[clang] [clang-tools-extra] [clang] check deduction consistency when partial ordering function templates (PR #100692)

Richard Smith via cfe-commits cfe-commits at lists.llvm.org
Fri Jul 26 11:32:00 PDT 2024


================
@@ -5399,11 +5434,85 @@ static QualType GetImplicitObjectParameterType(ASTContext &Context,
   return Context.getLValueReferenceType(RawType);
 }
 
+static TemplateDeductionResult FinishTemplateArgumentDeduction(
+    Sema &S, FunctionTemplateDecl *FTD, int ArgIdx, QualType P, QualType A,
+    SmallVectorImpl<DeducedTemplateArgument> &Deduced,
+    TemplateDeductionInfo &Info) {
+  // Unevaluated SFINAE context.
+  EnterExpressionEvaluationContext Unevaluated(
+      S, Sema::ExpressionEvaluationContext::Unevaluated);
+  Sema::SFINAETrap Trap(S);
+
+  Sema::ContextRAII SavedContext(S, getAsDeclContextOrEnclosing(FTD));
+
+  // C++ [temp.deduct.type]p2:
+  //   [...] or if any template argument remains neither deduced nor
+  //   explicitly specified, template argument deduction fails.
+  bool IsIncomplete = false;
+  SmallVector<TemplateArgument, 4> SugaredBuilder, CanonicalBuilder;
+  if (auto Result = ConvertDeducedTemplateArguments(
+          S, FTD, /*IsDeduced=*/true, Deduced, Info, SugaredBuilder,
+          CanonicalBuilder, /*CurrentInstantiationScope=*/nullptr,
+          /*NumAlreadyConverted=*/0, &IsIncomplete);
+      Result != TemplateDeductionResult::Success)
+    return Result;
+
+  // Form the template argument list from the deduced template arguments.
+  TemplateArgumentList *SugaredDeducedArgumentList =
+      TemplateArgumentList::CreateCopy(S.Context, SugaredBuilder);
+  TemplateArgumentList *CanonicalDeducedArgumentList =
+      TemplateArgumentList::CreateCopy(S.Context, CanonicalBuilder);
+
+  Info.reset(SugaredDeducedArgumentList, CanonicalDeducedArgumentList);
----------------
zygoloid wrote:

Doing this work for each parameter in turn seems unnecessarily expensive; can we do the setup steps once, and then for each parameter just do the subst and check?

https://github.com/llvm/llvm-project/pull/100692


More information about the cfe-commits mailing list