June 2007

I was up late last night experimenting with a new feature for Review Board. The idea was to introduce syntax highlighting for the code in the diff viewer through use of the Pygments library. Actually getting syntax highlighting to work was pretty easy, and before long I had it working. The trick, it turned out, was to get it working with our interline diff support.

Pygments generates a nice HTML-formatted string with span tags, but we need to insert our own span tags for the highlighting. This was a pain, since you need to make sure you place them at the correct locations, keeping in mind the existing span tags and entities. The code also needed to avoid messing up the nesting of tags. After working with it a bit, I had that working as well, and now the code is up for review. In a couple of days, Review Board will have nice syntax-highlighted diff output!

I have some new improvements in the works for the diff algorithm that should result in much better diffs (especially when moving functions around) and improved UI for that case as well. I’ll blog about that later when I have something to show 🙂

It’s only been a couple of weeks since the last update about Review Board, but too much is happening to stay quiet.

New Features

• Interdiffs

  Ever go through a long patch review process only to find it’s getting harder and harder to review any new changes due to the size of the patch? I’m sure most large projects have had to deal with this at one point or another.

  Review Board can now display interdiffs, which are diffs between versions of diffs. This makes it easy to see what was modified since the last change. Currently, there’s no UI for this, as there’s still a few backend changes that need to be made before we allow commenting on interdiffs, but it’ll be a full, proper feature very soon.

• New Diff Algorithm

  We were previously using Python’s SequenceMatcher, along with some hacks, to generate the diffs. SequenceMatcher, unfortunately, just isn’t a desirable diff generator, due to its algorithm, quirks (assuming every change between matching blocks consists entirely of inserts, deletes, or changes instead of a mix of them), and lack of extensibility (no leading whitespace trimming to simplify diffs containing lots of indents).

  I’ve known for a while that we needed a new algorithm, so I finally sat down to write one. The obvious choice was the algorithm used in GNU diff (amongst other programs) — Eugene Myers’ O(ND) Difference Algorithm. After a few frustrating nights that left me nearly bald (ok, not really), it worked! And the diffs were much better than the older diffs. I think this is the first Python implementation of this algorithm.

  This also gives us the ability to add some heuristics and other logic to clean up the diffs. The result is much more readable diffs that no longer show indentation changes or huge misaligned fragments.

• Improved support for project maintainers

  Project maintainers can now quickly download the uploaded diff from the diff viewer in order to apply the patch to their own tree. They’re also able to mark contributors’ review requests as submitted. While small additions, these make Review Board much more useful to open source projects.

• Better browser and SCM compatibility

  We’re working on better browser compatibility and better SCM compatibility.

  CVS is almost implemented. Contributors are working on Git and BZR support. We’re working on the best ways to get these distributed systems integrated cleanly into the system.

• Subversion post-review script

  We just received a patch for adding a Subversion-specific post-review script. Soon, we’ll combine the two and add support for CVS and other systems. This will give us one tool for creating review requests in all systems.

We’ve had a number of good contributors come out of the wood work, and from the sounds of it, several companies are using it now. I’m planning to put up a page listing teams, companies and projects using Review Board, so let me know if you’d like to be on the list! 🙂