On Mon, Feb 20, 2012 at 02:04:26PM -0800, Junio C Hamano wrote: > > ... Things may have changed since then, of course, but I at > > least know that they were sufficient in 34110cd^. > > Looking at where cache_tree_free() is called, I think back then the > two-way merge was deemed OK, but we did not trust three-way merge or > merge-recursive at all. Thanks, I'll take a look more closely at those cases. > It is OK to check that we do not over-invalidate for performance, but it > is a lot more important to make sure we do not under-invalidate for > correctness. I am a bit worried that you seem to be putting more stress > on the former. I think it is just selection bias of the specific parts of his tests that I was responding to. I completely agree that correctness is way more important, and I'm also trying to come up with tests to validate correctness. I just wasn't talking about them there. I still think replaying real-world test cases is going to be more likely to find issues in invalidation. I can come up with lots of simple test-cases, but they're not likely to find anything we wouldn't find in the code with trivial inspection. I think a combination of careful analysis and real-world validation is going to be more helpful in the long run than the kind of simplistic tests that are in t0090. -Peff -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html