================
@@ -2316,6 +2316,20 @@ void
ASTDeclReader::VisitCXXConstructorDecl(CXXConstructorDecl *D) {
}
VisitCXXMethodDecl(D);
+
+ // Microsoft CXX ABI specific:
+ // Restore the RecordToCopyCtor sidecar LUT entry so that `throw` expressions
+ // find the correct copy constructor for exceptions during codegen.
+ // There is no need to check the target info because the
+ // HasCopyConstructorForExceptionObject bit only gets set for the MS ABI.
+ if (D->isCopyConstructor()) {
+ // TODO What if this is not the same copy constructor which was chosen by
+ // LookupCopyingConstructor() in SemaExprCXX? Is there a better way?
----------------
rnk wrote:
I think we can rely on this assumption, and if it holds, then we don't actually
need this lookup table at all. I've observed that LookupSpecialMember is
cached, so once a special member (copy ctor) is looked up, it cannot change.
Since we know at the point of the throw site, all relevant copy constructors
are looked up, I think it should be safe to, in all cases, in
MicrosoftCXXABI.cpp, when building the catchable type array, iterate the
constructors and use the first constructor that returns true for
`isCopyConstructor`.
I propose you prototype that change (replace the call to getCopyCtor... with
the constructor search), run the tests, and see if that creates any fallout. If
all tests involving copy constructors with default arguments pass, then I think
it's safe to rely on this assumption in all cases, not just in the cases where
we're doing a modules build.
And, if it turns out we're wrong and we should be storing the result of
LookupCopyConstructor in a lookup table, that bug won't be hidden in a
modules-only build configuration. It's better for the behavior to be consistent
between regular builds and modules builds.
I propose the following experiment:
*
https://github.com/llvm/llvm-project/pull/114075
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
