueful idea on how to handle this, as regards the easybuild/modules case, well, it is a good moment to spit it out! Perhaps, the best option is to just provide an external "enumerating function" for now, to outsource the academic problem to the academics... It's important to realize that this idea should stay on "experimental" status, since it might go against the reproducibility great argument in favor of EB! -- echo "sysadmin know better bash than english" | sed s/min/mins/ \ | sed 's/better bash/bash better/' # Yelling in a CERN forum
