Thor Lancelot Simon <tls%panix.com@localhost> writes: > Usually most developers (even I) don't screw up. This is NetBSD-current, > not a release branch, and so I believe some amount of "screw up and fix > it later" is OK if it greatly speeds development. And I believe that, > because most developer usually don't screw up, the current situation > does, in fact, greatly speed development, compared to forcing everyone > to run all the anita tests before every commit. I think periodic runs > of the tests represent a good compromise between speed of development > and speed of debugging. I agree with your argument that very strict pre-commit testing and no requirements for quality would both be worse situations. And I agree with your characterization that most people are pretty careful. But, I think the question can be framed as whether where we are is the right place, or if we should be a little more (or a little less) careful, rather than whether we should have far stricter norms. As others have pointed out, periods of time when the tree doesn't build and periods where there are massive test failures seem to lead to cascading effects. The build-broken-often problem does seem to be milder now than last fall. So perhaps where we are now should be modified by a norm that says: If you've made commits more than a few times in the last 6 months that break the build (even if it is "just setlists"), you really really have to at least build whatever you are working on (and perhaps one of i386/amd64/sparc64, if different) before committing. This is especially true if you were not able to fix your previous breaks in the last 24 hours. Changes that are large in scope and don't obviously have a low risk of trouble should at least have a py-anita run or equivalent. I realize that won't catch everything, but it should make the two situations that seem to give rise to real trouble rarer, and only put more work where it is actually needed.
Description: PGP signature