Trying Git SVN

I’ve been experimenting with a Mercurial-to-Subversion workflow using hgsubversion. In brief, I have our Subversion repository checked out using that tool, and I work in a clone of that Hg repository. It works, but there are two main negatives.

The first is a consequence of the tool rewriting history. The tool works great at keeping you in sync with Subversion as you are fixing bugs (each Subversion checkin corresponds to a Mercurial changeset with the correct author and date as you would expect). But once you push your changes from Mercurial to Subversion, any clones from your bridge repository are now invalid. I knew this full well going into it. This would normally be fine for a workflow where I made an hgsubversion clone for each piece of concurrent development I want to do.

The problem is that, since our Subversion repository is so big (working copy is 2.5 GB and we use it for non-code assets), cloning it takes really a long time (about 7 minutes), which leads me to the second negative. It’s slow. I still get the benefits of fast local commits, but overall, using TortoiseHg and the command-line tools drag just enough that it’s left me unsatisfied and in search of a better solution.

So, I’m in the process of checking out our repository using Git, per the instructions here. I tried to just check out our “integration” branch, which is all I care about, but even though I pointed Git to that URL, it appears to be getting the whole kit and kaboodle. It’s been hours now since I started, but I’m hoping this is just the up-front cost I need to pay. Git is supposed to be really fast, so we’ll see…

Tuesday, November 29, 2011