The "two-byte UTF-8" and "three-byte UTF-8" tests in code.json and string.json fail because libbson emits raw characters instead of "\u" escape sequences. Based on mongodb/libbson@980546a, it seems these tests were modified long ago to account for this difference of behavior.

The current behavior seems like a spec deviation. If this behavior cannot be changed for BC reasons, the upstream tests should be skipped and new tests should be introduced for the modified versions (per CDRIVER-4016).