Skip to content

[concepts] Push a CurContext before substituting into out-of-line constraints for comparison #79985

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Jan 31, 2024

Conversation

zyn0217
Copy link
Contributor

@zyn0217 zyn0217 commented Jan 30, 2024

InjectedClassNameType is such that every template specialization within the Record scope ought to canonicalize to it, as outlined above the definition of that Type.

This invariant is maintained during the tree transformation at the rebuilding stage for a template specialization type; see RebuildTemplateSpecializationType. In that, we attempt to retrieve the current instantiation from Sema.CurContext, and if that fails, the transformation proceeds silently.

In terms of this issue, we previously set no CurContext other than that set by Parser, who had left the FunctionDecl before performing the constraint comparison. As a result, we would profile these types (i.e. InjectedClassNameType and its specialization type) into different values and fail to consider two expressions equivalent, although they are.

In passing, this also fixes a crash while attempting to dump the transformed expression. We failed to look into the template parameters from a CXXRecordDecl, which we should have done since the Decl we bound to a SubstTemplateTypeParmType can be of a CXXRecord type per the call to getTemplateInstantiationArgs within SubstituteConstraintExpressionWithoutSatisfaction.

This closes #56482.

…straints for comparison

InjectedClassNameType is such that every template specialization within
the Record scope ought to canonicalize to it, as outlined above the
definition of that Type.

This invariant is maintained during the tree transformation at the rebuilding
stage for a template specialization type; see RebuildTemplateSpecializationType.
In that, we attempt to retrieve the current instantiation from Sema.CurContext,
and if that fails, the transformation proceeds silently.

In terms of this issue, we previously set no CurContext other than that set
by Parser, who had left the FunctionDecl before performing the constraint comparison.
As a result, we would profile these types (i.e. InjectedClassNameType and its
specialization type) into different values and fail to consider two expressions
equivalent, although they are.

In passing, this also fixes a crash while attempting to dump the transformed expression.
We failed to look into the template parameters from a CXXRecordDecl, which we
should have done since the Decl we bound to a SubstTemplateTypeParmType can be of
a CXXRecord type per the call to `getTemplateInstantiationArgs`
within `SubstituteConstraintExpressionWithoutSatisfaction.`

This addresses llvm#56482.
@zyn0217 zyn0217 requested a review from erichkeane January 30, 2024 13:34
@zyn0217 zyn0217 marked this pull request as ready for review January 30, 2024 13:34
@llvmbot llvmbot added clang Clang issues not falling into any other category clang:frontend Language frontend issues, e.g. anything involving "Sema" labels Jan 30, 2024
@llvmbot
Copy link
Member

llvmbot commented Jan 30, 2024

@llvm/pr-subscribers-clang

Author: Younan Zhang (zyn0217)

Changes

InjectedClassNameType is such that every template specialization within the Record scope ought to canonicalize to it, as outlined above the definition of that Type.

This invariant is maintained during the tree transformation at the rebuilding stage for a template specialization type; see RebuildTemplateSpecializationType. In that, we attempt to retrieve the current instantiation from Sema.CurContext, and if that fails, the transformation proceeds silently.

In terms of this issue, we previously set no CurContext other than that set by Parser, who had left the FunctionDecl before performing the constraint comparison. As a result, we would profile these types (i.e. InjectedClassNameType and its specialization type) into different values and fail to consider two expressions equivalent, although they are.

In passing, this also fixes a crash while attempting to dump the transformed expression. We failed to look into the template parameters from a CXXRecordDecl, which we should have done since the Decl we bound to a SubstTemplateTypeParmType can be of a CXXRecord type per the call to getTemplateInstantiationArgs within SubstituteConstraintExpressionWithoutSatisfaction.

This closes #56482.


Full diff: https://github.com/llvm/llvm-project/pull/79985.diff

4 Files Affected:

  • (modified) clang/docs/ReleaseNotes.rst (+3)
  • (modified) clang/lib/AST/DeclTemplate.cpp (+4)
  • (modified) clang/lib/Sema/SemaConcept.cpp (+13-1)
  • (modified) clang/test/SemaTemplate/concepts-out-of-line-def.cpp (+23)
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index f0dea0c9bc89..eb35782b8185 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -135,6 +135,9 @@ Bug Fixes to C++ Support
   parameter where we did an incorrect specialization of the initialization of
   the default parameter.
   Fixes (`#68490 <https://github.com/llvm/llvm-project/issues/68490>`_)
+- Addressed an issue where constraints involving injected class types are perceived
+  distinct from its specialization types.
+  (`#56482 <https://github.com/llvm/llvm-project/issues/56482>`_)
 - Fixed a bug where variables referenced by requires-clauses inside
   nested generic lambdas were not properly injected into the constraint scope.
   (`#73418 <https://github.com/llvm/llvm-project/issues/73418>`_)
diff --git a/clang/lib/AST/DeclTemplate.cpp b/clang/lib/AST/DeclTemplate.cpp
index 7d7556e670f9..946a34ea8830 100644
--- a/clang/lib/AST/DeclTemplate.cpp
+++ b/clang/lib/AST/DeclTemplate.cpp
@@ -1583,6 +1583,10 @@ void TemplateParamObjectDecl::printAsInit(llvm::raw_ostream &OS,
 
 TemplateParameterList *clang::getReplacedTemplateParameterList(Decl *D) {
   switch (D->getKind()) {
+  case Decl::Kind::CXXRecord:
+    return cast<CXXRecordDecl>(D)
+        ->getDescribedTemplate()
+        ->getTemplateParameters();
   case Decl::Kind::ClassTemplate:
     return cast<ClassTemplateDecl>(D)->getTemplateParameters();
   case Decl::Kind::ClassTemplateSpecialization: {
diff --git a/clang/lib/Sema/SemaConcept.cpp b/clang/lib/Sema/SemaConcept.cpp
index 19a460f41175..5028879eda22 100644
--- a/clang/lib/Sema/SemaConcept.cpp
+++ b/clang/lib/Sema/SemaConcept.cpp
@@ -807,8 +807,20 @@ static const Expr *SubstituteConstraintExpressionWithoutSatisfaction(
       ScopeForParameters.InstantiatedLocal(PVD, PVD);
 
   std::optional<Sema::CXXThisScopeRAII> ThisScope;
-  if (auto *RD = dyn_cast<CXXRecordDecl>(DeclInfo.getDeclContext()))
+
+  // See TreeTransform::RebuildTemplateSpecializationType. A context scope is
+  // essential for having an injected class as the canonical type for a template
+  // specialization type at the rebuilding stage. This guarantees that, for
+  // out-of-line definitions, injected class name types and their equivalent
+  // template specializations can be profiled to the same value, which makes it
+  // possible that e.g. constraints involving C<Class<T>> and C<Class> are
+  // perceived identical.
+  std::optional<Sema::ContextRAII> ContextScope;
+  if (auto *RD = dyn_cast<CXXRecordDecl>(DeclInfo.getDeclContext())) {
     ThisScope.emplace(S, const_cast<CXXRecordDecl *>(RD), Qualifiers());
+    ContextScope.emplace(S, const_cast<DeclContext *>(cast<DeclContext>(RD)),
+                         /*NewThisContext=*/false);
+  }
   ExprResult SubstConstr = S.SubstConstraintExprWithoutSatisfaction(
       const_cast<clang::Expr *>(ConstrExpr), MLTAL);
   if (SFINAE.hasErrorOccurred() || !SubstConstr.isUsable())
diff --git a/clang/test/SemaTemplate/concepts-out-of-line-def.cpp b/clang/test/SemaTemplate/concepts-out-of-line-def.cpp
index 7323ad8d9ef2..b6fea2e0b4b3 100644
--- a/clang/test/SemaTemplate/concepts-out-of-line-def.cpp
+++ b/clang/test/SemaTemplate/concepts-out-of-line-def.cpp
@@ -537,6 +537,29 @@ template <class T>
 void X<T>::bar(decltype(requires { requires is_not_same_v<T, int>; })) {}
 } // namespace GH74314
 
+namespace GH56482 {
+template <typename SlotMap>
+concept slot_map_has_reserve = true;
+
+template <typename T> struct Slot_map {
+  constexpr void reserve() const noexcept
+    requires slot_map_has_reserve<Slot_map>;
+
+  constexpr void reserve(int) const noexcept
+    requires slot_map_has_reserve<Slot_map<T>>;
+};
+
+template <typename T>
+constexpr void Slot_map<T>::reserve() const noexcept
+  requires slot_map_has_reserve<Slot_map<T>>
+{}
+
+template <typename T>
+constexpr void Slot_map<T>::reserve(int) const noexcept
+  requires slot_map_has_reserve<Slot_map>
+{}
+} // namespace GH56482
+
 namespace GH74447 {
 template <typename T> struct S {
   template <typename... U, int V>

@zyn0217 zyn0217 merged commit ab70ac6 into llvm:main Jan 31, 2024
@zyn0217
Copy link
Contributor Author

zyn0217 commented Jan 31, 2024

I plan to backport this patch and #79698 to clang 18 a few days later to see if we're actually not breaking anything.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
clang:frontend Language frontend issues, e.g. anything involving "Sema" clang Clang issues not falling into any other category
Projects
None yet
3 participants