I am wrong here. I misunderstood the question, kind of. See #9 for my new answer.
I understand, but first, it’s SIDREF_array, and not IDREFS (there is an IDREF like type in the schemas but it’s not referenced.) Normally the reference to the <source> element would be an xs:anyURI type, which generally has open semantics, but for whatever reason the <source> reference of <input> is the custom type that restricts the URI-like reference to just a fragment, so that it must be local to the document. Presumably the requirement to use # is there so that a future revision/version might extend it to xs:anyURI. I don’t know the rational for this.
I wouldn’t worry about how to treat an array with more than one SIDREF. Frankly it’s not really necessary that the contents of the source are restricted to “_array” types, but the manual says it is, and I don’t know but the menu of types is probably limited to the types supplied by the schema.
That said, and assuming I am correct in my understanding, then the only real concern is that other software is unlikely to know what to do with such an input. But there is precedent for such handling in the <newparam> type, which can include <SIDREF> as a variant option. It’s very possible that there is no other use for a <SIDREF_array> based <source> but the only way to know would be to scour the manual exhaustively. What I’m trying to say, is this use may be why it is provided in the first place.
COLLADA isn’t “broken” so much as left unfinished. It looks like there is a lot of support for it, but it’s more like it was a fad for a while, so a lot of people worked on it enthusiastically, but then the bottom fell out. It gives the impression that it’s well supported, but it isn’t. In theory it might have good support under Maya and 3DSMax, but I doubt it. Still, files that come out of those tend to be better than the average application’s. Problem with that software is it’s out of reach for regular people, and so there are solid grounds for avoiding that software out of principle.
But you are probably right. I could find no alternatives either last year. So I ended up spending all of 2016 to make COLLADA a viable technology. I didn’t plan to, but the XML design makes it very heavy lifting. That’s part of the reason it’s failed to take hold. Another part is AutoDesk bought every commercial 3-D house, so the original thought behind COLLADA that it would help to be a bridge between these houses is kind of moot after there is only one house. Crazy hyper capitalist age we live in. Right now I think all interest in COLLADA is going to come from noncommercial interests, which for 3-D is kind of a new development.
I think “SAX” is a very awkward technique. Programmers are by nature lazy and shortsighted, so they lean on it. But I find its backward logic tedious and counterproductive. I am busy rewriting the original COLLADA-DOM library, with a space age approach to XML for C++ that I hope will ingratiate COLLADA to the likes of Blender. Which even though it long dropped official support for COLLADA still has it as its default export/import target. Which goes to show you, there’s no alternative to it. So we just either have to finish it or invent something from wholecloth. At this stage the former just makes more sense. Especially since we don’t have a lot of experience with this sort of thing. That isCOLLADA itself is experimental/groundbreaking futuristic technology.
When WebGL was brand new/experimental I developed a custom JSON loader that isn’t SAX based for my own use. There’s a design document here (http://en.swordofmoonlight.org/wiki/JSOM/object_reference) just to see what such a thing looks like. I only learned about glTF when I started investigating COLLADA. It didn’t exist at the time naturally, and it’s still relatively new. On the WWW glTF may make much more sense than COLLADA even though I think it’s unproven. All of the existing standards are heavily OpenGL based, and glTF and COLLADA are really two different problem domains.
I hope we are on the cusp of a COLLADA renaissance. I am thrusting myself in the middle of it. Slowly worming my way into the possibility of reviving it. It’s not the most brilliant thing in the world. But it got a lot of publicity and so it seems like the best candidate for the basis of an open 3-D standard. We can use all hands on deck