On 3/3/26 9:16 PM, Eric Biggers wrote:
Unfortunately the verity-compat-test script isn't in very good shape. I opened a pull request to clean it up: https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/890
The testing suite was written mostly as a volunteer effort for cryptsetup to cover userspace interoperability, so it uses mostly parameters that worked properly with many ancient kernels (root=2, etc.). The testsuite was not intended to test the kernel in the first place, similar to the dm-crypt one. Despite that, it has found many kernel bugs in the past, apparently because the kernel itself lacks proper tests. As I am the primary author of this test and have not been paid for working on cryptsetup for years now, I have really enough of this "your code is shit, I know better" approach without even cc'ing us. We try not to touch tests much (e.g., refactoring) unless really necessary, to prevent accidentally introducing regressions. If there is a QA team validating it, it would be much simpler, but each corporation cares only about its own corporate sh^W products. Your approach is going exactly in the opposite direction, rewriting only parts you are interested in, according to your taste. So now we have multiple approaches in different scripts. I am really tempted to reject this, as it will add us much more effort in the future. That said, I want to add new tests there, but please at least try to think about it from this point of view. Thanks, Milan
