Maetveis wrote: > @Maetveis > > All of the Visibility fields appear to be not null, but that's a good check > to add anyway. Thanks. After adding it though, the same failures occur. > Here's an example. Maybe there's a gap in my understanding: > > The script generates tests for both flang & clang -cc1as when looking at the > O_flag option. cc1as outputs the unexpected error message, but flang does not.
That looks to be by design, because you are using the `-help`, the drivers meant to be user facing (clang, clang-cl and flang) don't report unknown options with `-help`. A user might tack on a `-help` after their broken command line when they are looking for the exact spelling of their option so it makes sense that it is not an error to have unknown options when mixed with `-help`. Simply you shouldn't be using `-help`, instead passing options to perform a real compilation on the command line like `clang-cl <tested-options> -### -x c++ -c - < /dev/null` should work better. The other issues is that `-O_flag` is parsed as `-O _flag` for drivers that support `-O`, in runtime longer valid options have preference over shorter ones, in this case I think if there is a supported joined option (`Kind == JoinedClass`) that is a prefix of an unsupported option you now way to test the unsupported option so it should be skipped. https://github.com/llvm/llvm-project/pull/120900 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
