[clang-repl] Test explicit emission of dtors in runtime interface builder (NFC)

[weliveindetail]

This patch adds test coverage for an edge case that is supported already.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Stefan Gränitz (weliveindetail)

Changes

This patch adds test coverage for an edge case that is supported already.

---

Full diff: https://github.com/llvm/llvm-project/pull/89734.diff

1 Files Affected:

• (added) clang/test/Interpreter/force-codegen-dtor.cpp (+13)

diff --git a/clang/test/Interpreter/force-codegen-dtor.cpp b/clang/test/Interpreter/force-codegen-dtor.cpp
new file mode 100644
index 00000000000000..a299ea46d5eac0
--- /dev/null
+++ b/clang/test/Interpreter/force-codegen-dtor.cpp
@@ -0,0 +1,13 @@
+// UNSUPPORTED: system-aix
+
+// RUN: cat %s | clang-repl | FileCheck %s
+int *x = new int();
+template <class T> struct GuardX { T *&x; GuardX(T *&x) : x(x) {}; ~GuardX(); };
+template <class T> GuardX<T>::~GuardX() { delete x; x = nullptr; }
+
+// clang would normally defer codegen for ~GuardX()
+// Make sure that RuntimeInterfaceBuilder requests it explicitly
+(GuardX(x))
+
+// CHECK-NOT: Symbols not found
+// CHECK-NOT: _ZN6GuardXIiED2Ev

[weliveindetail]

I'd prefer a check like below that is more explicit, but we recognize expressions for value printing (the thing we parse as "semi-missing") only in trailing statements. If we added the code below, then (GuardX(x)) is not considered trailing anymore, because we pipe he whole file into clang-repl.

#include <cstdint>
auto x_addr = static_cast<uint64_t>(reinterpret_cast<uintptr_t>(x));

extern "C" int printf(const char *, ...);
printf("0x%016" PRIx64 "\n", x_addr);

// CHECK: 0x00000000000000000

[vgvassilev]

How about a printf in the dtor?

[vgvassilev]

LGTM!

[weliveindetail]

Oh interesting, the Windows test actually fails at this assertion:

DC->getLexicalParent() == CurContext && "The next DeclContext should be lexically contained in the current one.", file C:\ws\src\clang\lib\Sema\SemaDecl.cpp, line 1332

@vgvassilev Any ideas?

[vgvassilev]

We generally have to put a switch for the late parsed templates. We have it in some other tests.