Yeah, the image/canvas world is indeed in limbo. :-( You're right in that HTML5 isn't final, but overall I think it's pretty stable, especially with respect to canvas. For a year or more now, there are three existing canvas implementations in stable, shipped browsers: namely Safari, Firefox and Opera. There's enough deployment that makes breaking backward compatability highly unlikely.
It might seem like re-inventing the wheel to implement a Gears canvas, given these other implementations, but the motivation for the Gears canvas was to fill in the gaping hole (i.e. Internet Explorer) in the list of canvas-enabled browsers, and to also provide Gears-isms like canvas import from / export to blob, so you can do things like canvas renderings in a worker thread. Given that, I think that canvas is still the better API, even though you're probably right in that the imaging API is probably easier and simpler. As for a list of action items (i.e. features) to implement, I think it's simply "whatever bits of the HTML5 canvas spec are missing". There are plenty of "Unimplemented"s in canvas_rendering_context_2d.cc. A nice test suite under gears/test/ manual/canvas would also help. The text-related and shadow-related API are probably more a nice-to-have than a must-have-for-launch, and I'd drop the API additions in canvas_rendering_context_2d.cc (e.g. convolutionTransform, medianFilter, adjustHue, etc.) since they're unproven, and not necessary for an initial launch. As for what is required to get this released, I think if a good- enough, self-consistent feature set is implemented (and thoroughly tested) then I would argue for release. The main problem at the moment is simply that nobody is working on completing the feature set. If you are willing to write the code, then I will be willing to do the code reviews, and to land your patches. http://code.google.com/p/gears/wiki/ContributingToGears
