Fms is a replacement fabric mapper for both GM and MX myrinet interconnect networks. Along with mapping, it also gives a fair amount of valuable debugging information when something in the network is misbehaving, and is therefore recommended over the mx/gm_mapper. This relies upon the gm package submitted in bug 184355. If gm goes to sys-cluster, this package should as well. Reproducible: Always
Created attachment 124056 [details] fms-1.3.0_rc1.ebuild
Created attachment 124057 [details] fma-init.d
Created attachment 124058 [details] fms-conf.d
Created attachment 124059 [details] fms-init.d
Created attachment 124061 [details] fms-seperate-productinfo_dir.patch Patch allows us to move the productinfo directory to /usr/share/fms/productinfo. These files contain configuration options for fms relating to how it talks to various Myrinet Hardware and will not typically be modified by the user.
Looks good, small comments as usual: ebuild: einfo "the use of most fm_* commands without specifiying the einfo server using -s." "the einfo server"? What are the choices of server here? I don't particularly like the idea of suggesting the user to use EXTRA_ECONF under normal circumstances. eerror "fms cannot handle gm and mx" eerror "fms needs to know which fabric to use" These would be better as more descriptive errors. The first one should state that you can't have both USE flags set. The second one should say which use flags the user has as options, pointing out that only one can be set. make DESTDIR="${D}" install || die "make install failed" Please use emake..install in new ebuilds (unless it doesn't work that way, in which case accompany the make command with a comment) dodir /var/run/fms keepdir /var/run/fms keepdir implies dodir, no need for both fma init script: hostname=`hostname -s` please use $(..) rather than `..` (just a convention around here, plus you can nest the former) (this occurs in another place as well) if [ -z "FMS_SERVER" ]; then missing ${ } ? eerror "FMS_SERVER is not set." error message should give a hint how to correct it (i.e. ask them to edit the conf.d file) if [[ $USE_CHECK_TORQUE -ne 0 ]]; then please use ${foo} rather than $foo (just another convention, but you can also do stuff like ${foo}bar) check_torque if [[ $? -eq 0 ]]; then eend 0 return 0 fi can be slightly simplified: if check_torque; then eend 0 return 0 fi fms init script: if [ -z "$(which fm_switch)" ]; then better way that doesn't rely on which: if ! type fm_switch >/dev/null 2>&1; then type is a shell builtin, see the bash man page All your fm_switch invokations go off into the background. It is not guaranteed that they have even started to run by the time s-s-d starts /usr/bin/fms. Is that OK? eend $? return $? No need for that final return, it's implicit (eend returns the same value that is given in its first parameter, and bash uses $? as a default return value)
This package does not meet the current criteria needed to be a viable candidate for the Portage tree since the upstream sources are no longer available at the location provided by the ebuild. This criteria is currently listed at the following URL: https://wiki.gentoo.org/wiki/Project:Proxy_Maintainers/Maintainer_Wanted Metric: Upstream status: Exists; "Package must still be available and fetchable (did it move to github, bitbucket, ...?)" If you (or another user) are still wanting this package to make it into the Portage tree please let us know by re-opening this bug and include a new usable upstream source. Sadly, until that time we cannot move forward with this ebuild. Tom D
