Most of the devices I've seen are great by themselves. If I played with one in a store I'd want it. But as a group, when they're what my company has to target for distribution, things could get really ugly.
Practically; I was an early adopter for the HTC Magic. Despite that, all the apps designed for the G1 worked fine. I could even use Emacs in ConnectBot (ssh client) with the soft keyboard. Now that's abstraction!
When it comes to other features of the phone, e.g. camera or accelerometer, like one of the posters said- you'll have to do progressive enhancement. Intelligent web developers already know how to do this. You feature test, then you enable a particular feature. Its not all that difficult.
It works, but it means giving up a lot of control, and the results tend to show it.
I am actually an android app developer and some of the comments here make me question how many others actually are, versus relying on second-hand bullshit.
While it's slightly harder than developing for a single device, it is quite manageable. Some examples:
- different screen resolutions: this means that you must rely on UI layouts or make your pixels in your graphics code relative. This is an issue the iPhone will have in the future as well, as they WILL eventually upgrade the resolution on that screen.
- different input methods: make sure both gestures/swipes/touches and key presses both map to the correct functions
- missing features: accelerometer, cameara, etc: check and Handle gracefully or refuse to work. iPhone developers have to deal with this as well, except they can only refuse to work (by specifying minimum OS version to install).