[clang][bytecode] Reject dependent RequiresExprs

[tbaederr]

Fixes #152899

[llvmbot]

@llvm/pr-subscribers-clang

Author: Timm Baeder (tbaederr)

Changes

Fixes #152899

---

Full diff: https://github.com/llvm/llvm-project/pull/155230.diff

2 Files Affected:

• (modified) clang/lib/AST/ByteCode/Compiler.cpp (+2)
• (modified) clang/test/AST/ByteCode/cxx2a.cpp (+14)

diff --git a/clang/lib/AST/ByteCode/Compiler.cpp b/clang/lib/AST/ByteCode/Compiler.cpp
index e3235d34e230e..a98c0acb4f660 100644
--- a/clang/lib/AST/ByteCode/Compiler.cpp
+++ b/clang/lib/AST/ByteCode/Compiler.cpp
@@ -3879,6 +3879,8 @@ bool Compiler<Emitter>::VisitCXXUuidofExpr(const CXXUuidofExpr *E) {
 template <class Emitter>
 bool Compiler<Emitter>::VisitRequiresExpr(const RequiresExpr *E) {
   assert(classifyPrim(E->getType()) == PT_Bool);
+  if (E->isValueDependent())
+    return false;
   if (DiscardResult)
     return true;
   return this->emitConstBool(E->isSatisfied(), E);
diff --git a/clang/test/AST/ByteCode/cxx2a.cpp b/clang/test/AST/ByteCode/cxx2a.cpp
index 744c99eaa1e68..533173d84792f 100644
--- a/clang/test/AST/ByteCode/cxx2a.cpp
+++ b/clang/test/AST/ByteCode/cxx2a.cpp
@@ -239,3 +239,17 @@ namespace GH150705 {
   constexpr const A& a = b;
   constexpr auto x = (a.*q)(); // both-error {{constant expression}}
 }
+
+namespace DependentRequiresExpr {
+  template <class T,
+            bool = []() -> bool { // both-error {{not a constant expression}}
+              if (requires { T::type; })
+                return true;
+              return false;
+            }()>
+  struct p {
+    using type = void;
+  };
+
+  template <class T> using P = p<T>::type; // both-note {{while checking a default template argument}}
+}

[shafik]

Do you know which clause causes clang to choose to reject this, looks like we are the only one: https://godbolt.org/z/KhfeP5W6d

CC @hokein looks like came in w/ bd4662c

[tbaederr]

Do you know which clause causes clang to choose to reject this, looks like we are the only one: https://godbolt.org/z/KhfeP5W6d

CC @hokein looks like came in w/ bd4662c

No clue. The RequiresExpr ends up being dependent, but that does not end up being propagated to any parent nodes, so none of the previous dependency checks works. Whether the RequiresExpr (or its SimpleRequirement) should be dependent at all, I don't know.

[shafik]

CC @zygoloid is clang correct to reject the example in this code?

[zygoloid]

I don't see why that would be invalid. When forming p<T>::type, we should produce a default argument for p by substituting T = T into the specified default argument. That should produce a value-dependent initializer, which is fine, but makes p<T> a dependent type. So p<T>::type should be interpreted as a dependent type-id.

[zygoloid]

Looks like the problem is not a RequiresExpr issue at all, it's a problem with how we handle partial substitution into lambdas. Reduced to a crash on valid: https://godbolt.org/z/n46Knan3v

[shafik]

CC @cor3ntin