[RFC][PATCH] New style of revision id

[Goffredo Baroncelli]

On Friday 27 October 2006 11:48, Martin Pool wrote:
> On 25 Oct 2006, Goffredo Baroncelli <kreijack at alice.it> wrote:
[...]
> If this is going to be the main mechanism then I think we should go the
> whole way and make it actually the hash of the revision as stored, to
> avoid the extra translation to and from testaments.
>
> Doing so means that the revision-id depends on the representation of the
> revision, and that representation needs to stay supported in future
> versions.  Of course we would still have the option to record different
> types of revision in the future.

Using the testament has the big adavantage that the hash can be verified 
easily with other languages/systems. Theoretically the hash can be verifyed 
without knowing how bazaar stores the data.
But I don't think that is the problem.

The problem is that until the knit format use as index the revision-id, we 
have to access ( == read the file ) the changed files two times:
1) for computing the hash ( the hash depends on the checksum of the ALL file )
2) for storing the file in the repository. 

That implies that we have to check that the checksums evaluated at the step 
one and two are equal! 


> 
> The behaviour of upgrade on a repository would be somewhat different: it
> could rewrite the container format, but not the contents of the
> revisions.
> 
> As a concrete example, we could stop recording revisions in XML, but
> we'd have to still be able to read all the old ones.
> 
> -- 
> Martin
> 

-- 
gpg key@ keyserver.linux.it: Goffredo Baroncelli (ghigo) <kreijack at inwind.it>
Key fingerprint = CE3C 7E01 6782 30A3 5B87  87C0 BB86 505C 6B2A CFF9