The question was raised that there might be a problem with (UTF-8) bytes vs characters, but that's definitely not it - pythonrun.c:1362 UTF-8-decodes the line of source and then gets its character length to use as the new offset. So I don't think this is a duplicate of 2382.

(Side point: There appears to be quite a bit of complexity inside the CPython parser to cope with the fact that it does everything in UTF-8 bytes rather than simply decoding to text and lexing that. I presume that's for the sake of efficiency - that it'd be too slow to work through PyUnicode everywhere?)

Am looking into the rest.