[Monotone-devel] Re: What's the plans for cvs_import?

[graydon hoare]

On Wed, 04 Aug 2004 20:42:33 -0600, Derek Scherger <address@hidden> wrote:

> "it" being the manifest or the second cert? I'm pretty sure the manifest will 
> be synced
> but it sounds like the cert will too which is probably undesireable.

the second cert.

> so are you going back to allowing commits (and creating new revisions) that 
> don't include
> any manifest changes then? ;)

yes. revisions are explicitly constructed to enable this (perhaps "solely")

> I think I've noticed that if I commit a new branch directly to the database 
> file that a
> server is serving from it is not visible to someone pulling from that server 
> until the
> server is restarted. Are the merkle trees built in the database when the server 
> starts or
> when a new connection opens?

when the server starts, or when it receives input via netsync. a normal commit
does not (currently) update the merkle trie. if we keep the tries on disk, it
ought to; but then if I'm going to change it to just build the tries in memory
it ought not to.

> suppose we have A -> B -> C in net.venge.monotone
>              and A -> X -> Y -> Z in local.monotone.foo
>
> suppose I then cert Z into net.venge.monotone which leaves net.venge.monotone 
> not merged
> and with 2 heads C and Z
>
> will trying to merge resort to the 2 way merge because the ancestry of Z is not 
> part of
> the net.venge.monotone branch?

when you merge, since you have an ancestry path, you will get a 3-way merge.

> if we then sync and you don't get X and Y does this do anything particularly 
> nasty to the
> ancestry graph?

since I will not have the ancestry path, I will not be able to do a 3-way
merge. unless I teach netsync to try to load complete revision history
during a sync, which I could do. otherwise you have to do the merge if
you want the 3-way version. I can still do the 2-way version, of course.

-graydon