[RFC] bzr.jrydberg.versionedfile

[Jan Hudec]

On Wed, Dec 21, 2005 at 17:09:02 +1100, Robert Collins wrote:
> Some design level thoughts:
> I'm not sure what the goal of the rollback stuff is: we're aiming for
> lock-free read operation. This means that writing cannot in general ever
> rollback. heres why with your current implementation:
> ----
> lock a branch
> write a bunch of data to the end of knits, including some revisions.
> Decide to rollback
>  -> a reader reading one of those revisions is now fucked.
> ----
> 
> So I'd like some use cases for the rollback stuff or for 'commit' to not
> be a no-op. That is:
> Either there is no rollback,
> OR
> no data is actually available to lock-free readers until we commit. 

Just a note. I think there is a third variant which I think is correct -- the
data is actually available to lock-free readers before commit, but the
readers know the data are not commited yet and ignore them. Ie. you can write
the knits, but until the revision is written, readers see them but ignore
them.

-- 
						 Jan 'Bulb' Hudec <bulb at ucw.cz>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20051221/680aa48b/attachment.pgp