I'm not sure yt.physics is even appropriate, yt doesn't do physics, strictly
I don't anticipate that changing in the near future.
I think the best way to reorg is by hierarchical function, which is more or less what you've described in the google doc. Under yt. the dirs should be simple, like yt.data_io, yt.analysis, yt.plotting, yt.gui. And under those more refined, yt.analysis.stars, yt.analysis.haloes, yt.analysis.fields. Well, this is just my two cents. I'm just writing my thoughts down, I haven't really contemplated how hard it would be to reorg this way.
Perhaps that's the best way to go -- yt.analysis subpackage? Does anyone else have any ideas?
My suggestion is if we are to reorganize the directories, we should do it all at once, meaning it should coincide with a point release of yt, to keep the distributions (svn, hg) more or less comparable. Otherwise we'll just go crazy trying to juggle the two.
I can't imagine doing it any other way. One of the reasons I've been sitting on this document for a year is that there hadn't been a good time. I see this coinciding with pushing toward a 2.0 -- which would need to include first-class time series analysis, the software volume renderer (possibly a hardware one as well), the VTK stuff, the Parallel HOP, embedded support described in the docs, the new particle IO, the reorg, possibly some other stuff that hasn't been done yet. Anybody else have any thoughts on reorg? -Matt