know if we retained it or discarded it.  Especially useful for batch-
reading mode where the old code could look like it did the update when
it really silently failed.

implies the latter.

only process the updates from the batch file when the generator has
indicated that it is ready (which ensures that all the necessary
dirs have been created).

- Avoid calling local_child().
- Create a pipe that lets the generator send us index values.
- Set batch_gen_fd for the receiver to read the pipe.

to the f_out pipe before we return from recv_generator().  This gives
the receiver something to sync with so that it doesn't rush ahead of
us (which could be bad if we didn't have a chance to create the
destination dirs yet).

call to map_file() until the block size was known).

  call to map_file() until the block size was known).
- Got rid of the setting of max_map_size.

  new def_window_size member variable.
- Got rid of max_map_size global (we use def_window_size now).

- Renamed the cleanup_fd* vars.

- A few spelling fixes.

the BATCH.sh file.

call (with extra arg).

  server-excluded file.
- If get_tmpname() fails we need to discard the update using
  receive_data().

characters and spaces in args, omit all the source args, omit the
include/exclude args if we can).

user specified a trailing '=VALUE' or put the value in a separate
arg.

write_sbuf(), and write_byte().

- Fixed the sentence describing how to read standard input with
  --read-batch.

symlinks and devices.

functions seemlessly work on fds that aren't for the main socket. This
involved changing some fd-variable names (to make them clearer), adding
io_set_sock_fds(), and making input buffering have a better enabled
flag (via an allocated buffer, just like the output buffering).  I also
got rid of the fd arg to some functions where the fd arg could only
specify the input or output fd for the socket (which we already know).

- Call io_set_sock_fds().