Speaking as the author of that gist, “scaling” is … kind of a weird tag for this.
I shared it because I find linear-time sorts fascinating in general, and I haven’t seen this one documented anywhere. A more traditional counting sort or radix sort is likely to be faster, unless the data set is fairly small. The implementation is quite simple, though.
Interesting. In principle you should be also able to build an arbitrary number of indexes of the same array. Say for instance that the array is populated with tuple, you could build a sorting for each field. Like the option -k of sort. Just pass a “key” parameter (callback) and you are done :)