Our data provider of choice is collegefootballdata.com and they’ve been great. In the natural life stages of API developers everywhere they’ve introduced a V2 with breaking changes, final deployment coming real soon now. I’ve been there myself. After putting off any mods for a while I dusted off the codebase yesterday and got the fixes in. First real commit in quite some time.
I used my new favorite AI assistant: blackbox.ai. I asked it to read the new docs and identify any changes. That worked… pretty well. There were a few breaks that turned up in the actual runs. The AI assistant always works much better from a basis of actual failure stack traces.
The change itself is here and here. The biggest practical difference was a python change not necessarily brought on by the API itself, as datetime types are now getting converted from string inside the API. After the commit went through, the “github actions” regression test kicked in and verified against past data. Then the unit tests for ties coughed out a few new errors which I fixed. The system works?
The tests then started failing on API limits, since it turns out tiers have changed. This is due to what we’ll charitably call a naive approach in the code: we do an API call for every team, instead of pulling the whole year at a time and building a local lookup table. So that’s a nice to-do to put on the list. The whole codebase could also do with a refactoring for some other efficiencies. It feels quaint: I wrote this in the pre-AI assistant era with less than perfect python knowledge. There’s always a mythical v2 out there.