Another behaviour change in v7 is the seekable() return value is now inherited from underlying file, instead of always being True. I mentioned on Reitveld why this could be a bad idea, but as it is undocumented and untested and will leave the new behaviour unless someone complains.
Posting v8, with these changes from v7:
* Reverted _read_exact() and read() loop changes to fix tests
* Restored myfileobj attribute and test that it is closed
* Clean up gzip read() loop so that it always returns data if available, rather than raising EOFError
* Document the read(None) is now allowed
* Other minor changes addressing review comments