sort the prefixes by popularity

[VladimirAlexiev]

Run this several times

perl -MRDF::NS -e "print RDF::NS->new(1)->PREFIX('http://www.w3.org/2001/XMLSchema#')"

and you get different answers, eg xs xds etc.

The doc says "If multiple prefixes are defined, it is not determinstic which one is returned!" But I wonder why it has to be so:

If you ask http://prefix.cc about http://www.w3.org/2001/XMLSchema# it confidently answers xsd. It has these voting buttons so worse alternatives like xs xds are not returned. Can't the module use this voting info?

[VladimirAlexiev]

http://prefix.cc/popular/all.txt seems to return them in the right order:

• for http://www.w3.org/2001/XMLSchema#: xsd, xs, xds, xmls (the last two are wrong)
• crm is the official namespace, ecrm is erlangen, etc.

[kjetilk]

I suppose it comes down to an implementation detail, that the mapping is represented internally as a hash, and therefore unordered. So, it is probably just a matter of having time to change it so that there is a an array of prefixes... I tend to agree, this behaviour is rather odd, but someone has to set aside time to do it. :-)

[nichtich]

The current release returns them in sorted order - I don't mind changing the order to something else. This only affects reverse lookup which is less relevant because every RDF serialization should declare its prefix-namespace mapping anyway.

[VladimirAlexiev]

Version 20150725 returned xmls.
Version 20170111 returns xds. I guess by "sorted order" you mean "alphabetically sorted": but that's not good, it should be "sorted by popularity" (i.e. as in http://prefix.cc/popular/all.txt)

every RDF serialization should declare its prefix-namespace mapping

If your module does reverse lookup in prefix.cc, please do it right. Else remove this function.

[kjetilk]

Yeah, I agree that sorted by popularity would be a better solution, but I suppose that's a different issue?

[VladimirAlexiev]

The problem is that the reverse lookup returns BAD prefixes (eg xds) for known namespaces (eg xsd).
Now the issue title corresponds to what we're discussing. Please reopen it.

[nichtich]

This feature requires an extension of the internal database and the update mechanism for reverse lookup. Each prefix could get an additional popularity ranking (1,2...n) among all prefixes that share the same namespace (additional column in prefix.cc). Updates will not only cover changes of prefix-namespace-mappings but also changes in popularity. The main work is in updating the UPDATE method but this is not on the top of my TODO list, so pull requests are welcome.

By the way RDF::SN uses a different sort order to prefer shorter prefixes.