Code review comment for lp:~spiv/bzr/no-sigwinch-583941

Looks good to me. There was a lot of detailed discussion on IRC about the specifics of getting the logic maximally compatible. Trying for a minimal change on 2.1 and investigating further for trunk sounds like a solid plan. So, this branch only really needs a tweak to make the initial result of _terminal_size weaker than COLUMNS before landing.

The question of speed has been raised a couple of times. Andrew timed _ioctl_terminal_size at 11μs and that can be trivially improved upon if needed. The common case could just call fcntl.ioctl and do a string comparison, no need for function-level imports or struct packing and unpacking.