Currently this fails with a (misleading in the case of SEEK_END) "Can't do nonzero cur-relative seeks" error, but there's no conceptual reason it shouldn't be possible.

The offset should simply be taken as a character offset, without any pretense that the "file" represents bytes in some Unicode encoding. This is already done for SEEK_START and tell, and has not caused any problems.

can you provide more details? Python version and code/steps to reproduce. That'd be more helpful as it seems the methods are already implemented in latest version https://docs.python.org/3/library/io.html#io.TextIOBase.seek

That documentation isn't specific to StringIO, and in any case, the limitation in question isn't documented. The actual implementation is at https://github.com/python/cpython/blob/HEAD/Modules/_io/stringio.c#L484

But if examples would help, they're simple to come up with:

>>> f = io.StringIO('t\xe9st')
>>> f.seek(-1, io.SEEK_END)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: Can't do nonzero cur-relative seeks
>>> f.seek(2, io.SEEK_CUR)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: Can't do nonzero cur-relative seeks
# demonstration that SEEK_SET works treating all characters as one unit
>>> f.seek(2, io.SEEK_SET)
2
>>> f.read()
'st'

As far as I know this is the case in all currently maintained versions of Python 3, since the C-based unicode StringIO implementation was added in 2008.

This is a duplicate of issue25190.