MaskRay added a comment. Sorry, my previous main comment had been written before I introduced `LIT_CLANG_STD_GROUP` in `llvm/utils/lit/lit/llvm/config.py`. The multiple `%clang_cc1` approach actually looks like the following. Note the use of `%stdcxx_17-` to make the test future-proof. (It is non-trivial to run one `RUN` line multiples times with different `LIT_CLANG_STD_GROUP`. For now I just test locally with different `LIT_CLANG_STD_GROUP`.)
// RUN: %clang_cc1 %s -fsyntax-only -verify=expected,precxx17 %stdcxx_11-14 -fdata-sections -fcolor-diagnostics // RUN: %clang_cc1 %s -fsyntax-only -verify %stdcxx_17- -fdata-sections -fcolor-diagnostics ... TypedefAligned4 TA8c = TA8a + TA8b; // expected-warning {{passing 4-byte aligned argument to 8-byte aligned parameter 'this' of 'operator+' may result in an unaligned pointer access}} \ // expected-warning {{passing 4-byte aligned argument to 8-byte aligned parameter 1 of 'operator+' may result in an unaligned pointer access}} \ // precxx17-warning {{passing 4-byte aligned argument to 8-byte aligned parameter 'this' of 'StructAligned8' may result in an unaligned pointer access}} If this is changed to use `#if __cplusplus >= 201703L`, there will be multiple lines with relative line numbers (e.g. `@-2` `@-4`) register int ro; // expected-error {{illegal storage class on file-scoped variable}} #if __cplusplus >= 201703L // expected-error@-2 {{ISO C++17 does not allow 'register' storage class specifier}} #elif __cplusplus >= 201103L // expected-warning@-4 {{'register' storage class specifier is deprecated}} #endif Personally I prefer multiple `%clang_cc1` over `#if`. The first few lines give users a first impression. The dispatch makes it clear the test has different behaviors with the `%stdcxx_*` described dialects. ================ Comment at: llvm/utils/lit/lit/llvm/config.py:579 + l = h - clang_std_group % (h-l+1) + self.config.substitutions.append((s, '-std=c++' + clang_std_values[l])) + ---------------- aaron.ballman wrote: > One thing we should consider is whether we want to run in *all* the specified > language modes instead of just the newest mode. This will make running tests > slower because we'll run significantly more of them, and it might get awkward > if a lot of tests change behavior in the different language modes, so I don't > suggest it as part of this patch. This is difficult in lit. Will answer in my main comment. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D131464/new/ https://reviews.llvm.org/D131464 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits