experiments with claude, part ⅲ: JMAP-Tester coverage
Here’s another post in which I fiddle around with Claude Code and determine that it is not just spewing out nonsense and bugs, but instead is doing kinda what I would’ve done, at least enough to reduce my total effort. This time, writing tests.
Project Three: JMAP-Tester test coverage
Okay, I progressed from code I’d throw away to code I would keep but not look at. I was progressing up the totem pole of how much cultural value we put on code. What was the next least respected code? Tests, of course.
Now, I actually love tests, and like treating them like first-class code, and building libraries to make testing better. One of those libraries is JMAP::Tester, which we use in tons of our testing. Until pretty recently, it didn’t have all that much testing of its own. That is: JMAP-Tester was used to test things, but was not itself tested. In December, as part of adding some features to JMAP::Tester, I started to expand its test coverage. This was rewarding and useful, but I didn’t get to 100% coverage. I used to strive for 100% (well, 95% coverage) on my code, but these days… well, who has the time?
Turns out, Claude has the time. This one was pretty darn impressive. You can read the whole transcript, but here’s an accurate summary:
That was it! You can read the pull request.
Well, there were a couple more bits, mostly me saying, “Make it look like how I’d have written it” (literally), and some git faff. I did go through and tweak the code to be more readable and better organized. I could’ve given Claude clearer instructions to avoid most of that, or had standing instructions about it… but really, it was a good way to keep my mind engaged while I reviewed that the code was testing what it should be, anyway.
If “write more coverage tests” was the only thing I could get out of Claude, it’d still be huge, but obviously there’s more. I’ll keep posting…