I think it's a nice idea, but seems difficult in practice. First, I think embedding language knowledge into the VCS is fraught with peril. For one, does that m

I think it's a nice idea, but seems difficult in practice. First, I think embedding language knowledge into the VCS is fraught with peril. For one, does that mean that you need to rev your VCS version every time your language changes? What about when your language revs its AST, but not the language itself? Is your VCS version now no longer backwards compatible with old versions? Second, I think there's a significant amount of overhead and new technology here. Most DVCS's currently use hash-based filesystems for storing history. If you replace simple data diffs with semantic transformations then you have to find some portable way of encoding that. If you don't want the implementation to be language-specific than you have to find some language-agnostic encoding system that can also recognize that the textual diff and the alpha-rename are identical commits. IMHO, I would rather have metadata on commits. That way you can always fall back to plain text and all the old tools (like an ancient vi) continue to still be usable, but more advanced language-specific tooling could recognize these things and provide a simple view to the user.