That's fine. Dojo doesn't make this decision for you. Instead it gives you a full range of tools to choose for yourself how you want this to work.

Some UI widgets (Dijits) choose to use a client-side HTML fragment, but others do not have that requirement. Some of the work on projects like xstyle and put-selector seek to reduce this reliance further.

Dijits can be invoked either through extra markup in your base HTML page, or through a JS constructor. In the HTML templated widgets, there are attributes that bind DOM nodes to reference variables in your JS, so the widget knows where to insert content, or subwidgets, etc., and where to attach DOM events. The HTML isn't stored in JS, it's stored in an HTML file that gets combined through build tools (so you don't need to do this mix in process, it just works).

But again, that's an optional feature that's used primarily by the widget system and some of the widgets in Dojo.

I also think that using html tags is a poor practice. That said, Dojo does not require it for anything.

I use dojo daily, and I find everything to be well-written and high quality. It's the documentation that has always been a train wreck. Their idea of improving the documentation is reorganizing it on a new website.

It is modular, you only use the modules you need, and you only load the javascript you need. As for data-*, you only use that if you want, it is a feature provided by another module. If you don't like it then don't use that feature and don't load that module ('dojo/parser'). You can create widgets programatically like in jQuery (and thats what I do).

So to answer the original question:

Does anyone actually think dojo has a future compared to the power that jquery gives to make your own "widgets" but far simpler than dojo?

The answer is: yes, it has a future in a wide range of places. But in the subset of DOM manipulation, it might be more appropriate to use jQuery.