It has never been the case that upper() or lower() are guaranteed to preserve string length in Unicode. For example, some characters decompose into a base plus combining characters. Ligatures are another example. See here for more details:

https://unicode.org/faq/casemap_charprop.html


However, this example surprises me. In Python 2, I get the result I expected:

py> c = unichr(304)
py> unicodedata.name(c)
'LATIN CAPITAL LETTER I WITH DOT ABOVE'
py> unicodedata.name(c.lower())
'LATIN SMALL LETTER I'


If I am reading the UnicodeData.txt file correctly, I think that the right behaviour is for LATIN CAPITAL LETTER I WITH DOT ABOVE to lowercase to LATIN SMALL LETTER I, as it did in Python 2.

ftp://ftp.unicode.org/Public/UCD/latest/ucd/UnicodeData.txt