alternative for %r mentioned in FAQ

[baallan]

I have a suggestion if the %r mentioned in the FAQ is still open for discussion.
I have implemented on the application end a generalization of the idea of %r.
Let attribute values contain additional nodename format strings of the form %NT.
N is an integer value 0-9. At present, even a cray does not use more than 10 segments in a name.
T is a segment type
%NT is replaced by the Nth segment of type T appearing in the node name.

In my initial implementation the supported values of T are 'd' (integer substrings), 'm' (substrings separated by '-'), and 'u' (substrings separated by '_'). Note that multiple separators of the - or _ types are collapsed to one, so '--' is equivalent to '-' in this scheme.

E.g.

node-gw23 fastnet=node-gw%0d-ib 
node-gw23 otherthing=ot-%1m

yields

node-gw23 fastnet=node-gw23-ib 
node-gw23 otherthing=ot-gw23

This notation allows collapsing definitions in a manner similar to the FAQ %r idea, but with more control.

The exact semantics are given by a bash script below. Equivalent c/c++ are not hard to cook up (I have a c++ version).

gender_substitute () {
        ghost=$1
        val=$2
        if test -z $2; then
                echo ""
                return
        fi
        ilist=$(echo $ghost | sed -e 's/[-_a-zA-Z]*/ /g')
        wlist=$(echo $ghost | sed -e 's/-/ /g')
        ulist=$(echo $ghost | sed -e 's/_/ /g')
        vnew=$val
        n=0
        for i in $ilist; do
                sub="%${n}d"
                vnew=$(echo ${vnew//$sub/$i})
                ((n++))
        done
        n=0
        for w in $wlist; do
                sub="%${n}m"
                vnew=$(echo ${vnew//$sub/$w})
                ((n++))
        done
        n=0
        for u in $ulist; do
                sub="%${n}u"
                vnew=$(echo ${vnew//$sub/$u})
                ((n++))
        done
        echo $vnew
}