[clang][CUDA] Assume unknown emission status for skipped function definitions

[kadircet]

Emission status seems to be only used by cuda/openmp/hip compiles, to figure out
when to emit diagnostics. Current logic emits "uknown" when definition is
missing, so i extended that to skipped-function-bodies as well.

[llvmbot]

@llvm/pr-subscribers-openmp

@llvm/pr-subscribers-clang

Author: kadir çetinkaya (kadircet)

Changes

Emission status seems to be only used by cuda/openmp/hip compiles, to figure out
when to emit diagnostics. Current logic emits "uknown" when definition is
missing, so i extended that to skipped-function-bodies as well.

---

Full diff: https://github.com/llvm/llvm-project/pull/100124.diff

2 Files Affected:

• (modified) clang/lib/Sema/SemaDecl.cpp (+2-1)
• (modified) clang/unittests/Tooling/ToolingTest.cpp (+5)

diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index bb25a0b3a45ae..c04c66d8c445a 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -20148,7 +20148,8 @@ Sema::FunctionEmissionStatus Sema::getEmissionStatus(const FunctionDecl *FD,
     // be emitted, because (say) the definition could include "inline".
     const FunctionDecl *Def = FD->getDefinition();
 
-    return Def && !isDiscardableGVALinkage(
+    // We can't compute linkage when we skip function bodies.
+    return Def && !Def->hasSkippedBody() && !isDiscardableGVALinkage(
                       getASTContext().GetGVALinkageForFunction(Def));
   };
 
diff --git a/clang/unittests/Tooling/ToolingTest.cpp b/clang/unittests/Tooling/ToolingTest.cpp
index 2e3da2cd2a701..f41a44fa0922a 100644
--- a/clang/unittests/Tooling/ToolingTest.cpp
+++ b/clang/unittests/Tooling/ToolingTest.cpp
@@ -586,6 +586,11 @@ TEST(runToolOnCode, TestSkipFunctionBody) {
   EXPECT_FALSE(runToolOnCodeWithArgs(
       std::make_unique<SkipBodyAction>(),
       "template<typename T> int skipMeNot() { an_error_here }", Args2));
+
+  EXPECT_TRUE(runToolOnCodeWithArgs(
+      std::make_unique<SkipBodyAction>(),
+      "__inline __attribute__((__gnu_inline__)) void skipMe() {}",
+      {"--cuda-host-only", "-nocudainc", "-xcuda"}));
 }
 
 TEST(runToolOnCodeWithArgs, TestNoDepFile) {

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[kadircet]

this is fixing crashes when using clangd on cuda files. as clangd skips function bodies inside headers to speed up parsing times.

[ilya-biryukov]

Could you reformat the code so that the clang-format presubmit is happy?
Otherwise LG.