On Wed, Mar 11, 2009 at 11:20 PM, Gregory Maxwell <[email protected]> wrote:
> On Wed, Mar 11, 2009 at 11:14 AM, Daniel Cheng
> <[email protected]> wrote:
>> In DVCS, everybody own the whole copy of repository.
>> If some old data fall out of the network, you can:
>>
>> 1) ask the repository owner to insert his repository with jSite.
>> egit-freenet will pull from that automatically.
>>
>> /OR/
>>
>> 2) have any developer send you his repository.
>>
>> As all commits are secure-hashed. You can verify the authentic
>> of repository easily . This is not as bad as you think.
>
> Since people have bit-identical copies of the objects why can't
> everyone re-push these objects if they fall out of the network? The
> key holder should only be needed for the initial publication.
You means insert those object as CHK@ ...?
Then you have another set of problems:
1) Some kind of pointer (be it SSK@, KSK@ or metadata in some file)
must exist. The egit-freenet client *have* to know the hash for your
c...@..
If the pointer (let's assume it is a SSK@) fall out from the network,
we need the private key owner to recover.
-- this is no better then the scheme i proposed.
2) objects are always identical, but packs are not.
To generate exactly the same CHK@, we have to either keep all
object loose (lots of very small files, too slow), or come up with
a packing order / scheme (just plain impossible)
<some-random-idea-that-i-hate-to-implement>
we can insert the objects as KSK@ redirect to the packs,
and have git resolve them instead of using .idx file.....
but this involve too much git internals. I don't think this kind
of invasive patch should/can be merged upstream
</some-random-idea-that-i-hate-to-implement>
--
_______________________________________________
Devl mailing list
[email protected]
http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
