[clang] [clang] disallow narrowing when matching template template parameters (PR #124313)

Erich Keane via cfe-commits cfe-commits at lists.llvm.org
Fri Jan 24 12:46:22 PST 2025


================
@@ -6930,17 +6934,20 @@ ExprResult Sema::CheckTemplateArgument(NonTypeTemplateParmDecl *Param,
     IsConvertedConstantExpression = false;
   }
 
-  if (getLangOpts().CPlusPlus17) {
+  if (getLangOpts().CPlusPlus17 || PartialOrderingTTP) {
     // C++17 [temp.arg.nontype]p1:
     //   A template-argument for a non-type template parameter shall be
     //   a converted constant expression of the type of the template-parameter.
     APValue Value;
     ExprResult ArgResult;
     if (IsConvertedConstantExpression) {
-      ArgResult = BuildConvertedConstantExpression(Arg, ParamType,
-                                                   CCEK_TemplateArg, Param);
-      if (ArgResult.isInvalid())
+      ArgResult = BuildConvertedConstantExpression(
+          Arg, ParamType,
+          PartialOrderingTTP ? CCEK_InjectedTTP : CCEK_TemplateArg, Param);
+      if (ArgResult.isInvalid()) {
----------------
erichkeane wrote:

ExprResult always being initialized isn't a bad idea... but a much bigger change than i'd request here.  That said, the `Build/Act` functions SOMETIMES return uninit (even if by accident sometimes, particularly when used as an out param).  So the `isUsable` is both what we typically mean, and the safer option.

I realize we are bad about `isInvalid` all over the place, I've actively been trying to stop using it as we never MEAN it, so I show a strict preference for `isUsable`, particularly in these cases.  IMO, `isInvalid` becoming deprecated is perhaps of value.

https://github.com/llvm/llvm-project/pull/124313


More information about the cfe-commits mailing list