Well, I mean, the author goes on to make a bunch of salient points like the difficulty of diffs of nested directories (solved in [1] but not widely implemented)

Well, I mean, the author goes on to make a bunch of salient points like the difficulty of diffs of nested directories (solved in [1] but not widely implemented) and saving all of the binary blobs in every checkout (solved by [2] which has not "won out"). I think the big thing with DVCSes is that you can pretend you have a client-server model with a handful of directories on your local machine. At one point at my present job I replayed a bunch of SVN history through Hg so that I could by-hand divide the work that had been done into a few named branches; this helped me to figure out where things had "gone wrong" in the project. It was really effective to just have a day of SVN update, hg diff, copy files to ../branch_name, commit to hg, rinse, repeat. What I really needed was indeed the "killer feature" that he's saying -- sane branching and merging -- but the fact that it was all easily contained in my filesystem was a nice plus. [1] http://www.cs.utexas.edu/~ecprice/papers/confluent_swat.pdf http://www.cs.utexas.edu/~ecprice/papers/confluent_swat.pdf [2] http://darcs.net/Internals/CacheSystem#lazy-repositories-and-the-cache-system http://darcs.net/Internals/CacheSystem#lazy-repositories-and...