Traverse attributes in AttributedStmt.

[16bit-ykiko]

Fix #117687.

[llvmbot]

@llvm/pr-subscribers-clang

Author: ykiko (16bit-ykiko)

Changes

Fix #117687.

---

Full diff: https://github.com/llvm/llvm-project/pull/117692.diff

1 Files Affected:

• (modified) clang/include/clang/AST/RecursiveASTVisitor.h (+4-1)

diff --git a/clang/include/clang/AST/RecursiveASTVisitor.h b/clang/include/clang/AST/RecursiveASTVisitor.h
index b2dd51319ba9ea..6ab9d0e71d71aa 100644
--- a/clang/include/clang/AST/RecursiveASTVisitor.h
+++ b/clang/include/clang/AST/RecursiveASTVisitor.h
@@ -2442,7 +2442,10 @@ DEF_TRAVERSE_STMT(GotoStmt, {})
 DEF_TRAVERSE_STMT(IfStmt, {})
 DEF_TRAVERSE_STMT(IndirectGotoStmt, {})
 DEF_TRAVERSE_STMT(LabelStmt, {})
-DEF_TRAVERSE_STMT(AttributedStmt, {})
+DEF_TRAVERSE_STMT(AttributedStmt, {
+  for (auto *I : S->getAttrs())
+    TRY_TO(getDerived().TraverseAttr(const_cast<clang::Attr *>(I)));
+})
 DEF_TRAVERSE_STMT(NullStmt, {})
 DEF_TRAVERSE_STMT(ObjCAtCatchStmt, {})
 DEF_TRAVERSE_STMT(ObjCAtFinallyStmt, {})

[16bit-ykiko]

A test failed after the modification, but it really confused me

// Contexts where there is no function call, no diagnostic.
bool bad();

template <bool>
requires requires { bad(); }
void g() [[clang::nonblocking]] {}

void g() [[clang::nonblocking]] {
    decltype(bad()) a; // doesn't generate a call so, OK
    [[maybe_unused]] auto b = noexcept(bad());
    [[maybe_unused]] auto c = sizeof(bad());
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wassume"
    [[assume(bad())]]; // never evaluated, but maybe still semantically questionable?
#pragma clang diagnostic pop
}

  File /home/ykiko/C++/clice2/deps/llvm/clang/test/Sema/attr-nonblocking-constraints.cpp Line 396: function with 'nonblocking' attribute must not call non-'nonblocking' function 'bad'
error: 'expected-note' diagnostics seen but not expected: 
  File /home/ykiko/C++/clice2/deps/llvm/clang/test/Sema/attr-nonblocking-constraints.cpp Line 384: declaration cannot be inferred 'nonblocking' because it has no definition in this translation unit

I don't know what happened here. They seem to have no connection at all.

[Sirraide]

I don't know what happened here. They seem to have no connection at all.

What seems to be happening here is that we’re now traversing the argument of the [[assume]] attribute, which causes us to warn about the call to bad(). This wasn’t a problem before because TraverseAttributedStmt wouldn’t traverse the attributes. I think you need to override TraverseAttr() in the FunctionBodyASTVisitor in SemaFunctionEffects.cpp to do nothing, as there currently are no statement attributes that we’d want to traverse in the first place afaik.

[AaronBallman]

Thanks for the fix! Please be sure to also add some test coverage for the changes to clang/unittests/AST/RecursiveASTVisitorTest.cpp.