[clang] [WIP] [Modules] Delay reading type source info of the preferred_name attribute. (PR #122250)

[Ilya Biryukov via cfe-commits]

================
@@ -9860,6 +9860,33 @@ void ASTReader::finishPendingActions() {
     }
     PendingDeducedVarTypes.clear();
 
+    ASTContext &Context = getContext();
+    for (unsigned I = 0; I != PendingPreferredNameAttributes.size(); ++I) {
+      auto *D = PendingPreferredNameAttributes[I].D;
+      QualType InfoTy = GetType(PendingPreferredNameAttributes[I].TypeID);
+      TypeSourceInfo *TInfo = nullptr;
+      if (!InfoTy.isNull()) {
+        TInfo = getContext().CreateTypeSourceInfo(InfoTy);
+        if (auto Loc = TInfo->getTypeLoc().getAs<ElaboratedTypeLoc>()) {
+          Loc.setElaboratedKeywordLoc(
+              PendingPreferredNameAttributes[I].ElaboratedTypedefSL);
+          Loc.setQualifierLoc(PendingPreferredNameAttributes[I].NestedNameSL);
+          if (auto TypedefLoc = Loc.getNextTypeLoc().getAs<TypedefTypeLoc>())
+            TypedefLoc.setNameLoc(PendingPreferredNameAttributes[I].TypedefSL);
+        }
+      }
+
+      AttrVec &Attrs = getContext().getDeclAttrs(D);
+      PreferredNameAttr *New = new (Context) PreferredNameAttr(
+          Context, PendingPreferredNameAttributes[I].Info, TInfo);
+      cast<InheritableAttr>(New)->setInherited(
+          PendingPreferredNameAttributes[I].isInherited);
+      New->setImplicit(PendingPreferredNameAttributes[I].isImplicit);
----------------
ilya-biryukov wrote:

Duplicating the code here may lead to errors later.

Could we ensure that's the only code path that's being used to read preferred_name attributes? That would entail asserting that the common `readAttr` asserts the attribute is **not** PreferredName and I'd also suggest adding a documentation comment in `Attr.td` that PreferredName has a custom deserialization logic pointing to that function.

https://github.com/llvm/llvm-project/pull/122250