Re: random access delta streams

Greg Hudson wrote:

> Branko wrote:
>
>> But don't forget that a window can describe differences relative to
>> existing target, not source data. A pure compression would be a
>> stream of such windows, so when seeking you'd have to decompress
>> from the beginning of the file up to the seek offset. (SVN doesn't
>> create such deltas now, although it would take only a small change).
>
>
> Actually, the whole architecture of how we apply deltas right now is
> predicated on the notion that views are only used from the source file
> and that source views come in ascending order. Otherwise we couldn't
> apply deltas to non-seekable streams using a constant amount of
> memory.

I'm aware of that. :-)

> So it would be a relatively big change.

I disagree. We could keep the streaminess and constant memory usage by
constraining the comparison to the previous window of target data. It
would be a very small change in the encoder, and I suspect in the
applyer, too.

It /is/ true though that this would make random-access delta application
much more expensive, at least on those bits of the file that use such
encoding.

-- 
Brane �ibej
    home:   <brane_at_xbc.nu>             http://www.xbc.nu/brane/
    work:   <branko.cibej_at_hermes.si>   http://www.hermes-softlab.com/
     ACM:   <brane_at_acm.org>            http://www.acm.org/