Re: [Duplicity-talk] Non-integer mtime

[Peter Schuller]

> Using the metadata provided by the filesystem itself is indeed the
> fastest method since you only have to read the files that the FS claims
> to have changed (via metadata).  Hash based would mean reading every
> file in the source tree and that would be very slow.

The idea was to use the filesystem meta data, *EXCEPT* if the mtime
happens to be such that the file *COULD* possibly be subject to
modification-during-backup or
modification-right-after-in-the-same-second. This basically means that
mtime is within the time window during which the previous backup was
taken (with single seconds added appropriately to guarantee
correctness). So normally no overhead is added; only in edge case
timing cases would there be unnecessary checksumming.

So the cost should be small, while still eliminating the possibility
of silent data loss.

> It would be a rare thing indeed to have a file changed in the scenario I
> mentioned above.  Your method would close that issue, but I think it's
> more academic than realistic.

The solution or the problem? Because with the behavior above (i.e.,
not hashing except on changed mtimes or edge case timed mtimes) the
solution feels very realistic to me.

-- 
/ Peter Schuller

PGP userID: 0xE9758B7D or 'Peter Schuller <address@hidden>'
Key retrieval: Send an E-Mail to address@hidden
E-Mail: address@hidden Web: http://www.scode.org

pgpmM8W6VIbQV.pgp
Description: PGP signature