efficient file appends

[rsync at ka9q.net]

>This really is a minor point because as Martin pointed out, the fact that
>rsync's pipelining would be defeated means that the idea would have a
>drastic effect on throughput.

Okay, everyone has convinced me that my problem (excessive traffic in
the reverse direction) is best solved by simply specifying a large
checksum block size when transferring files that change only by
appending.

Now if we can only find the bug that corrupts data when compression is
enabled and a large block size is specified, I'll be in good shape!

Phil