Right - or if a flag is set, to ignore the dataset header info and assume it's all "orig" view. Heck, one could imagine that, since AC-PC mode is rarely used anymore, that we have "orig", "unconstrained" and "model-space" as the options there.
Now, of course, the real solution is to have anyone working with NIFTI really keep track of this info. Shouldn't things like MNI-space atlases be distributed with sform codes of 4 (or at the very least 3) rather than 2 (or 0)? While at it, shouldn't any program that is applying a set of transformations and writing NIFTI let you (or force you) to, at that point, specify the space the output image is in? It's great to have transformations built into the header, but honestly, I don't see that used much. Really using the sform code to say what space the image is in though is a great bit of housekeeping. As it stands, though, I find myself running nifti_tool and batch changing the qform and sform codes to either 0 or 3 on everything I'm working on (which one I choose is often a coin toss) just so I stop banging into conflicts (in effect, akin to chown -R a+rwx dirname). Sigh - sorry for the rant - time to grab some coffee.
Craig