amccarth added inline comments.
================
Comment at: lib/CodeGen/CGDebugInfo.cpp:1688-1689
@@ -1687,3 +1687,4 @@
- if (CXXDecl->hasDefinition() && CXXDecl->isDynamicClass())
+ if (CXXDecl->hasDefinition() && CXXDecl->isDynamicClass() &&
+ !CXXDecl->hasAttr<DLLImportAttr>())
return true;
----------------
rnk wrote:
> David, since we're doing this check to work around a limitation of PDB-based
> debuggers, do you think we should guard check with EmitCodeView, so that we
> don't emit the full type if we're emitting DWARF?
>
> After that, I'd add a comment like:
> // Only emit complete debug info for a dynamic class when its vtable is
> emitted. However,
> // Microsoft debuggers are unable to resolve type information across DLL
> boundaries,
> // so skip this optimization if the class is marked dllimport.
Comment added.
I don't think the EmitCodeView check is necessary because it seems unlikely the
DLLImportAttr would be set with DWARF debug info, and, even if it is, I
wouldn't expect it to do any harm. But I'll defer to dblaikie or other DWARF
experts.
https://reviews.llvm.org/D23462
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
