improved atime patch

[Wayne Davison]

On Tue, Apr 20, 2004 at 05:14:25PM -0400, Assar wrote:
> Here's a new version of the patch (relative to CVS) that adds
> -A/--copy-atime instead.

I agree that making it a separate option is a better choice.

> Any feedback on this patch and/or the previous one that I posted?

I've looked over your patch, made a few changes, and checked it into
the patches dir.  Here's what I changed:

  - The protocol should not sent the atime value if --copy-atimes was
    not specified.

  - Since the value never appears without the option, a protocol bump
    is not needed (since any remote rsync that can't support the
    atimes value will die with an option error).

  - There's no need to base the file-sending decision on a difference
    in the atime -- the atime will still get updated without the file
    getting rebuilt (which is faster, and the atime will settle down
    instead of being constantly bumped around by the copy process).

  - The time-setting logic was wrong when the atime changed but the
    mtime did not (uncovered because of we no longer force a resend
    when atime is different).

  - It's not useful to set the access time of a directory since the
    transfer constantly changes this on both sides (i.e. the values
    won't match when we're done, even if we tried to set it).

  - The code would constantly mention every directory name as being
    modified, even when it was up-to-date.

  - Giving the "tls" test program an unknown option would cause it to
    crash.

  - The tls output had the wrong spacing around the extra time values
    in the -u output.

Thanks for the patch!

..wayne..