Are you thinking that these interfaces would be real (potentially freezable) XPCOM interfaces, or C++ pseudo-interfaces that would be bound to particular GRE versions?

The device-layer (win32GDI, X, bitmap, PS, whatever) would be almost impossible to XPCOMify in a performant manner. But the "shell" isn't hard, and the document (DOM) can already be created without a presentation... how hard would multiple presentations be?

In my implementation the presentation is a real XPCOM interfaces. Nothing here *has* to be xpcom-ified, as long as the abstraction is easy to use internally--but nsIDeviceContext is already xpcom-ified, and that's our device abstraction. There will probably be pseudo-interfaces as well for non-frozen stuff, but in the main I want a frozen interface, which is why I am asking now.

The device-layer is not hard to do, it's just that you need specific interfaces that let you create an nsIDeviceContext--which is the abstraction for the device. I have internal interfaces that create one of these presentations against an nsIDeviceContext for my new printing architecture.

If the shell isn't hard, I'd love to hear how to do it :) I need a scripting interface and docshell/webshell, without a window preferably. Multiple presentations are "easy," I'm already doing that. It's my impression that our current world is just not designed for the shell separation yet but a little work could make it so. If we can solve this problem we can have Mozilla print documents from the command line and do several other things that are not currently possible.

Well, (doc|web)shell have several functions that are currently intertwined, but separating them has some significant benefits for simplicity's sake as well as for windowless presentations like printing.

1) history management (and URL loading)
2) network loading management (this can mostly be hidden, I hope)
3) focus management (which really needs to be rewritten)
4) random attributes (flags whether to load subframes, plugins, objects, scripts, follow redirects, follow meta refreshes)

if these can be split up, most of the printing/static viewing stuff can implement these interfaces in trivial ways (no hx management, no JS, no focus, no scripting).