we expected.

at EOF -- it should be a rapidly arriving event, so only hang around
for 10 seconds, max., and then go ahead and report the "unexpectedly
closed" error.

io_timeout is 1.

- Call do_hard_links() with its new args.

the inplace processing doesn't make a second backup.

  the sake of MS Windows.
- Improved some comments.

of exiting.

the end of the processing.

potential keep-alive message that may never come.

--delete-after is happening.

- Call maybe_send_keepalive() from delete_in_dir() so that a really
  long delete-pass won't timeout.
- Renamed delete_files() -> do_delete_pass().
- Renamed delete_file() -> delete_item().
- Call get_dirlist() with its new arg.

- Made filesystem_dev a global that the generator can set during its
  delete pass.
- Fixed an off-by-one boundary check in send_directory().
- Added an arg to get_dirlist() so that delete_in_dir() can call it
  efficiently.

  "code" from recv_generator() to generate_files() so that they
  don't get recomputed for every file (we pass them as args now).
- Allow protocols < 29 to call maybe_send_keepalive() so that we can
  try to flush the output buffer if things are slow (though we can't
  send a keep-alive packet without protocol_version >= 29).

now supported, though we can only try to flush the output
buffer if we're in a lull.

- A few minor tweaks to the "log format" setting.

- Improved the description of --log-format.

interacts with --verbose.

not set (we used to only call if verbose or log_format were also
set).

not getting delete messages, which ensures that the daemon's log-
file output includes all deletions when per-file logging is enabled.