Macintosh ABI non-zero vtable offsets

[riidefi]

• [x] Add a VPtrOffset to RecordLayout, specifically for CodeWarrior (reference the Microsoft-specific VBPtrOffset field)
  • [x] Add a getVPtrOffset() getter as well, referencing getVBPtrOffset()
• [x] Add a VPtrOffset to ItaniumRecordLayoutBuilder
• [x] At the RecordLayoutBuilder:1061 area, set VPtrOffset to DataSize before you increase the buffer to account for the vtable pointer (in the CodeWarrior case)
• [x] Extend the RecordLayout C++ constructor to take in vptroffset, and have the non-CodeWarrior calls specify CharUnits::fromQuantity(-1) for it. (calls are in RecordLayoutBuilder.cpp:3248 or so)
• [x] Add this edge case to VTableBuilder.cpp:1710
  • TODO: Make sure to account for AddressPointLocation being in units of sizeof(void*), not bytes.
  • ex CharUnits PtrWidth = Context.toCharUnitsFromBits(Context.getTargetInfo().getPointerWidth(0));

    if (/* abi check */ && !VTableIndices.empty()) {
    // FIXME: No idea if this is valid or not. Under what cases can a class have multiple vtables not from a base?
    assert(VTableIndices.size() == 1 && "Cannot have multiple vtables in the same sub-object");
    AddressPoints.insert(
    std::make_pair(BaseSubobject(RD, OffsetInLayoutClass),
                 VTableLayout::AddressPointLocation{
                     RD->getVPtrOffset(),
                     unsigned(AddressPoint - VTableIndex)}));
    } else { ... }

[JoshuaMKW]

All tasks have been implemented, whether they work correctly or not is untested currently.

[riidefi]

Looks like this task still needs to be done:

At the RecordLayoutBuilder:1061 area, set VPtrOffset to DataSize before you increase the buffer to account for the vtable pointer (in the CodeWarrior case)

[JoshuaMKW]

VPtrOffset is now being properly set, although codegen seems to be unaltered? I'm continuing investigation, it intrigues me that changes being made aren't having any affect on the output.

[JoshuaMKW]

Late to this, but all issues presented were fixed months back so I'm closing this issue now.