Hi, With regards to the recent discussion of the pain in merging changes, and also generally, I think that the Sparse test suite needs to be enhanced. My impression is that most tests do not validate the generated linearized IR. The sparse-llvm tests do not validate the generated LLVM IR either. Without such validation it is hard to detect when some change breaks stuff. I think that having robust tests will also ease the pain of merging contributions as you will be able to detect breaks more reliably. Because of this reason, I have been reluctant to merge changes into my repository as I first want to have a set of regression tests that I can run whenever I merge changes. I have done some work on this. As validating sparse IR and LLVM IR is relatively difficult since even a spurious change in the IR output will cause tests to fail - the approach I have taken is that all tests are designed to be run. The success of the test is therefore determined by whether the compiled test runs correctly and generates the required outputs. Ideally we need a mix of simple unit tests and complex real life programs, as not all issues can be caught by simple tests. I am happy to contribute the tests I have been creating. I have sourced many of them from other opensource projects: I think that they are all permissively licensed. But even if some license conflicts, all that is needed is excluding the tests from the distribution I think. In my view it is still worth having these tests run every time there is a change. Unfortunately all my tests need a working sparse-llvm component. So until all the changes I have done are reflected in sparse, the tests will not compile I think. As I have reported before the current patch series does not cover all the fixes, therefore additional fixes are needed. Regards Dibyendu -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
