[NIT] Managing repositories in MC

Colin Putney cputney at wiresong.ca
Mon Sep 22 17:52:06 UTC 2003


On Monday, September 22, 2003, at 01:45 AM, Daniel Vainsencher wrote:

> Avi Bryant <avi at beta4.com> wrote:
>> I don't like this at all.  It doesn't matter how little or how much
>> they're duplicated - they're still immutable.  I *really* don't think
>> you want to change that.  You have no idea how many .mcz files are
>> sitting around on disk with the old ancestry information.
>
> I think immutable versions is really critical for having all three of
> distribution, coherence and simplicity.

Agreed. I plead temporary insanity. I have no idea what I was thinking.

> [Colin said something about versions being content addressable instead
> of URL addressable, freenet style]

Though with the caveat that Versions are by identified UUIDs, not hash 
codes. This means that we can't implement Freenet's search algorithm. 
That's ok, our goals are not the same as Freenet's.

> I dunno. Seems to me that having a url back to where the version was
> published is nice. Especially since this makes the source of a package
> verifiable - compare it to what it claims to be. But also because it
> allows a form of discovery we dont have now, which is that I can find
> all the repositories whose changes get into Gorans SMBase line, and
> decide to track them myself.

Yeah, that is interesting, but I don't think it's worth the problems 
we'll get from having URL addressable versions.

> This all applies to content addressable scheme, except we're not using
> such a medium now. Lets get the benefits now, and discuss the perfect
> medium later...

Well, actually we *are* using content addressable versions at the 
moment. We don't have the fancy migration algorithm that Freenet does, 
but that's fine. We have the really important benefits of content 
addressability, which is being able to move versions between 
repositories. I do that all the time. I've already reorganized my 
wiresong repository twice and I'm about to do it a third time. I 
routinely copy versions from one repository to another without worrying 
about breaking URLs.

I'm arguing against *switching* to URL-addressable versions as a 
bandwidth optimization. We'd give up too much.

Colin



More information about the Squeak-dev mailing list