Bug description:

Since GH-157861, the first write of a str subclass instance into an empty PyUnicodeWriter keeps that object as the buffer, so PyUnicodeWriter_Finish() returns the subclass instance instead of a str. In io.StringIO, getvalue() then returns the written object itself, and the next write re-reads it through its __str__(), which changes the contents:

import enum, io

class Color(str, enum.Enum):
    RED = "red"

f = io.StringIO()
f.write(Color.RED)
print(repr(f.getvalue()))
f.write("!")
print(repr(f.getvalue()))

<Color.RED: 'red'>
'Color.RED!'

When __str__() returns a shorter string, a later read() reads past the end of the internal buffer.

Expected: 'red' and then 'red!', as on 3.15.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-158213: Don't return a str subclass from PyUnicodeWriter_Finish() #158214