This came up in fuzz testing, although I had trouble recreating with the standard test runner. This solution makes wide-string value printing slower but it's already on the debug path. There are other solutions (passing the alignment when calling intern or just aligning based on the byte length), but this seemed like the least invasive and doesn't involve extra storage.