[clang] [clang] Implement P2582R1: CTAD from inherited constructors (PR #98788)

[Matheus Izvekov via cfe-commits]

================
@@ -931,12 +935,73 @@ getRHSTemplateDeclAndArgs(Sema &SemaRef, TypeAliasTemplateDecl *AliasTemplate) {
   return {Template, AliasRhsTemplateArgs};
 }
 
+struct InheritedConstructorDeductionInfo {
+  // Class template for which we are declaring deduction guides
+  // This is `C` in the standard wording
+  TemplateDecl *DerivedClassTemplate;
+
+  // `template<typename> CC` in the standard wording
+  // This is the type of template that is substituted in the deduction guide
+  // return type `CC<R>`
+  TypeSourceInfo *CCType;
+};
+
+// Build the type for a deduction guide generated from an inherited constructor
+// C++23 [over.match.class.deduct]p1.10:
+// ... the set contains the guides of A with the return type R
+// of each guide replaced with `typename CC<R>::type` ...
+TypeSourceInfo *buildInheritedConstructorDeductionGuideType(
+    Sema &SemaRef, const InheritedConstructorDeductionInfo &Info,
+    TypeSourceInfo *SourceGuideTSI) {
+  auto &Context = SemaRef.Context;
+  const auto *FPT = SourceGuideTSI->getType()->getAs<FunctionProtoType>();
+  assert(FPT && "Source Guide type should be a FunctionProtoType");
+
+  // This substitution can fail in cases where the source return type
+  // is not dependent and the derived class is not deducible
+  Sema::SFINAETrap Trap(SemaRef);
----------------
mizvekov wrote:

I see. But why then are we suppressing the diagnostics in this case?
If there is a substitution failure otherwise, is that currently being diagnosed somewhere else?

https://github.com/llvm/llvm-project/pull/98788