[clang][Interp] Enable existing source_location tests

[tbaeder]

Summary

As requested.

This depends on a few other patches though.

[tbaeder]

clang/test/SemaCXX/source_location.cpp

764–768

The comment block just before this #if seems to say that the new interpreter is correct here...?

[cor3ntin]

clang/test/SemaCXX/source_location.cpp

764–768

Yes, which is interesting. Maybe modifying the existing interpreter properly would be doable then? I'll have to look into that...
I'm surprised it's only an evaluation issue

[hazohelet]

I was rebasing D155064 locally and saw test failures for the new interpreter in this file.
The cause seems to me that the new interpreter is not handling some cases of SourceLocExpr correctly when they appear inside constant-evaluated or immediate-function context.
e.g.

struct A {
  int n = __builtin_LINE();
};
struct B {
  A a = {};
};
#line 100
consteval void f() {
  constexpr B c = {};
  static_assert(c.a.n == 101, "");
}

Live demo: https://godbolt.org/z/9Y7bzj56G

D155064 pushes constant-evaluated context against initializers of constexpr variables, so this bug appears in the test file.
I'm not in any hurry to push D155064 upstream, but if it might take several weeks for this bug to be fixed, I'd like this change to be reverted until it gets fixed.

[tbaeder]

@hazohelet Thanks for noticing. Should be fixed via https://github.com/llvm/llvm-project/commit/f8074942c5a6f266abb1c4d1465a4d8115b22d43.

[hazohelet]

@tbaeder Now it's working correctly. Thanks!

clang/test/SemaCXX/source_location.cpp

 // RUN: %clang_cc1 -std=c++1z -fcxx-exceptions -fexceptions -verify %s
 // RUN: %clang_cc1 -std=c++2a -fcxx-exceptions -DUSE_CONSTEVAL -fexceptions -verify %s
 // RUN: %clang_cc1 -std=c++2b -fcxx-exceptions -DUSE_CONSTEVAL -DPAREN_INIT -fexceptions -verify %s
 // RUN: %clang_cc1 -std=c++1z -fcxx-exceptions -fms-extensions -DMS -fexceptions -verify %s
 // RUN: %clang_cc1 -std=c++2a -fcxx-exceptions -fms-extensions -DMS -DUSE_CONSTEVAL -fexceptions -verify %s
+//
+/// FIXME: The -DPAREN_INIT one is missing for the new interpreter.
+// RUN: %clang_cc1 -std=c++1z -fcxx-exceptions -fexceptions -fexperimental-new-constant-interpreter -DNEW_INTERP -verify %s
+// RUN: %clang_cc1 -std=c++2a -fcxx-exceptions -DUSE_CONSTEVAL -fexceptions -fexperimental-new-constant-interpreter -DNEW_INTERP -verify %s
+// RUN: %clang_cc1 -std=c++1z -fcxx-exceptions -fms-extensions -DMS -fexceptions -fexperimental-new-constant-interpreter -DNEW_INTERP -verify %s
+// RUN: %clang_cc1 -std=c++2a -fcxx-exceptions -fms-extensions -DMS -DUSE_CONSTEVAL -fexceptions -fexperimental-new-constant-interpreter -DNEW_INTERP -verify %s
 // expected-no-diagnostics
 
 #define assert(...) ((__VA_ARGS__) ? ((void)0) : throw 42)
 #define CURRENT_FROM_MACRO() SL::current()
 #define FORWARD(...) __VA_ARGS__
 
 template <unsigned>
 struct Printer;

 // However Clang does not yet implement this as it would introduce
 // a fair bit of complexity.
 // We intend to implement that functionality once we find real world
 // use cases that require it.
 constexpr int test_init_capture(int a =
                 [b = SL::current().line()] { return b; }()) {
   return a;
 }
-#ifdef USE_CONSTEVAL
+#if defined(USE_CONSTEVAL) && !defined(NEW_INTERP)
 static_assert(test_init_capture() == __LINE__ - 4);
 #else
 static_assert(test_init_capture() == __LINE__ );
 #endif
 
 namespace check_immediate_invocations_in_templates {
 
 template <typename T = int>
 struct G {
     T line = __builtin_LINE();
 };
 template <typename T>