D132324 was soft-reverted, this this wouldn't be by fiat prohibited
However, builds were still failing as reported in https://reviews.llvm.org/D132324#3739383.
The issue was that D132298 still relied on this case not happening.
Since it in fact does happen with the soft-revert, I need to accommodate
that possibility in the if condition.
Additionally make the soft-revert closer to a real revert with
SEND_ERROR -> WARNING so CMake doesn't have a non-zero exit status
after competing configuring.