For the record, the python-dev thread referenced in the original post is <https://mail.python.org/pipermail/python-dev/2009-April/088212.html>.

The obvious fix to me is to have FileIO.close() call IOBase.close() to invoke the flush() before continuing with its own closing. This is what Serhiy’s patches appear to do, so I agree with them in general.

I do wonder what the point of duplicating test code in test_io and test_fileio is. The only difference seems to be calling open() versus calling FileIO() directly. Wouldn’t it be better to merge the code together, or maybe just assert open() returns a FileIO instance?