So the obvious guess is that it fails to find a bounding box in the case where the shape list only has FILLs in it.
Obviously this "bug" is not at all critical - there is no practical reason for wanting this to work.
It should do better. While checking this out I discovered valid cfdg input that causes Context Free to hang. Do you get the error message "Invalid UTF-8 encoding: ASCII null character" if you have startshape FILL with no line feed at the end?
The bug is that the code that filters out shapes that are too small or infinitely large is supposed to always draw FILL shapes. But if there is nothing but FILLs then it filters them out too. You are right that this is caused by the lack of a bounding box.
I also discovered that the cfdg parser handles end-of-file incorrectly in some cases and can hang or report weird errors.