For fragments, there is this: https://github.com/blazegraph/BlazegraphBasedTPFServer

We may want to look into supporting this.

the node.js version of the TPF server is actually why I created this issue.

My concept of the fragment server was that it could decentralize a big dataset by distributing data fragments to it with selectors, so I am not exactly sure how implementing a Blazegraph TPFServer is in the overall WDQS scope. It could help perhaps for decentralizing your data internally, or for caching named graphs, though it would probably not by itself facilitate using WDQS as a SPARQL datasource for external TPF implementations.

it seems that with a CONSTRUCT query, sending an Accept: text/turtle works.

this is actually all that is required to get a linked data fragment from the SPARQL interface.

The problem with TPF access to the WDQS SPARQL interface is that the very simple query (required for the metadataCallback)

SELECT (COUNT(*) AS ?count) WHERE {?s ?p ?o}

cannot return turtle. Also, there seems to be a minor difference with the COUNT implementation in OpenVirtuoso, that allows a count to be unbound like:

SELECT COUNT(*) WHERE {?s ?p ?o}

I have worked around the counting problem. The experimental TPF Server is here:
http://orbeon-bb.wmflabs.org/

This wikidata datasource uses SPARQL interface at http://query.wikidata.org/sparql

I think that this issue can be closed.