================
@@ -19804,3 +19803,36 @@ void
Sema::ActOnFinishFunctionDeclarationDeclarator(Declarator &Declarator) {
}
InventedParameterInfos.pop_back();
}
+
+bool Sema::BuildCtorClosureDefaultArgs(SourceLocation Loc,
+ CXXConstructorDecl *Ctor, bool IsCopy) {
+ assert(Context.getTargetInfo().getCXXABI().isMicrosoft());
+
+ if (!Ctor->getCtorClosureDefaultArgs().empty()) {
+ // If we build args for default constructor closures, those will have
+ // been generated *before* building args for any copy constructor closures.
+ assert(IsCopy || Ctor->getCtorClosureDefaultArgs()[0] != nullptr);
+ return false;
+ }
+
+ unsigned NumParams = Ctor->getNumParams();
+ if (NumParams == 0)
+ return false;
+
+ CXXDefaultArgExpr **Args =
+ new (getASTContext()) CXXDefaultArgExpr *[NumParams];
+
+ if (IsCopy)
+ Args[0] = nullptr; // Copy ctor closure will provide the first argument.
+
+ for (unsigned I = IsCopy ? 1 : 0; I != NumParams; ++I) {
+ ExprResult R = BuildCXXDefaultArgExpr(Loc, Ctor, Ctor->getParamDecl(I));
+ CleanupVarDeclMarking();
----------------
zmodem wrote:
Setting `RebuildDefaultArgOrDefaultInit` doesn't help.
IIUC the root of the problem is that when the default argument is a constexpr
variable, and we go to mark it as ODR-used:
https://github.com/llvm/llvm-project/blob/5c57289bf017f7aa216f242080b7244e4cea88b7/clang/lib/Sema/SemaExpr.cpp#L20748-L20755
we take the "delay the marking" path and put it in `MaybeODRUseExprs`.
The only code I see picking up the var from that list is
`CleanupVarDeclMarking()`, so that needs to get called somehow. There are only
a few places that do this:
- `Sema::PopExpressionEvaluationContext()` for unevaluated or constevaluated
contexts: but as you point out, that's conceptually wrong for us.
- `Sema::MaybeCreateExprWithCleanups`: this is not appropriate for the
arguments since any cleanups should run after the full function call.
- `Sema::MaybeCreateStmtWithCleanups`: not applicable
- `SemaOpenMP::EndOpenMPClause`: not applicable
- `Sema::ActOnGCCAsmStmt`: not applicable.
So I think it makes sense to call `CleanupVarDeclMarking()` ourselves here.
https://github.com/llvm/llvm-project/pull/203554
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
