3 ms·
Google's response to Chesterton's Fence is: "if you liked it, then put a test on it". I used to update the internal version of numpy for Google and if people a
by dekhn 4y ago
Google's response to Chesterton's Fence is: "if you liked it, then put a test on it".
I used to update the internal version of numpy for Google and if people asked me to rollback after I made my update (having fixed all the test failures I could detect), and they didn't have a test, well, that's their problem. The one situation where that rule wouldn't apply is if I somehow managed to break production and we needed to do an emergency rollback.
I shed a tear when some of my old, unused code was autodeleted at Google, but nowadays my attitude is: HEAD of your version control should only contain things which are absolutely necessary from a functional selection perspective.
- Forge36 4y agoI like that philosophy. In a similar vein: if it's important why aren't we testing it. How do you encourage testing?
- compiler-guy 4y agoAt Google, because everything is in the mono repo and nearly everything around it is in some level of flux, if you don’t have a test for your code, it gets broken pretty regularly. Most teams are pretty understanding when they break you on some untested code, and will work with you to fix it, but the very first step in working together is for you to write a test that shows the issue. The resultant culture is heavily biased toward writing tests before something bad happens to you.