[PATCH] D132874: [clang] Don't emit debug vtable information for consteval functions

[Luke Nihlen via Phabricator via cfe-commits]

luken-google marked an inline comment as done.
luken-google added inline comments.


================
Comment at: clang/lib/CodeGen/CGDebugInfo.cpp:1761
 
-  if (Method->isVirtual()) {
+  if (Method->isVirtual() && !Method->isConsteval()) {
     if (Method->isPure())
----------------
shafik wrote:
> It is not clear to me if this is the core issue or not. Can you explain this a little better.
Thanks for the feedback!

Sure, here's what I'm thinking. This code calls ItaniumVTableContext::getMethodVTableIndex(GlobalDecl) on line 1771 below. That method asserts in VTableBuilder.cpp:2281 because it tries to look up a vtable entry for the consteval method and fails to do so. Any consteval methods are prevented from gaining vtable entries by the logic in VTableContextBase::hasVtableSlot(const CXXMethodDecl *MD):

```
bool VTableContextBase::hasVtableSlot(const CXXMethodDecl *MD) {
  return MD->isVirtual() && !MD->isConsteval();
}
```

This call is peppered throughout the VTableBuilder code, but the relevant call for our purposes is in ItaniumVTableBuilder::AddMethods() in VTableBuilder.cpp:1492, which causes any consteval method to be skipped when the code is iterating through all virtual member functions and adding them to the vtable.

My fix mirrors the logic in VTableContextBase::hasVtableSlot here, because the code assumes that if the method is virtual it has a vtable index, which consteval functions don't. Writing this, I've convinced myself it's better to just call that method directly, so I've refactored the code to avoid the duplicate logic.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D132874/new/

https://reviews.llvm.org/D132874