[Clang][Sema] Push an evaluation context for type constraints

[zyn0217]

This helps getTemplateInstantiationArgs to properly recover template arguments of an enclosing concept Decl.

Fixes #93821

[llvmbot]

@llvm/pr-subscribers-clang

Author: Younan Zhang (zyn0217)

Changes

This helps getTemplateInstantiationArgs to properly recover template arguments of an enclosing concept Decl.

Fixes #93821

---

Full diff: https://github.com/llvm/llvm-project/pull/93945.diff

3 Files Affected:

• (modified) clang/docs/ReleaseNotes.rst (+1)
• (modified) clang/lib/Sema/SemaTemplateDeduction.cpp (+14)
• (modified) clang/test/SemaTemplate/concepts-lambda.cpp (+12)

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index 49ab222bec405..83eb285e9ca48 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -710,6 +710,7 @@ Bug Fixes to C++ Support
 - Correctly treat the compound statement of an ``if consteval`` as an immediate context. Fixes (#GH91509).
 - When partial ordering alias templates against template template parameters,
   allow pack expansions when the alias has a fixed-size parameter list. Fixes (#GH62529).
+- Fixed a type constraint substitution issue involving a generic lambda expression. (#GH93821)
 
 Bug Fixes to AST Handling
 ^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/clang/lib/Sema/SemaTemplateDeduction.cpp b/clang/lib/Sema/SemaTemplateDeduction.cpp
index 853c0e1b50619..56529b4d852e7 100644
--- a/clang/lib/Sema/SemaTemplateDeduction.cpp
+++ b/clang/lib/Sema/SemaTemplateDeduction.cpp
@@ -5004,6 +5004,20 @@ static bool CheckDeducedPlaceholderConstraints(Sema &S, const AutoType &Type,
     return true;
   MultiLevelTemplateArgumentList MLTAL(Concept, CanonicalConverted,
                                        /*Final=*/false);
+  // Build up an EvaluationContext with an ImplicitConceptSpecializationDecl so
+  // that the template arguments of the constraint can be preserved. For
+  // example:
+  //
+  //  template <class T>
+  //  concept C = []<D U = void>() { return true; }();
+  //
+  // We need the argument for T while evaluating type constraint D in
+  // building the CallExpr to the lambda.
+  EnterExpressionEvaluationContext EECtx(
+      S, Sema::ExpressionEvaluationContext::ConstantEvaluated,
+      ImplicitConceptSpecializationDecl::Create(
+          S.getASTContext(), Concept->getDeclContext(), Concept->getLocation(),
+          CanonicalConverted));
   if (S.CheckConstraintSatisfaction(Concept, {Concept->getConstraintExpr()},
                                     MLTAL, TypeLoc.getLocalSourceRange(),
                                     Satisfaction))
diff --git a/clang/test/SemaTemplate/concepts-lambda.cpp b/clang/test/SemaTemplate/concepts-lambda.cpp
index fac790d09f9cf..280be71284f97 100644
--- a/clang/test/SemaTemplate/concepts-lambda.cpp
+++ b/clang/test/SemaTemplate/concepts-lambda.cpp
@@ -225,3 +225,15 @@ void foo() {
   }(x);
 }
 } // namespace GH73418
+
+namespace GH93821 {
+
+template <class>
+concept C = true;
+
+template <class...>
+concept D = []<C T = int>() { return true; }();
+
+D auto x = 0;
+
+} // namespace GH93821

[cor3ntin]

Should that not be UnevaluatedContext?
https://eel.is/c++draft/temp.concept#6

(I appreciate the great comment, thanks!)

[zyn0217]

Good question. Honestly, I copied the line from Sema::CheckConceptTemplateId:

llvm-project/clang/lib/Sema/SemaTemplate.cpp

Lines 5662 to 5670 in f34dedb

	EnterExpressionEvaluationContext EECtx{
	*this, ExpressionEvaluationContext::ConstantEvaluated, CSD};
	if (!AreArgsDependent &&
	CheckConstraintSatisfaction(
	NamedConcept, {NamedConcept->getConstraintExpr()}, MLTAL,
	SourceRange(SS.isSet() ? SS.getBeginLoc() : ConceptNameInfo.getLoc(),
	TemplateArgs->getRAngleLoc()),
	Satisfaction))

I'll look into the details later and hopefully get back with an answer. (Of course I'd appreciate it if @erichkeane could beat me to it.)

[erichkeane]

I think Corentin is right. And that link from SemaTemplate might actually be wrong too, but I'm not completely sure. Though I would have expected us to see a bunch of bugs for evaluating that incorreclty.

[zyn0217]

I changed these two to Unevaluated and all our tests were still passed.

a bunch of bugs for evaluating that incorreclty.

Probably? Or it could be we have been pushing another unevaluated context during the evaluation, so the context kind here doesn't matter.