Would it be better to pass in the expected root element? It doesn't make sense that if you call Workbook("foo.tds") we wouldn't give a meaningful error. It would fail, but not with a meaningful error message.

The only way to know would be to force the user to explicitly say a Workbook should be a 'workbook', which feels redundant, or go back to checking the file extension -- which I can add, and it would actually let us optimize not opening every single file in the zip, and only open *.tds *.twb files

I think I've gotta scrap this approach and take a step back.

With the commit in its current state you can't rename a twbx to a tdsx and have it work, but you can load Workbook(TDSX_FILE) because all the xfile stuff doesn't have context for what root tag to look for. I can plumb that through, not sure how it feels yet.

@RussTheAerialist I think you should move open_xml into xfile and then we can ignore this bug for 0.2.

Then we can chat about a cleaner approach to making sure Workbook is actually getting a workbook file

I'm okay with that. My thought was that when you invoked it from Workbook or Datasource, you would just put in the root tag you expected. That way xfile remains generic and uncoupled to the concept of Workbook/Datasource, and workbook/datasource can specify the root tag to check (and only check if specified), same for file extension, I suppose.

Yeah, I like the idea of xml_open passing that argument, once you merge it I'll rebase and give it a go (I've got a long flight anyway)

Replaced by #62