Yeah, this makes a lot of sense to have. What about this approach?

Each link between a Wikibase entity and an article on some wiki currently consists of entity id, site id and page title. I propose we group the site ids (ie { "wp": ["enwp", "dewp"], "incubator": ["foo", "bar"] } ) and then have the client wikis specify for which group(s) they want to get the links. So each Wikipedia would then just have a single setting in their LocalSettings file ensuring they only get actual Wikipedia links. And it would be easy to allow overriding this behaviour on article basis if desired.

I was planning on creating the API mode for obtaining the links in one of the next days, so we should try to decide on this soonish :)

I don't know much yet about Wikidata's structure, and I don't know what has been decided so far, but I'd go for the simplest approach (concerning Incubator): allow language wikis in any language code, and if that wiki does not exist, make that go to Incubator.
Something like:
site id = dewp -> de.wikipedia.org
site id = grcwp -> incubator.wikimedia.org/wiki/Wp/grc

(In reply to comment #2)

Something like:
site id = dewp -> de.wikipedia.org
site id = grcwp -> incubator.wikimedia.org/wiki/Wp/grc

This kind of fallback seems unrelated to what this bug is about. If you think we should have something like this, It'd be better to create a separate feature request.

I think it is not a good idea, to make commons, wikispecies and incubator simple interwikis which may not be schown, as they are not part of the wikipedia project but projects of their own.

As other projects like Wikiquote and Wikisource have their own interwiki-lists and may want to refer by default to commons too, every project should have its own Section I think.

All these project have langlinks that link to wikipedias. Bots currently update langlinks on these projects if langlinks on the wikipedia target articles have changed.

If they would have their own interwiki lists we need a bot that keeps these list uptodate based on corresponding wikipedia langlist.

There are two different usecases for sitelinks; one for interlanguage links between Wikipedias, and one as a means to identify the item to get to the structured content. The alter is typically to create the inforbox about an external entity. Its nothing that really stops any project and/or site from being included, its only business logic for how to identify the groups that should be included, and sites that of some reson should be excluded. For now all sites in the group "Wikipedia" are included, it could very well be that commons, wikispecies and incubator should be included.

The only thing that seems somewhat clear is; the present usecases for sitelinks are about linking pages that describe the same entities, linking pages that describes different entities will not be supported. That does not mean that pages at those sites can not include the sitelinks generated by Wikidata on individual pages.

Perhaps give some examples. A page at Commons with pictures from Oslo (http://commons.wikimedia.org/wiki/Oslo), and articles at Wikipedia about Oslo (http://en.wikipedia.org/wiki/Oslo) is basically about the same entity. Pictures from within Oslo (http://commons.wikimedia.org/wiki/File:Holmenkollen.jpg) can be more specific than the entity Wikidata is about. A page about Oslo from a Wikipedia at Incubator should not be a problem as long as it is about the same entity.

The same can be said about Atlantic Cod for example (http://en.wikipedia.org/wiki/Atlantic_cod, image from Commons http://commons.wikimedia.org/wiki/File:Atlantic-cod-1.jpg but no page, there is a page at Wikispecies http://species.wikimedia.org/wiki/Gadus_morhua (Atlsntic Cod)), or Zebra (http://en.wikipedia.org/wiki/Zebra, http://commons.wikimedia.org/wiki/Zebra, http://species.wikimedia.org/wiki/Equus_quagga (really plains zebra)).

The common thing is to describe the same entity, not that the page is at "Wikipedia". But also this makes it difficult to define hard limits. An article at Wikinews could be about the same entity as a page on Wikipedia, but it would happen very seldom.

To clarify, there are two aspects to consider:

1. which sites should be in the "wikipedia" site group on wikidata? Answer: from the experience with interwiki bots, these need to be all sites that have interlanguage links that point to Wikipedia. This includes Commons, Wikispecies and Incubator. Which sites are in this group on Wikidata is determined by the sites table of the Wikidata wiki.

2. Which sites should be shown as interlanguage links on a given client wiki, i.e. on a Wikipedia page corresponding to an item? This should, I think, be determined by the sites table of the *client* wiki. This would typically not include Commons, Wikispecies or Incubator.

So, interlanguage links on a wikipedia will be shown to all wikis that are:
a) in the Wikipedia group of Wikidata's sites table
b) marked for interlanguage treatment in the Wikipedia's sites table.

I think this would resolve this issue cleanly.

Incubator is different from Commons and Wikispecies. Incubator contains many language versions of the same topic, while Commons and Wikispecies contain one (supposedly multilingual) page about a unique topic. That is both multilingual, but in a different way.

Unlike Commons and Wikispecies, Incubator should not exist in Wikidata. Instead, it should be possible to link to Incubator pages as being a page in that language version of Wikipedia (or Wikiquote/...) and not as being an Incubator page.
It is already this way: it is possible to add interlanguage-links to any language in the "langlist" even though the subdomain does not exist (using e.g. [[ace:Page]] on a Wikiquote will go to the Aceh Wikiquote on Incubator). I would like to keep it this way, or even extend it to languages known to MediaWiki (or maybe even all languages in CLDR or ISO 639).

As for Commons and Wikispecies, they could perhaps be shown as interlanguage links since they are a multilingual version of a certain topic, or Wikidata could be a used as a solution for bug 708, in which case Commons and Wikispecies could be shown in the sidebar as interproject links. These are the two possible approaches, I don't know what's the best one. In the second case, Incubator could be part of them if you use it to link articles about a certain language to its Incubator test wiki (i.e. like https://en.wikipedia.org/wiki/Template:Incubator).

I'm going to close this. We have the ability to define which language link groups get shown on the projects that Wikidata supports.