On Fri, Oct 27, 2006 at 06:12:58PM +0100, toad wrote:
> On Fri, Oct 27, 2006 at 07:04:31PM +0200, bbackde at googlemail.com wrote:
> > On 10/27/06, toad <toad at amphibian.dyndns.org> wrote:
> > >
> > >Well, any external keys may not be "clean" CHKs. They might even be
> > >files inside containers etc. Although not if they are big files. So if
> > >you want to cleanly support externally originated keys you may need to
> > >support multiple keys per SHA-256.
> > 
> > Good to know. So I think I have to add this support for multiple keys.
> > I hoped the encoding would be deterministic, but I accept if you say
> > this is not the case (especially for big files).
> 
> It is basically deterministic, as long as you specify the same MIME type
> and filename. However there may be manifest lookups, container lookups,
> (either way, using keys embedded originally in freesites), different
> MIME types, and different filenames.

The following might be helpful, feedback would be useful:

1. Provide a "canonicalise" function via FCP. Either as a flag during a
normal request, or as a separate function, it might be useful to be able
to follow all the redirects until it ceases to be a simple URI (i.e. you
reach a splitfile or similar structure).

2. For any file over some arbitrary size, don't include the top-level
metadata in a container.

3. Allow selective inserts. It has long been planned to support
reinsert-on-demand via:
a) On a failed request, providing a list of the failed keys (or some
unique identifier thereof)
b) On an insert, allowing such a list to be included to indicate that we
should only insert specific blocks.

We can extend this concept further. We could tell the node to only
insert the top 1 or 2 levels of a splitfile. We could even support a
don't-insert list as well as a do-insert list, and include in the
don't-insert list all blocks from another file with the same contents.
(This is the long way around; we can macro this as an OverlapKey=<URI>
option, which downloads the specified file, and includes all blocks
therefrom).

Something useful to develop an API for, IMHO, although not needed
immediately.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: 
<https://emu.freenetproject.org/pipermail/tech/attachments/20061027/077aef8b/attachment.pgp>

Reply via email to