Re: [Monotone-devel] partial pull #2 - gaps instead of a single horizon

[Markus Schiltknecht]

Hi,

Timothy Brownawell wrote:
> So all we have to worry about is whether the history graph is
> well-formed (because that graph is based entirely on (1), which we don't
> know), and the worst problem we can have is that your database becomes
> broken and unusable.

Uhm.. I've lost you here. Who's 'we'?

Let's stick with the server vs. client notation for a moment. Assuming 
the server has the full history. On an initial pull, the client is 
trusting the server to provide it what's advertised. To authenticate 
revisions, the client can then check the rev certs.

Thus, an initial partial pull is no different in any way: the client 
normally trust the server to be who it claims to be, but the client can 
check the certs on the revisions later on (assuming it already knows the 
pubkeys).

Even if the server only has a partial repository, i.e. its repository 
has gaps in it as well, it's all the same process. Thus I'm beginning to 
think that checking gap information is totally unnecessary.

This would drive me away from using normal revisions (with a correct 
hash and possibly with certs) as sentinels, because that's overkill. 
Better stick with my original idea of having sentinels for revisions. 
Those sentinels happen to use a very similar textual description as 
revisions, but their description certainly doesn't hash to the revision 
id they sentinel.

Am I missing something important here?

Regards

Markus