Welp, another day another attempt.  As mentioned in the PR 4847, atexit is not the answer.  If the raw/buffered file pair are part of a reference cycle and the GC cleans it before atexit runs, then the buffered data can get lost.

I attempted to implement my weakref idea (i.e. raw file keeps a weakref to the buffered file, calls flush before the raw file gets closed).  That doesn't work either because the GC clears the weakref before calling __del__.

The attached patch "buffer_register_flush.txt" does seem to work.  The downside is that it creates a reference cycle between the raw and buffered file objects.  Perhaps that is not a problem since unless you call close() on the raw file, you will be leaking resources anyhow.  In the patch, calling close() removes the reference cycle.

I still feel like this is worth fixing, as ugly as the implementation is.