I expect Apple will solve this problem in their usual way; pick slightly less stupid settings and lock those in. In this case, the difference will be stunning and people will marvel at how those apple tv's can look so good.
Hrrm ... that was RMS's point, way back when, with some printer, if memory serves ...
As someone who's dealt with the inimitable joy of trying to get an MBP to work with an HDTV, Apple's "usual way" appears to be to code to the standards and to hell with anyone who breaks them (or the users stuck with non-compliant products). This appears to apply to wifi as well. There's been an open bug in OSX for years involving OSX assgining its own DHCP lease when it fails to negotiate one with the router. This results in the dreaded "Self-assigned IP" message, which is nigh-on impossible to rid yourself of short of voodoo dolls and wifi dances.
Getting a bit off-topic here, but that's not a bug: it's assigning itself a valid zeroconf* address because the DHCP server is not responding.
I think bug in this case just means that the software does not work as intended even though it follows the spec.
Also, in my experience Apple's DHCP agent will re-request an IP address after having assigned itself a self-assigned IP address. This generally takes about a minute or so, in that time the DHCP server can then reply once again. I've never had issues with this at all.
Zeroconf is a historical mistake. It shouldn't be there any more; it does far more harm than good.
But getting back to the original complaint, if not for zeroconf you'd have no IP address at all; I don't see how that's any better.
I'll admit to having only a rudimentary knowledge of how wireless networks work, but it's better than 99% of users and I find OSX frustrating (in this regard). At the end of the day, the user doesn't care whether the router manufacturer isn't following the spec, or whether Apple's implementation is buggy. The simple fact is that it works on Windows but not on OSX, and that's a failure on Apple's part.
What Windows does do wrong is that it then also drops the self-assigned IP address which may already be in use for communication, this can cause issues with other hosts that are communicating with it over zeroconf.
You are correct for Ad-Hoc networks.
In RMS's worldview, software freedom is an inherent good, not a derivative good. "Many eyes make all bug shallow" is nice, but for him it's a side-benefit.