We should provide a set of test data for people to play with. This should include:
EDIT: struck through as these are out of date....
These should come with instructions about how to import these, and some configuration snippets needed for setting up the wiki in a way that is compatible with the dumps (items in the main namespace, etc)
I would make this a maintenance script that pulls the items via the wikidata api.
if there are id conflicts for the properties, they can be renumbered and any usages of the properties (or referenced items) can be remapped during "import" (via api)
actually have done this to have all wikidata properties in my dev wiki (and I think some items), although it was a hacky script and should be made into something nice.
and sites table contents should be available via api, although populateSitesTable script generally works fine as well to replicate the contents of wikidata's sites table.
For generating the dump, it would be nice to have a script that can generate a dump based on a list of entity IDs, automatically pulling in all referenced entities (one level, not recursively).
@aude Is it possible to get that script you wrote? A hacky version that works is better than nothing so far.
https://github.com/filbertkm/WikibaseImport is an initial version so far.
my approach is to fetch a list of properties from the wikidata (or other wikibase api) and import them via maintenance script. property and item ids get renumbered (as needed) and tracked, so links from one entity to another can work.
it doesn't import site links yet (because badge items need to be imported) nor stuff like qualifiers.
and i'll be making the code a bit nicer and add tests. also, want to put this on gerrit.
it might be possible to make a xml or json dump mode, but not sure it's necessary yet.