rovka added a comment. In D126291#3577159 <https://reviews.llvm.org/D126291#3577159>, @mstorsjo wrote:
> In D126291#3577133 <https://reviews.llvm.org/D126291#3577133>, @rovka wrote: > >> I had the same idea about switching the tests to using target triples >> instead of having separate files for it, but initially I had some issues >> getting that to work properly. When specifying a triple, we need to provide >> an architecture. Leaving the triple as `unkown-linux-gnu` or just >> `linux-gnu` gives us an error along the lines of `flang-new: error: unknown >> target triple 'unknown-unknown-linux-gnu', please use -triple or -arch`. >> OTOH, hardcoding an architecture like x86 or aarch64 fails if we're not >> building that specific backend. > > If you actually execute code generation, then yes, it fails if that specific > arch isn't enabled. But for general compiler driver level tests, which just > print out the command the would have executed (when running with `-###`), it > should work without the actual code generation target being available. This > is at least how it's done for clang's corresponding tests, most files in > `clang/test/Driver` have hardcoded arch triples, without any REQUIRES lines > or other exclusions. Oops, you're right, I was testing with a wrong x86 triple >.< I'll simplify the patch then. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D126291/new/ https://reviews.llvm.org/D126291 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
