mizvekov wrote: I don't think them being autogenerated changes whether a test is a regression test or not.
The biggest issue of the current tests, is the maintainability concern regarding the context of diagnostics. So for example, if a patch causes a new note to be produced, it's hard to understand from the compiler output what error is associated to it, and what other notes are part of the same diagnostic chain. If the patch then has a change to add the expectation of this note, it's impossible, in the general case, for reviewers to see that's correct, instead of just a change that makes the test pass. The format you proposed for the DR tests, together with this patch, helps that maintainability concern. But not as far, in that we'd still need to convert a bunch of existing tests, and that updating these tests manually is tedious. I think a criticism along the lines of the reasoning that it makes it easier to do things blindly, would also shut down all kinds of automation. https://github.com/llvm/llvm-project/pull/179835 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
