[webkit-dev] to reitveld or not to reitveld

[Ojan Vafai]

Sorry in advance for the long email. I'm trying to be thorough.

There's been a lot of discussion on #webkit about possibly using a code
review tool like reitveld for webkit reviews. There's been various
suggestions and a few misunderstandings, so it seems worth having a more
formal discussion about this with the larger WebKit community.
The things I'd like to assess here are:
1. Pros/Cons of using a system like reitveld. I listed some below. Please
add any that I missed.
2. Whether the WebKit community is interested in pursuing something like this.
3. If there is interest, what is the best way to move forward.

WHAT IS REITVELD
It's a code review tool. Reitveld doesn't allow you to do anything that is
impossible with the current review process, however, it makes the review
process more efficient and less error-prone. As such, it makes it easier and
less time-consuming to do good, thorough code reviews.

The basic gist of reitveld is that it allows you to put comments inline and
send them all in one chunk. Then it lets the reviewee easily respond to each
comment individually and send all the responses in one chunk.

EXAMPLE CHROMIUM PATCH
http://codereview.chromium.org/119103

Note that you can view the patch in each version that was uploaded and that
you can diff between versions. Also, if a comment was made in the version
you were looking at, then you can see all the comments/responses.

To see this nicely, under "Delta from patch set" in patch set 3, click on 2.
That is where most of the comments in this review were made. For example,
http://codereview.chromium.org/119103/diff2/14:27/29. You can see all the
comments and responses along with the diff in the patch to see that the
reviewer comments were implemented as intended.

Keyboard shortcuts to try out:
-n/p to switch between diff chunks
-shift n/p to switch between comments
-j/k to go to the next/previous file

*Please don't actually click the "Publish all my drafts" button on the
publish page as you'll be modifying a real code review.*

Other things to try
-try the side-by-side diff and the unified diff views
-adding comments (double click)
-replying to comments
-go to the publish page (click the publish link or type "m")

Also note that the "Committed" URL is automatically added when the patch is
committed and the reitveld issue is marked closed. So there isn't extra
overhead in maintaining list of outstanding code reviews.

HOW TO TRY IT OUT
Here's the process for trying out reitveld with a webkit patch. The current
workflow is a bit janky, but some scripting and some simple reitveld fixes
would make this a lot more natural and automated (e.g. chromium uses
commandline "gcl upload" to put up a new patch).

1. Find a non-git patch
2. Go to http://codereview.chromium.org/new
3. Login with a Google account (e.g. any gmail or Google search account)
4. On that page, type in a subject and paste in the URL to the patch in the
URL field.
5. Click "Create Issue"

There's a couple apparent bugs that are easily fixable:
1. The ChangeLog files don't get downloaded correct, so the diffs don't
work. This is an AppEngine problem that Chromium works around with the gcl
upload script.
2. With an old patch there are often diff chunk mistmatches, which breaks
the side-by-side diff view (you can use the unified diff in those cases).

PROS
For the reviewer:
-easier to write thorough review comments since adding comments is so
light-weight
-easier to make sure that all review comments actually got implemented

For the reviewee:
-easier to see which line the reviewer's comment addresses
-easier to make sure you've completed all the reviewer's comments (you can
mark them as "done" in reitveld as you go)

For everyone:
-efficient keyboard navigation (e.g. j/k to navigation between diff chunks
and n/p to navigate between files
-easier to follow the progression of a code review and see what changed over
the course of the review
-shows image files, so you can see before/after before commit

CONS (most of these are easy to fix/improve)
-There's no pretty printed view of all the files in the patch at once that
lets you insert comments
-The UI is a bit cluttered
-It takes using it for a couple patches before you're totally comfortable
with it
-Currently doesn't work with diffs generated by git
-Reitveld's current implementation requires running on AppEngine
-A couple issues with reitveld on appengine that Chromium uses a script to
workaround (line-endings differences and large files like ChangeLogs don't
upload correclty).

PATH FORWARD
As far as reitveld versus another code review tool, I don't have strong
opinions. I hear http://www.review-board.org/ is good, but I haven't used
it. One advantage of using reitveld is that a lot of the work on reitveld
was done by Chromium team members and so modifying it to meet WebKit needs
(e.g. running an instance that isn't tied to Chromium, changing the UI,
etc.) should be relatively painless.

I think the transition to using a new tool would need to be gradual, so that
people can continue use the current webkit process and not be forced into
using a new tool. Like using git vs. svn, it ought to be possible to use
either process without putting an extra burden on the webkit community.

If we agree that something like reitveld would be good for the webkit
community and wanted to make reitveld possible to use with the current
WebKit workflow, we *could* work out some of the simple kinks listed above
and do a lightweight integration with bugzilla such that any updates to
reitveld are reflected in bugzilla (but not vice versa). There was pushback
to this on #webkit, but it seems to me like the simplest non-invasive way to
let the webkit community try this out to see if it's worth investing more
time into.

That said, I'm totally open towards other approaches that aren't an
unreasonable amount of effort.

Ojan

P.S. For those interested, here's a talk about
Mondrian<http://www.youtube.com/watch?v=sMql3Di4Kgc> [Reitveld's
closed-source predecessor] and how Google moved from text-diff base reviews
to Mondrian.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-dev/attachments/20090606/1fe21db5/attachment.html>